The revenue lexicon

It puts a measurement where an opinion was, and its worth depends on having chosen beforehand which number counts and how long the test runs.

Example. An e-commerce manager uses the web's historical archive to reconstruct how competitors changed their calls to action over the years. Those variants are A/B tests somebody else already ran, and reading them points to the tests worth running on your own site.

It puts marketing, sales and customer service in the same programme. It starts from the observation that in business-to-business sales a large share of revenue comes from existing customers, and lifetime value grows when the experience after the sale lives up to the promise.

Example. An industrial maintenance provider, after the first contract with a plant, prepares a quarterly report of avoided downtime for the production manager. At renewal the customer adds a second site.

It pays off when a few dozen accounts carry most of the revenue, and it requires Marketing and Sales to work the same list of names in the same order of priority. The work moves in steps: the ideal customer profile is defined, the target account list is drawn from it, and interest is measured per company with the marketing qualified account instead of per contact. The first conversation is opened through outbound marketing (calls, cold emailing, introductions from shared contacts, invitation-only events), while inbound marketing still helps the listed companies find you. When the programme continues after signature, through renewal, it is called account-based experience.

Example. The case below is constructed, since the term does not appear in the articles yet. An industrial supplier with a few dozen possible customers works from a list rather than an audience: each row carries the name of the company to be won, the rep who owns it and the reason that company should change supplier. Marketing material is built for those rows, and cost per lead stops being the measure of the work.

It automates repetitive operations and handles tasks that require advanced decision making, meaning the ability to choose between alternatives, with reduced human supervision.

Example. Checking availability in the calendar, offering the client three time slots on WhatsApp, booking the appointment and sending the deposit link; or checking stock and reordering a product when it falls below the threshold.

The articles on this site stress one point: "The focus of AI adoption should extend far beyond short-term cost savings." When the push comes only from the top and frontline staff are left with their doubts, the company creates an internal disconnect, and that is where Shadow AI grows: the use of AI tools outside any company rule. For this reason adoption goes hand in hand with AI governance and with change management that involves the people who will actually use the tools.

Example. A manufacturer of mechanical components starts with a single process, technical quotes: it chooses the tool, trains the sales office, sets rules on customer data and measures response time before and after. Only when that process holds does it move to the next one.

What separates it from automation is who picks the next step, and that shifts the governance question from «is the rule well written?» to «how far can it go alone, and who notices when it crosses the line?». In Enrico Giubertoni's articles the agent also appears on the other side of the counter: the consumer delegates search and comparison to it, and the brand ends up talking to a machine that represents its customer.

Example. A company hands payment reminders to an agent: it reads the overdue invoices in the management system, picks a tone, writes to the customer and schedules the follow-up. As long as nobody has written down which customers the agent may chase on its own, the system draws that line one case at a time, and whoever reads the log after the incident is the one who finds out.

It does not arise from a lack of information but from an excess of it: complexity, perceived risks, the speed at which tools change and the stakes involved. It is a form of analysis paralysis, and recalls the paradox of choice described by the psychologist Barry Schwartz (2004): the more alternatives there are, the harder it is to pick one. The results are stalled projects, missed opportunities and a competitive disadvantage that grows while the market moves on. The way out is to shrink the decision: one use case, one scope, one measure of success.

A committee spends a year evaluating five AI platforms while waiting for the definitive one, while a competitor adopts one on a single process and learns.

In Enrico Giubertoni's articles it is a matter of architecture before compliance, and the recurring question runs like this: if the board asked today for undeniable evidence that AI adoption is a controlled asset, could management produce it? Whoever gives up on building it falls into AI governance paralysis, the direct consequence of working without a guide: nobody knows how far they are allowed to go, and the projects stand still.

Example. At a manufacturing company three departments use three different tools, none of them passed through legal and none of them has an owner. AI governance starts with one box on the org chart with a name inside it: the guidelines document comes afterwards, and without that name it stays on the desk.

It is a leadership weakness more than a technical problem: the gap between ambition and action where competitive advantage is lost. It differs from AI decision paralysis, which blocks before adoption: here adoption has already happened, often as shadow AI, and nobody is accountable for it. The European AI Act and the ISO/IEC 42001 standard make the problem concrete, because they require knowing which systems run in the company and who controls them. The first step out is to measure the real starting point.

The board asks how many AI tools the company uses and with what data, and nobody can answer.

In Enrico Giubertoni's articles the danger sits in perceived credibility rather than in the error itself: as these systems enter daily workflows, people overrate the reliability of the direct answer and give up on checking the sources. Mitigating that risk belongs to whoever brings the tool into the company, alongside transparency and respect for privacy.

Example. A salesperson preparing a quote for an industrial client asks the assistant which European directive governs a component. Back comes a precise number, stated with confidence, that exists nowhere. The quote goes out, the client has an engineering department, and the lost credibility costs more than the saved hours.

The AI Act, the European regulation on artificial intelligence, names it in one place only, Article 4, which since July 2026 asks organizations to «take measures to support the development» of that literacy and sits outside the list of duties carrying a penalty. Hence the point Enrico Giubertoni highlights: the one provision covering the ground where shadow AI grows (the artificial intelligence tools people use for their work outside any company choice and any company inventory) has no sanction attached and has just been softened.

Example. At a distribution company the sales director has an assistant summarise the open deals and trusts that text as much as a spreadsheet. Nobody explained that it is a statistical prediction to be read again: the missing literacy costs a meeting spent arguing over a figure the system invented.

In place of the fixed web of roles and reporting lines it puts a flexible structure based on collaboration between human talent and machine intelligence, and redesigns who governs data, decisions and experimentation. Adding an AI box under the IT department is not enough: rigid hierarchies and silos work only where information changes little. Paul Daugherty and H. James Wilson, in Human + Machine (2018), described the new roles emerging between people and machines; Ethan Mollick, in Co-Intelligence (2024), the daily work with AI as a colleague.

A company appoints in each department a person accountable for the AI systems in use, and a committee that decides which ones to test and with what data.

Born as an experiment in 2023 under the name Search Generative Experience, it was launched in the United States in May 2024 and then extended to other countries. It changes the deal search used to rest on: fewer clicks to websites for simple questions, more value for in-depth and authoritative content, which is more likely to be cited as a source. It is the ground of generative engine optimisation.

Someone searching "what is churn" reads the definition on the results page itself, next to the sites it was taken from.

An effective policy also states where it is published, who checks it and how often it is updated. What the law looks at is whether measures are adequate and applied: a policy that nobody checks protects little, and a policy that slows work down too much pushes employees towards Shadow AI (the use of tools the company has not chosen).

Example. A financial firm allows the company AI assistant for summarising internal documents and prohibits it for customer data. The policy says so in writing, names the owner and provides for a review every six months.

For a brand the damage is to reputation: readers recognise mass-produced content and stop trusting the source. The remedy is a review step with an owner before the content reaches the customer.

Example. A financial services firm publishes ten AI-generated articles a week, all with the same structure and a few wrong figures. Within months traffic falls and customers start pointing out the errors.

It takes the form of questions that challenge the foundations of the business model: instead of asking how to do something faster, it asks why it is done at all. Marco Iansiti and Karim Lakhani, in Competing in the Age of AI (2020), showed that companies built around data compete with a different operating model, not with one more tool; Ajay Agrawal, Joshua Gans and Avi Goldfarb, in Prediction Machines (2018), read AI as a fall in the cost of prediction, which changes which decisions are worth taking.

A company that wanted to automate its quotes realises, by asking why, that customers want a public price rather than a quote.

The phrase spread from 2015 through the research firm Gartner, which presented it as the step after the big data economy. Raw data gains value when a system turns it into a choice that matters to the customer: a recommendation, a price, a message at the right moment. This is where the question of algorithm governance begins.

Two retail chains hold the same sales data; the winner is the one whose algorithm offers each customer the product about to run out in their pantry.

Every digital channel is made of a medium and an algorithm which, like a film director, decides what content to show, to whom and when. Governing it means knowing its signals and building content and data that the director rewards, turning a black box into an ally. The subject has a public side too: in Europe the Digital Services Act requires large platforms to explain the criteria their recommender systems use to rank content, one piece of the European Digital Identity.

A company studies what a social network's algorithm rewards and reorganises formats and posting times instead of raising the advertising budget.

The result is answers that are systematically distorted against some groups.

Example. If a search for «woman» returns almost only images of white Caucasian women, the algorithm has identified the concept of woman with one of its groups alone. In any automated assessment the same mechanism penalises whoever looks little like the past cases the model learned from.

Applied to marketing, it describes the brand's continuous presence beside that person, assisting and informing them every day instead of appearing only in launch campaigns. The second meaning is the answer to the first: campaigns in bursts, designed for an audience that watched television at fixed times, leave empty the days on which the connected consumer searches, compares and decides.

A household appliance brand publishes user guides every week, answers in chat in the evening too and sends a maintenance reminder, instead of putting its whole budget into the Christmas campaign.

It says something precise: the advantage does not come from owning data, which everyone now owns, but from reading it sooner than others and turning it into a decision. The work consists of widening that distance.

Example. Two competitors receive the same signal: a long-standing customer stops asking for quotes. The first notices at quarter end while reading revenue, the second the following week while reading the CRM. Between the two there is a signal advantage, and amplifying it means making it systematic rather than lucky.

Example. A company launches three artificial intelligence pilots, each with its own dashboard, and none of them changes a decision: a new form of Analytical Inertia disguised as innovation.

In the articles it appears as anchoring bias, the systematic distortion of judgement produced by the first figure or the first data point encountered, among the traps that warp competitive analysis. Recognizing it is the first step towards reducing its impact on decisions.

Example. A salesperson opens the negotiation at full list price and then grants ten per cent: the customer values the offer starting from that first figure. The same final price, presented straight away as a «net price», feels higher.

He should not be confused with the curator of the TED conferences, who shares his name. His main ideas come in three books: - «The Long Tail», first a Wired article in 2004 and then a book in 2006. The digital economy makes an almost endless supply of niche products economically sustainable. The logical step is this: when shelf and warehouse cost almost nothing, many products selling a little are together worth as much as the few hits; - «Free: The Future of a Radical Price» (2009), a conceptual basis for understanding the forms of free and designing freemium strategies (the model where the basic service is free and advanced features are paid for); - «Makers» (2012), on digital manufacturing and additive manufacturing bringing production within reach of small teams. The site's Italian blog calls «The Long Tail» one of the essential books for living in the new century.

Example. A distributor of laboratory chemicals lists online thousands of rare reagents no physical warehouse would stock: each sells little, but together they bring in revenue the shorter catalogue never reached. This is Anderson's long tail applied to an industrial market.

It is used for multi-year contracts and subscriptions, and allows customers with different durations to be compared. It guides the choice of sales model: a high ACV justifies dedicated sellers and account-based marketing programmes, a low ACV calls for more automated channels. Together with customer acquisition cost it shows how long a customer takes to pay back what it cost to win them.

Example. A maintenance services company signs a three-year contract worth ninety thousand euros: the ACV is thirty thousand. Against an acquisition cost of fifteen thousand euros, the customer pays back the sales spend in six months.

An open interface turns a product into an ecosystem, because other developers add services the manufacturer never imagined. Its commercial weight shows through its absence: two systems that each work perfectly on their own hand the data over manually, and the manual hand-off is where the data starts to age.

Example. The digital storefront of a retail chain has to read stock levels from the warehouse system. With an interface between the two, the promise on screen holds all day; with a file exported every night, it holds until the first sell-out in the afternoon.

The difference: Augmented Reality overlays digital elements on the real world framed by the camera, so the lipstick appears on your own lips in real time; Virtual Reality replaces the real world with an entirely digital environment, usually entered with a headset, for example a shop or a beauty cabin rebuilt in three dimensions.

Example. The case below is constructed. A manufacturer of packaging machines sends the customer an augmented reality app: the plant manager points the camera at the shop floor and sees the machine at full scale, with its footprint and the lanes for forklifts. With a virtual reality headset the same customer walks the whole line before it exists, and objections about space surface before the order rather than during installation.

In his texts it has two uses: the values a brand narrative draws on, and the representative figure a target persona (the typical portrait of the customer being addressed) is built from. Selection matters more than quantity: a few archetypes the departments actually use beat many portraits that stay on a slide.

Example. A window manufacturer builds two customer portraits: the surveyor who orders a hundred units and the family renovating once in a lifetime. Those are two archetypes, and the sales network actually uses them because there are few of them and they are recognizable.

In Enrico Giubertoni's articles the term almost always arrives with a caveat about method: whoever holds it to be a tool to be managed reduces it to a cost centre to be optimised, while the competitive advantage comes from the organization that governs it, from the org chart, the roles and the culture, before the model that was bought. That is where companies slip: buying a platform is far easier than redesigning roles, decision flows and data responsibilities.

Example. A management team buys a licence, hands it to Marketing and Sales and announces the project, while roles and responsibilities stay exactly as they were. The tool sits on top of an unchanged process, and the pilot closes without having changed a single decision.

The rule gets written first, and it holds for everyone.

Example. A customer checks the price on a phone while standing in the shop, then buys online. The act is one, and inside the company its attribution is contested between two divisions with two sets of accounts and two bonus schemes: the fix starts with a single record that keeps the lead's origin and the place of delivery together.

Unlike virtual reality it does not cut the user off from reality: it enriches it. The phrase was coined in 1990 by Tom Caudell, a Boeing researcher, for a system that showed workers the layout of the cables to fit; in 1992 Louis Rosenberg built the first working system for the US Air Force. It reached the general public through phones: in 2016 the game Pokémon Go, then social media filters and virtual try-ons of make-up, glasses and furniture. Unlike the metaverse, augmented reality is firmly built into marketing processes, and increasingly so thanks to artificial intelligence and phygital: it lets people try a product without touching it, and in cosmetics it has become a sales tool.

A customer frames their living room with the phone and sees the sofa already placed between the walls, in different upholstery colours, before ordering it.

It was formulated by the American Douglas Engelbart in 1962 in the essay Augmenting Human Intellect, following two earlier visions: the Memex imagined by Vannevar Bush in 1945 and the human-computer symbiosis described by J. C. R. Licklider in 1960. To develop it Engelbart founded the Augmentation Research Center at the Stanford Research Institute, and in 1968 he showed the mouse, hypertext and video conferencing in a single public demonstration. It is the conceptual ancestor of AI used as a copilot, and of the idea of human-in-the-loop.

An analyst who uses AI to explore in an hour hypotheses they would have tested alone in a week, keeping the decision for themselves.

It applies to people and to websites: an authoritative salesperson knows the product in depth and handles objections professionally, and a site that is authoritative on a topic passes that credit to its own pages, to the point that search engines and artificial intelligence systems pick it as a source.

Example. In a retail outlet the assistant who knows the product in depth, proposes tailored solutions and handles objections professionally builds a professional authority from which a more solid relationship with the public grows, and it shows in the outlet's overall performance and in customer loyalty. On a website, in-depth articles linked to each other around a pillar page, the page that covers a topic in full, concentrate authority on that topic and pass it on to the linked pages.

In Enrico Giubertoni's articles it serves as the term of comparison for agentification (the shift to systems that choose by themselves the steps for reaching a goal they have been assigned), and the recurring misjudgement is precisely confusing the two: an agent is delegated an executive autonomy of reaction inside defined purposes, while the vision, the strategy and the decision that direct that purpose stay with a person. In marketing it takes the shape of marketing automation: software that crosses complex data to segment an audience, recognise when a segment is ready to buy and adapt the messages accordingly.

Example. A bank connects its customer database to its sending platform and sets one rule: whoever opens a current account receives the message about the savings plan (investing in small monthly instalments) twenty days later. The Marketing manager chose the criterion, the machine carries it out, and responsibility for that message stays with whoever wrote the rule.

It was described by the Israeli psychologists Amos Tversky and Daniel Kahneman in 1973, within the research programme on mental shortcuts (heuristics) that founded behavioural economics. It leads to overrating conspicuous risks and underrating quiet ones, and it is why a well-told story weighs more than a statistic.

After news of a cyber attack on a competitor, a company moves its whole budget to security and neglects the slow loss of customers.

The word comes from the Sanskrit avatāra, "descent", which in the Hindu tradition means a deity taking earthly form. The digital sense was born with Lucasfilm's online game Habitat (1986) and spread with Neal Stephenson's novel Snow Crash (1992), where avatars live in the metaverse. For a brand the avatar is also a customer to dress and to serve: fashion and cosmetics sell digital garments and make-up designed for it. Outside gaming, however, this kind of experience has not yet managed to become a stable part of companies' marketing processes.

A person enters a gaming platform with their avatar and buys it a digital jacket from a fashion brand, identical to the one sold in the shops.

It reports what happened at the till, while the reasons why it happened sit in the signals upstream: footfall, the ratio of visitors to buyers, dwell time. Those are the signals you can act on the following week.

Example. A chain's head office measures sign-ups, attendance and satisfaction for its channel training; the shop measures average order value and reordering. Each side measures what it controls, the stretch joining the two measures has no owner, and that is precisely where the money leaks.

These traits make tools such as the ideal customer profile, account-based marketing and marketing and sales alignment central in B2B, and a customer's value shows in lifetime value more than in a single order.

Example. An industrial valve manufacturer sells to plant builders. Each order goes through engineering, purchasing and management, takes six months and brings repeat orders for ten years.

That is why mass channels, direct email marketing, advertising and the conversion rate of physical and online stores weigh more in B2C.

Example. A bank offers the same account to both businesses and households. Households open it in ten minutes in the app, businesses go through a meeting with a relationship manager: two journeys for the same product.

The more an idea spreads, the more followers it attracts, regardless of the evidence for it. For a company it is the risk of chasing market trends without asking whether they serve its own customers; for marketing it is a lever, related to social proof.

A company opens an account on an emerging social network only because its competitors have, without knowing whether its audience is there.

The brain perceives a third sound, a beat equal to the difference between the two frequencies. The phenomenon was discovered in 1839 by the German physicist Heinrich Wilhelm Dove, and brought back to attention in 1973 by the biophysicist Gerald Oster in an article in Scientific American. Some theories hold that the beat can steer brain waves towards states such as relaxation or concentration; the scientific evidence is still mixed, which is why the subject calls for caution when it enters a product's communication.

A 200 hertz tone in the left ear and a 210 hertz tone in the right produce a perceived beat of 10 hertz.

The simplest picture is an account book of which every participant holds an identical copy: each new line is checked by everyone before it is written, and if someone tries to change an old line in their copy, the other copies contradict it at once. That is why no central authority, such as a bank or a notary, is needed to guarantee the register is true: the fact that everyone holds it is the guarantee. Each block carries a digital fingerprint (hash) of the previous one, and changing one would break the chain. The idea of timestamping digital documents in a chain comes from Stuart Haber and W. Scott Stornetta (1991); the first working blockchain was born in 2009 with Bitcoin, the digital currency described in 2008 by an author who remains anonymous, Satoshi Nakamoto. In 2015 Ethereum, conceived by Vitalik Buterin, added smart contracts, small programs that carry out an agreement on their own when the conditions are met. Since 2018 the European Union has been building its own infrastructure, EBSI (European Blockchain Services Infrastructure), to certify diplomas and public documents across member states. Beyond digital currencies it is useful wherever several parties who do not know each other must trust the same data: tracing a supply chain, a degree, a certificate of authenticity.

A wine bottle carries a code: by scanning it, the customer sees on a blockchain the vineyard, the harvest date and every step to the shelf, recorded by the different players in the supply chain and which none of them can touch up afterwards.

Example. The category pages of a company site pull traffic from search and show a high bounce rate. Visitors arrive and stop there: the site's internal structure sends them back to start the search over, and the category taxonomy is steering navigation badly.

The brand orchestrates every touchpoint, digital and physical, and moves from promoting products to building an ecosystem that accompanies the customer. Enrico Giubertoni developed the concept in his work on digital strategy in cosmetics, where the beauty brand becomes the place where the customer finds diagnosis, advice, virtual try-on and community, even before the product.

A sports brand whose app records workouts, suggests training plans, organises events and sells the right equipment.

David Aaker, who theorised it in Managing Brand Equity (1991), describes it as a set of assets and liabilities linked to the brand, which can add value to the offer or take it away; in 1993 Kevin Lane Keller read it from the customer's side, as the difference that knowing the brand makes to their response. It is measured by the price premium customers accept, by loyalty and by resilience in a crisis.

Two drills with the same specifications, made in the same factory: the one from the well-known brand sells at a higher price and with fewer discounts.

They are pinwheel calculators: at the end of the nineteenth century the company licensed the design of Willgodt Theophil Odhner, and its engineer Franz Trinks improved it until it was fit for large-scale production. In 1921 the company became a joint-stock company, and the Brunsviga name covers a family of models found in German and European offices for much of the twentieth century. Among them: - the Brunsviga Trinks MR, the compact model from the years of the First World War; - the Brunsviga 13, produced between the 1920s and the Second World War. From 1922 to 1927 the neighbouring Rema Patent maker, also from Braunschweig, effectively merged into the firm, and some of its models were sold under the Brunsviga brand. It is one of the best-known European names among mechanical technologies of intelligence.

Example. The case below is constructed, because the term does not yet appear in the articles. A German bank in the 1930s calculates interest on accounts at the end of the quarter: every clerk has a crank calculator on the desk, and the word Brunsviga on the cover is enough to name the type of machine, as would later happen with the name of a computer brand.

It is a desk machine with three elements that explain its use in the office: - the setting levers, one per digit, which push out the pins of the discs; - the sliding carriage, which holds the result register and the revolution counter and shifts one position to multiply by ten; - the side crank, turned forwards to add and backwards to subtract. The mechanism is the one devised by Willgodt Theophil Odhner and developed in Germany by Franz Trinks, and it belongs to the technologies of intelligence that brought mechanical calculation into offices before the computer.

Example. The case below is constructed, because the term does not yet appear in the articles. The accountant of a chemical company in the 1930s divides the cost of a batch by the kilos produced: he sets the cost with the levers, shifts the carriage and turns the crank backwards until the bell rings, then reads the quotient in the revolution counter.

MR stands for Miniature Rechenmaschine, meaning "miniature calculating machine": it is a compact version of the firm's desk machines. The Trinks name refers to Franz Trinks, the engineer who developed the company's pinwheel models on the design of Willgodt Theophil Odhner. According to the Rechnerlexikon the label Trinks-Brunsviga is an addition to the name meant to stress the quality of the machines, so finding it on a machine does not identify a variant. The model belongs to the mechanical technologies of intelligence and predates the Brunsviga 13.

Example. The case below is constructed, because the term does not yet appear in the articles. A manufacturing company in the 1920s sends its inspector to audit the accounts of its branches: a compact calculator fits in the suitcase and travels with him from city to city, where a desk machine would stay at headquarters.

The figure on its own says little, because having a budget is no guarantee of allocating it well: the return depends on the capability of the team steering it. The recurring choice in the articles is about destination: less spend on interrupting the consumer with traditional advertising, more on accompanying and assisting them.

Example. In the fourth quarter management allocates next year's budget and adds a line for training on new tools, artificial intelligence included. The same sum spent entirely on advertising space buys twelve months of visibility; spent partly on the team's skills it leaves a capability that is still there the year after.

They work on outbound marketing, while the sales development representative more often handles contacts coming from marketing. In account-based marketing programmes the BDR works the target account list and takes target companies as far as the first meeting.

Example. A technical packaging manufacturer wants to enter the pharmaceutical sector, where it has no customers. A BDR maps sixty companies, identifies the purchasing and quality managers in each, and within three months brings the sellers nine meetings.

The term in its modern sense was first used by Hans Peter Luhn, an IBM researcher, in the 1958 article «A Business Intelligence System»; Howard Dresner, a Gartner analyst, spread it at the end of the 1980s. Its value lies in cross-referencing data. The data of a single department tells half the story, while BI places side by side, for example, e-commerce sales and physical store sales, and reveals behaviours that no department sees on its own.

Example. A European brand with an e-commerce site and stores cross-references the data of the two channels and finds that people living near a store use the online cart as a reminder and then buy in the store.

In companies organised by business unit, each unit tends to choose its own tools and methods: that is where the data a company owns and does not turn into decisions accumulate, along with the misalignments between departments that slow shared work down.

Example. A chemical group has three business units: paints, adhesives and additives. Each has its own price list, sales force and accounts, and each has adopted a different AI assistant.

Enrico Giubertoni sets it against the target (the group of customers a company wants to sell to, described in broad categories): the target stops at socio-demographic variables, age, sex, role and location, while the persona digs into psychographic ones, identity and interests, buying habits, the places people spend time in and the way they decide to buy. It is one of the variants of the persona, the method born with the designer Alan Cooper in 1999; the purchase-oriented variant was developed above all by Adele Revella, founder of the Buyer Persona Institute, in the book Buyer Personas (2015).

Example. Two customers share an age, a role and a city. One walks into the shop after reading reviews, the other orders from a phone at eleven at night. To the target they are the same row; to the buyer persona they are two different people, and the site has to be designed so both reach the end.

The buzz is that murmur of excitement. Emanuel Rosen was the first to describe it systematically, in The Anatomy of Buzz (2000), which studies how word of mouth travels through networks of relationships. It works when it grows from a shared experience worth telling, because word of mouth influences buying decisions more than advertising; it differs from viral marketing, which relies on a single piece of content to pass around.

A surprise launch for a few guests, who tell their contacts about the event before the official announcement.

Example. On a price list page «Request a quote» works with people ready to buy, while those arriving from an informational search respond better to «Download the comparison of the three options». Rereading the archived versions of your own pages, and of your competitors', shows which calls generated clicks and which were ignored.

It is measured as a rate: abandoned carts over carts created. The most frequently cited causes are shipping costs discovered at the last step, mandatory registration and a complicated checkout. A high rate does not always point to a flaw in the site; on closer analysis it can reveal a different use of the cart, for example as a reminder to buy later in a store. That is why the figure should be cross-referenced with the other channels before acting.

Example. An e-commerce site has many full carts and few sales. Cross-referencing its data with the stores' data shows that abandoned carts come mostly from people living near a store, where they complete the purchase.

Example. A digital training course from 2010 becomes a case study because the examining board's minutes record a measurable fact: most of the internships continued as working relationships beyond the agreed term, something the agreements did not provide for. The document makes the result checkable by the reader, and the lesson transferable to another organization.

Enrico Giubertoni treats it as the real content of any technology project, because an organisation is a living system of structures, processes and cultures, and the work consists of creating the conditions in which it learns to evolve on its own.

Example. An AI assistant lands in the sales force and produces results only once the way a commercial request gets qualified changes too. Reps need sharpened skills for the role they hold (upskilling) and, for some, retraining toward a new one (reskilling). Without that work, the technology simply adds itself to the problems already there.

The model reads signals such as falling purchases, calls to customer service or lower use of the service, and gives each customer a probability of leaving. Born in telecoms and banking, where customers have a contract and churn shows on the day of cancellation, it is used today in every subscription business. It allows action before the customer leaves, with targeted retention, and its worth is measured by the customer lifetime value it saves.

A telecoms company calls the customers the model flags as at risk with a dedicated offer.

The first large experiment was South Korean: OhmyNews, founded in 2000 by Oh Yeon-ho with the motto "every citizen is a reporter". The American journalist Dan Gillmor theorised it in We the Media (2004), and in 2006 Jay Rosen spoke of "the people formerly known as the audience". The 2004 Indian Ocean tsunami and the 2005 London bombings were reported by witnesses before newsrooms; in Italy it happened in the hours after the 2009 L'Aquila earthquake. It arrives before professional journalists, but without their verification: that is why it lives alongside the risk of false news, and why newsrooms created teams dedicated to verifying user content.

In the hours after the 2009 L'Aquila earthquake the first updates came from residents, not from newsrooms.

In email it is read alongside click-to-open rate, which counts clicks on opened emails only and measures the strength of the content apart from that of the subject line.

Example. According to a 2023 Schema App analysis of its own clients' pages, pages that adopt structured data, the labels that tell search engines what a page contains, recorded a click-through rate 40% higher than pages without them. The same content, displayed better on the results page, gets chosen more often.

It is read alongside CTR (clicks over all emails sent) and open rate: a high open rate with a low CTOR means the subject line promises something the text does not deliver.

Example. A machine tool manufacturer sees that its technical newsletter has a good open rate but few clicks. CTOR shows the problem lies inside the message: the main link sat at the bottom after four paragraphs, and moved to the top it doubles the clicks.

Each message names the recipient and their problem, and usually belongs to a sequence of three to five emails a few days apart. It is a tool of outbound marketing, and in account-based marketing it often opens the first conversation with target accounts. It differs from direct email marketing in number and care: a few dozen messages written for one person against thousands of copies of the same message. The measures that count are reply rate and meetings booked, more than opens.

Example. An industrial process consultant writes to thirty plant managers in the chemical sector. Each email opens with a public fact about the plant, such as an announced expansion, and asks for a twenty-minute conversation: seven reply.

The expression entered management vocabulary with Judith M. Bardwick, whose book «Danger in the Comfort Zone» (1991) describes organisations in which security lowers commitment; behind it lies the Yerkes-Dodson law (1908), according to which performance grows with a moderate level of tension and falls when tension is missing or excessive. For marketing the concept works in two directions. A smooth buying experience creates a comfort zone for the customer that brings them back; a department that stays in its own stops seeing market changes.

Example. A sales network that renews the same contracts with the same customers every year hits its targets effortlessly, until a competitor takes away the three largest customers with an offer nobody had foreseen.

In Enrico Giubertoni's writing it has moved from owning something to knowing how to use it: with data now within everyone's reach, the advantage comes from the mastery with which algorithms turn it into decisions, and from how fast those decisions reach the market.

Example. A bank builds its fraud detection system in house, because that capability is its competitive advantage, and buys payroll processing from outside. The build-or-buy choice resolves into one question: which of the two could a competitor copy within a quarter.

The British psychologist Peter Wason brought it to light with his experiments in the 1960s; Daniel Kahneman, winner of the 2002 Nobel prize in economics, described its role in decision-making in Thinking, Fast and Slow (2011). In Italy Andreina Mandelli has worked on bias in marketing decisions and in our relationship with algorithms. In companies it leads people to read data as proof of the chosen strategy and to ignore the weak signals of change; social media algorithms amplify it, because they show each person what confirms their tastes.

Blockbuster's managers found in the data confirmation that in-store rental was irreplaceable, and gave no weight to customers' irritation at late fees and unavailable films.

In Enrico Giubertoni's articles the word names the leaning that someone or something carries around. In November 2007 the blog Ecoweb becomes Buzzes.eu, because its author's European connotation pushes him towards the .eu domains (the European suffix of web addresses) that had just been created. At SANA 2016, the Salone Internazionale del Biologico e del Naturale, the organic and natural products fair held in Bologna, his talk deals with the echo on social networks generated by natural-connotation cosmetics, where the word qualifies a product category. Much of the work on a campaign happens on this plane, and it often happens while people believe they are working on product features.

Example. A delicatessen puts two oils of the same origin in the window: one in a dark glass bottle with a rough paper label, the other in a clear plastic bottle. The literal level is identical, the connotation is not, and the price the customer accepts follows the connotation.

Treat it as paperwork to sort out at the end of the project and you discover it once the flow is already built and has to be rebuilt.

Example. A customer community works as a channel when the company holds the list itself, with consent collected and the identity of its members known. The same conversations living only on someone else's platform remain rented ground, and the day that platform changes its terms the relationship stops.

It ensures consistency of tone and quality, prevents duplicate or outdated pages, and keeps the site's structure intact over time even when people change. It is part of content strategy, a discipline made known by Kristina Halvorson with the book Content Strategy for the Web (2009).

Every page has an owner and a review date; when the date comes, it is updated or retired with a redirect.

In the articles the strategy starts from the problem to be solved and reaches the product at the end, and it is judged on indicators declared up front: when those indicators are missing, the count of published pieces becomes the only measure available.

Example. A machinery manufacturer publishes a technical sheet every week and counts the sheets. Rebuilt around the customer's problem, the same output changes shape: the sheets become answers to the questions reaching customer service, and the indicator becomes how many of those questions stop arriving by telephone.

Today the sector starts from selling inside a conversation on messaging apps or chats. It is set to spread through the personal shopping assistant built into e-commerce sites, which works as a new search engine based on dialogue.

Example. The case below is constructed. A tool manufacturer puts an assistant on its website that asks the customer which material they need to drill, with which machine and at what workload, and after the third answer suggests the two suitable bits with a link to order them. The catalogue is still there, and the customer reaches the product through the questions.

Its meaning changes with where the ratio is cut, so numerator and denominator get declared every time. It is one of the quantitative indicators of how much an audience moves, while the reasons it moves live in qualitative data.

Example. A retail network reads four indicators together: footfall, conversion rate, average basket and total sales. A drop in sales with stable footfall is a conversion problem, with stable conversion it is a traffic problem, and the combination points to the department where the cause sits.

It starts by locating the real barrier: sometimes it is doubt about the product, more often it is the path, and in that case sending more people at the same obstacle multiplies the loss.

Example. In an industrial services company on recurring subscription, business-to-business, over 50 million in revenue in its reference market, ninety-eight paid visitors out of a hundred left the site without even starting the enquiry form, and the conversion gap between phone and desktop was roughly twenty times. Perimeter: the core business line (the main activity, the one most of the revenue comes from) on the paid search network, direct measurement of 16 July 2026, first half of 2026 against first half of 2025. Conversion rate optimisation starts at that form and at the path leading to it, and an immersive experience added to the product page would have left that number where it was.

It carried online advertising measurement for twenty years and carries less of it today: browsers now restrict cookies set by parties other than the site being visited, so measurement shifts onto the data a company collects itself. Anyone reading an advertising dashboard today is reading an estimate, and it is worth knowing that before taking it into a meeting. In Europe cookies are governed by Directive 2002/58/EC on privacy in electronic communications (known as ePrivacy), amended in 2009 by Directive 2009/136/EC, the so-called cookie law: apart from strictly necessary technical cookies, a cookie may be placed only with the visitor's consent, which is where consent banners come from. In Italy the data protection authority, the Garante, updated its guidelines in 2021, requiring among other things that refusing be as easy as accepting.

Example. An online store sees a hundred conversions in the ad platform dashboard and sixty attributed orders in its management system. The gap has several causes at once, from cookies lost between one device and another to attribution windows the platform sets by itself (the period within which an order still counts as the effect of an ad), and the useful question is which of the two numbers goes into the commercial review, and who owns that gap.

The comparison that places it is with automation, where the criterion was chosen by whoever wrote the rule, and with an agent, which picks the next step by itself inside an assigned perimeter: here the person decides every single time. In Enrico Giubertoni's texts the word appears as a product name, Bing Copilot, next to Search GPT and Google Gemini, among the engines that read a site's structured data to build their own answers.

Example. On the shop floor a sales assistant queries a digital assistant about the model the customer is holding and gets availability, alternatives and warranty terms in three seconds. The system proposes, the person chooses what to say and owns the advice.

The author keeps the rights and allows everyone to use, modify and redistribute the work, on one condition: whoever redistributes it, modified or not, must grant the same freedoms under the same terms. The name plays on right and left: from «all rights reserved» to «all rights left». The practical consequence is that the work's freedom travels down the chain of derivative versions, and nobody can lock it inside a proprietary product. The best-known software example is the GPL; in Creative Commons licences the equivalent is the SA clause (Share Alike). Copyleft differs from permissive licences, such as MIT or BSD, which only ask for credit to the author and leave whoever reuses the code free to close their own version.

Example. The case below is constructed, because the term does not yet appear in the articles. A university chemistry lab publishes a safety manual under CC BY-SA; a chemical company adapts it to its own plants and shares it with suppliers: the adapted version carries the same licence, so anyone can improve it in turn.

Protection arises when the work is created, with no registration, and in the European Union it lasts for the author's life plus seventy years. The © symbol with the holder's name and the year shows who owns the rights; the phrase «all rights reserved» says every use must be requested. Anglo-American copyright and European author's rights differ in approach. The former mainly protects the economic exploitation of a work; the latter adds moral rights, such as the right to be credited as author, which stay with the author even after the economic rights are transferred. The international framework is the Berne Convention for the Protection of Literary and Artistic Works, and in the European Union the most recent reference is Directive (EU) 2019/790 on copyright in the Digital Single Market. The topic is back in the spotlight with generative AI: the site's articles list «AI bias in business and copyright» among the concerns of entry-level team members. Creative Commons offers a middle path, «some rights reserved», through which the author grants certain uses in advance.

Example. A financial institution has a generative AI write the text of a brochure: before publishing it, the institution checks that the text does not reproduce protected passages by other authors, the risk the site's articles link to copyright in AI adoption.

They count because they feed into ranking, and because a slow page loses the visitor before the content has had its chance.

Example. A manufacturer's site has image-heavy product pages that keep loading while the visitor is already reading, and the text jumps under their thumb. Phone visitors leave before the enquiry form, and the fault shows up in Core Web Vitals before it shows up in the number of enquiries.

It is both strategic and tactical: it helps set the ultimate goals of the campaign and provides the operational tools to reach them. It builds on the Pain/Gain model, the customer's pain points and gain points, to identify the values and content that differentiate the product and give the reason why, to be turned into an exponential approach.

Example. The case below is constructed. A sun-care brand starts from the customer's pain point, the fear of shiny skin in the city, and from the gain she is after, protection that stays invisible under make-up. From those two points it draws the reason why, the texture nobody can see, and on the eight steps of the CMP it builds a content calendar in which every post proves that single promise.

It is used mostly in digital advertising, where it is often also the way the campaign is paid for. It differs from cost per lead, which counts contacts, and from customer acquisition cost (CAC), which adds all sales and marketing spend for each new customer. A low CPA on an action loosely tied to revenue is only an apparent result.

Example. A machinery rental company spends three thousand euros on advertising and gets sixty quote requests: the CPA is fifty euros. Only six requests become contracts, and the figure that matters to management is the cost per signed contract.

Example. A components manufacturer watches its cost per click rise from one year to the next, with the same settings and the same agency. The cause is the number of competitors who entered the same auction: raising the bid buys space that is already crowded, and the useful move is to change the channel mix.

Example. Cost per lead improves meeting after meeting while the value of signed contracts stays flat. The line of configuration asking the system to optimise for the number of leads had sat there for months, doing exactly what it had been asked to do.

It exists to compare the price of advertising space across channels: it states what it costs to be seen, while what being seen is worth sits in other indicators.

Example. A retail chain compares two platforms and finds the same cost per mille. The decision moves elsewhere, onto which of the two brings people who then walk into a store, because the price of visibility is identical and the value of what follows is not.

Lessig, promoting the idea of a «free culture», raised core questions about copyright in the digital age. The practical answer is a shift from «all rights reserved» to «some rights reserved»: the author keeps the copyright and states in advance what others may do, so nobody has to ask permission case by case. The licences are built from four clauses: BY (attribution, credit the author), SA (Share Alike, derivative works carry the same licence), NC (Non Commercial, no commercial use) and ND (No Derivatives, no changes). These are the six licences: - CC BY: anything goes, commercial use included, as long as the author is credited; - CC BY-SA: as above, and derivative works stay free in the same way; - CC BY-NC: use and changes allowed, for non-commercial purposes only; - CC BY-ND: redistribution allowed, commercial too, but without changes; - CC BY-NC-SA: no commercial use, and derivatives carry the same licence; - CC BY-NC-ND: the most restrictive, the whole work is shared with credit, unchanged and without profit. CC0 stands apart: with it the author waives their rights as far as the law allows and moves the work towards the public domain. Every licence has its own badge showing the clause codes. Creative Commons licences are designed for text, images, music and data; for software the organisation itself recommends dedicated licences such as the GPL.

Example. This glossary is published under the CC BY-NC 4.0 licence: anyone can reuse an entry in university lecture notes or on a consulting firm's blog, as long as they credit the author and do not use it to sell a product.

It differs from up-selling, which offers a higher version of the same product. In e-commerce it took the form of the "customers who bought this also bought" suggestions made famous by Amazon; in a shop it works only if the complementary product is really in stock, which is where integration with the ERP becomes decisive.

Someone buying a camera finds the right lens and memory card already in stock in the shop.

On its own it prices the entry, while the margin that entry brings sits in another figure, customer lifetime value. The two are read together.

Example. A company sees cost per acquisition rise across every channel in the same quarter, and the first guess is that the agency is underperforming. When it rises everywhere at once the cause is usually market saturation, the number of competitors buying the same space: change the channel mix before changing the number.

In these articles it is a cultural outcome before a commercial one: the most advanced platforms leave churn untouched where a customer culture is missing. Predictive models earn their keep by showing who is about to leave while there is still time to call.

Example. A company applies machine learning to its sales data to predict which customers are at risk of churning, and gets a list of names to call. The list is worth as much as the organisation receiving it: with someone owning the call it becomes a negotiation, otherwise the model has merely forecast a loss.

It unifies what is already defined, and it stays powerless in front of ambiguity: if "lead" means two different things in two departments, the platform replicates that difference faster and at greater cost. The working order puts definitions first and tooling second.

Example. For Marketing a lead is someone who downloaded a guide; for Sales a lead is someone who asked for a quote. While both definitions live side by side, the platform delivers one unified profile that the two departments keep reading in two ways, and the monthly meeting argues again about which number is the real one.

Customers do not live it channel by channel; they judge it as a single experience. The idea that experience is itself an economic offering was set out by B. Joseph Pine II and James H. Gilmore in the article «Welcome to the Experience Economy» (Harvard Business Review, 1998). It follows that a flaw in a single touchpoint weighs on the judgement of the whole brand, and that improving it requires departments that work together.

Example. A customer finds a product available on the website, goes to the store and discovers it is not there. The product is the same, but the experience has worsened, and the negative judgement falls on the whole brand.

Enrico Giubertoni also uses it as the perimeter where damage gets measured, because friction at a single step interrupts the entire journey and hands the customer to whichever competitor works at that exact point.

Example. A lawyer buys from an online shop that breaks form autofill in her browser and then never shows her the order. The purchase is lost there, and with it goes the story she will tell about the brand: a technical fault at one step turns into a reputation problem.

It sets how much is worth spending to win that customer: it sits on the other side of acquisition cost, and with that figure alone in view spending gets judged on the price of the entry.

Example. A subscription services company finds that customers from one channel are cheap to win and cancel quickly, while customers from another cost twice as much and stay for years. Acquisition cost favours the first channel, lifetime value favours the second, and the decision comes from reading both.

Its value depends on how it gets filled in: a CRM rich in rows and poor in consistent fields feeds bad decisions as fast as it would feed good ones. The step change comes when salespeople stop being CRM form-fillers and become predictive hunters: the machine processes the noise, the person decides who to act on.

Example. A sales director opens the CRM and counts, across a hundred recent rows, how many have the source field empty or spelled in different ways. That share is the ceiling on every campaign analysis, because no dashboard can rebuild information nobody wrote down when the record was created.

Example. The case below is constructed. An electrical wholesaler draws up the customer persona of the installer who stops at the counter every morning on the way to the site: in a hurry, knows the product codes, wants the part ready. That profile produces ordering by message the evening before and a fast pick-up lane, and repeat sales grow because the service is built around the customer the business already has.

Example. A company runs three artificial intelligence pilots, each with its own dashboard, and by the end of the quarter no decision has changed. The dashboards describe, and reporting that stops at description is what Enrico Giubertoni calls Analytical Inertia.

It is worth building once the business knows which joins it actually asks for, and waiting for one before starting is the most common way to lose a year: a single promotion traced end to end on a defined perimeter produces a usable answer in weeks. That first trace tells you which data belongs in the warehouse, so it designs the warehouse better than any preliminary workshop.

Example. A sales director wants to know whether the discount on the flagship product moved channel revenue or merely pulled forward orders that were coming anyway. One promotion traced across twenty stores answers in three weeks, and that measurement names the tables from the management system that are genuinely needed.

A data-driven culture has three conditions: reliable data accessible to decision-makers, people able to read them, and the habit of changing one's mind when the data require it. The term spread from the late 2000s, together with the profession of data scientist (the specialist who draws knowledge from data), and machine learning multiplied its reach. The opposite risk is the wrong metric: steering by data that do not measure the result, such as vanity metrics, is worse than steering by instinct, because it gives false confidence.

A company stops choosing promotions by the sales director's taste and decides them by comparing, shop by shop, sales in the weeks when they ran with sales in the weeks when they did not.

The quality of the dataset decides the quality of the model: incomplete or skewed data produce wrong predictions and algorithmic bias. The case that showed how much this matters is ImageNet, the large archive of labelled images built from 2009 by the group of the researcher Fei-Fei Li at Stanford: on that data, in 2012, a neural network demonstrated the power of deep learning. In Europe the AI Act requires the datasets of high-risk systems to be relevant, representative and as free of errors as possible.

Three years of customer purchase history, flagged with who stopped buying, used to teach a model to recognise who is about to leave.

It works on two fronts: the company's decision-making process and the customer's, which in the articles on AI online chat is what a virtual assistant understands, supports and strengthens. The Italian articles use «processo decisionale», the English ones «decision making», and alongside both stands the adjective data-driven, used of the company whose decisions start from measurement rather than from the loudest opinion in the room. Speed is the delicate part: a slow decision-making process turns fresh data into a late observation.

Example. The customer management system flags that a long-standing client has stopped asking for quotes. If the signal reaches the salesperson only with the end-of-month report, the decision is formally data-driven and in practice arrives after a competitor's signature.

The difference with classic Machine Learning: there a person has to tell the system which variables to watch, such as age, purchase frequency or skin type, while Deep Learning finds them on its own, and for this reason it can process unstructured data such as images, video and voice. It is the technology behind skin analyzers, which read wrinkles, spots and pores from a selfie.

Example. The case below is constructed. A chemical company inspects the granules coming off the line with a camera. With classic machine learning a technician has to tell the system what to measure, such as diameter and colour; a deep learning model trained on an archive of images already classified by the technicians learns by itself which details separate a good batch from one to reject, including the fine cracks nobody had put on the list.

Negotiation happens here, communication happens one level above, and the two levels have to stay aligned.

Example. A pump's data sheet states flow rate, head and consumption: that is denotation. The catalogue that calls it «the silent heart of the system» works on another plane, and a salesperson who mixes the two ends up defending a promise with a number.

Many Italian users published their personal homepage there, alongside those who chose GeoCities, Xoom or Interfree. It belongs to the first season of the Italian web, when publishing a site was a personal act before a commercial one.

Example. A group of friends opens a page on Digiland for their five-a-side football team, with the league table updated by hand every week.

In Enrico Giubertoni's writing it lives on two levels: the historical one, born in the years of the open network and free software, and the corporate one, where it names the mindset that decides whether a transformation succeeds, because where internal digital culture stands still the clunky processes go on being tolerated inside the company just when customers have stopped tolerating them.

Example. Two companies in the same sector receive the same new tool. In the first, managers ask to test it on a real case within two weeks and report the failed attempts as well. In the second, everyone waits for instructions from above and goes back to the spreadsheet: the difference sits in the digital culture long before it sits in the budget.

In the legal sense it is the set of credentials a citizen uses to prove who they are when accessing online services. Here Europe has led the way: Estonia has had an electronic identity card since 2002, Italy has had SPID since 2016, and the European eIDAS regulation (Reg. (EU) 910/2014, updated by Reg. (EU) 2024/1183) provides for a digital identity wallet valid across the Union. It is one of the pillars of the European Digital Identity.

A person's professional profile on LinkedIn is digital identity in the first sense; the SPID login they use to download a certificate is digital identity in the second.

It entered into force in November 2022 and has applied since 2 May 2023. In September 2023 the European Commission designated the first six gatekeepers, Alphabet, Amazon, Apple, ByteDance, Meta and Microsoft, joined by Booking in 2024; since 6 March 2024 they must comply with its obligations. Among them: allowing users to uninstall pre-installed apps and to choose their default search engine and browser, not favouring their own services in results, letting users take their data with them and letting businesses access the data they generate on the platform. For a European company it means fairer conditions on the platforms it depends on to sell.

Example. Anyone switching on a new Android phone in Europe sees a screen asking them to choose the default search engine: it is one of the obligations the DMA places on Alphabet as a gatekeeper. For a manufacturer selling online the practical consequence follows: part of its customers search with an engine other than Google, so it pays to cover Bing and the other engines offered on that screen as well.

Enrico Giubertoni uses it to describe the ground everyone now works on: where digital maturity is widespread, technology stops telling one company from another and the contest returns to the customer relationship.

Example. Two suppliers in the same market both have the website, the automated assistant and the customer data in order: the technology they own is the same, so something else decides. The customer, better informed and alert to how their data gets handled, stays with the one that remembers the previous conversation and spares them the questions already answered. Where digital maturity is widespread, trust is what tells two equivalent offers apart.

The physical shop stops being a warehouse open to the public and becomes a hub of experience and logistics linked to the digital channels: it shows its stock online, ships web orders and handles pick-ups. It is the general term of which New Retail and smart retail are two variants.

A shop that shows its stock online, ships web orders from its own storeroom and lets customers collect app purchases at the till.

It entered into force in November 2022; since August 2023 it has applied to very large online platforms and search engines, those with more than 45 million monthly users in the Union, and since 17 February 2024 to all intermediary services. Platforms must let users report illegal content and give reasons for removals, say why an ad is shown and who paid for it, explain how their recommender systems work and offer at least one not based on profiling; profiled advertising aimed at minors or based on sensitive data is banned. Together with the Digital Markets Act it forms the package through which Europe chose to regulate the large platforms.

Example. Every very large platform must keep a public repository of its ads: a company advertising on Instagram finds its own ads there, with who paid for them and which audience they were shown to. If a third party posts illegal content about its product, the company reports it through the tool the platform is required to provide and receives the reasons for the decision.

Example. An Italian SME selling industrial spare parts online receives an order from Lyon. Since December 2018 Regulation (EU) 2018/302 has prohibited redirecting that customer to a national version of the site, or refusing their card, merely because they are French. It is a concrete piece of the Digital Single Market: one website sells to the whole Union on the same terms.

Enrico Giubertoni lists four pillars, and the first is the one the traditional list leaves out: culture, people, processes, technology. Company culture acts as the engine or the brake, and where it stays still even the largest technology investment returns little. The phrase took hold in the 2010s, when cloud, phones and data made digital a factor of the whole business model and no longer of the IT department alone. Europe has measured its progress since 2014 with DESI, the European Commission's Digital Economy and Society Index, and since 2022 has steered it through the Digital Decade programme, which sets 2030 targets for skills, infrastructure, businesses and public services.

Example. A company buys an advanced customer management system and leaves its salespeople's targets untouched, still measured on this month's contracts. Data gets entered at quarter end, all at once and from memory, and what remains of digital transformation is the licence fee.

It runs on a sending platform, the email service provider, and is measured with open rate, click-to-open rate and conversions. It works on large numbers, which places it at the opposite end from cold emailing, which writes to a few people one at a time. In Italian practice DEM mostly means promotional communication, while a newsletter carries content on a fixed schedule.

Example. A distributor of industrial chemicals sends its end-of-quarter promotion to ten thousand customers and contacts. The manufacturing segment opens at twice the average, and from the next send it receives a message written for that sector.

Before it each country protected them in its own way; the directive set a common definition and common remedies against anyone who acquires, uses or discloses them unlawfully. Information is a trade secret if it meets three conditions: it is secret, not known to people in the sector; it has commercial value because it is secret; the company has taken reasonable steps to keep it confidential. Italy transposed it with Legislative Decree 63 of 2018, which updated articles 98 and 99 of the Industrial Property Code. The third condition is the one most often missing: information shared with third parties without a confidentiality obligation loses its protection, even if nobody stole it.

An employee pastes the confidential price list into a free AI service, with no contract: the company risks no longer being able to defend that list as a secret.

It reaches people who are busy with something else at that moment, so the measurement that counts sits downstream of the ad, in the behaviour that follows it.

Example. A machinery manufacturer buys graphic ad space, the banners, on trade portals and judges it on impressions. Read instead against the quotation requests that reached the website in the following weeks, the same spend tells a different story, and the budget moves to the portals that produced conversations.

The GDPR makes it mandatory (art. 28) and sets its content: what the vendor may do with the data, how it protects them, to whom it may pass them and what happens when the relationship ends. For AI tools, the DPA is the document that says whether what is pasted in may be used to train the vendor's models.

Example. A components manufacturer adopts an AI service to analyse customer complaints. Before signing, it checks that the vendor's DPA excludes using the data for training and states where they are stored.

The GDPR (the EU General Data Protection Regulation) makes the role mandatory for public bodies and for companies that process data on a large scale or special categories of data, such as health data (arts. 37-39). The DPO advises and checks: decisions on processing stay with the company. That is why, when AI enters processes, the DPO is the person with whom a company settles which data may reach which tools.

Example. A regional bank wants an AI assistant to answer customer requests. Before switching it on, the DPO checks which personal data the tool receives, where the vendor stores them and which processing agreement protects them.

In the articles it is a channel to be engineered rather than a catalogue to be published: the experience of use decides whether a visitor buys, and the way an own shop tells itself apart from the large marketplaces is what keeps it standing next to them.

Example. A company measures that customers living near its stores fill the basket and abandon the payment, while customers further away complete the order online. Read by area, the same figure changes the decision: a basket abandoned two streets from the store becomes a click and collect offer, and the salesperson at the shelf enters the journey.

For a company, and for a system of companies such as Europe's, these are two innovation strategies, and both are opportunities. The early adopter creates new worlds and takes on the risk of a market that does not exist yet. The late adopter works with hindsight: someone has invented something and opened a path, but the invention still has flaws, so whoever comes next enters with the sturdy, powerful, refined version. It follows that coming second is a choice, and often it is the choice that wins the market. For Europe it means that an opportunity missed as an early adopter can be taken up again as a late adopter, with its own view of the world and the values of the European Union.

Example. The Nabaztag, launched in France in 2005, was a European case of early adoption: it was already a smart speaker (a speaker that talks and listens) a decade before the market asked for one. Apple is often a late adopter: the smartphone already existed in the devices of Research In Motion, later BlackBerry, which had their limits, and the iPhone spread the idea with a sturdier and easier version. In the same way Apple did not invent the computer, it invented a simple way to use one, and so it democratised digital technology.

It brings somebody back to where the content is, and it works when it arrives at the moment that content is useful to them: out of context, even a subscriber reads it as noise. It covers two different practices: direct email marketing, the same message to a large list, and cold emailing, individual emails to people who do not know the sender yet. The basic measures are open rate and click-to-open rate, and sending goes through a dedicated platform, the email service provider.

Example. A company announces a timed sale before it opens: a letter to subscribers previewing what will be on sale, social posts, a countdown and a collaboration with a voice from the sector. The anticipation is built in the days before, and when the sale opens the subscribers arrive ready to buy, because the date and the terms were spelled out to them in advance.

Brevo, Mailchimp and similar tools are ESPs. Beyond sending, it looks after delivery to the main inbox: it authenticates the sender and watches the reputation of the addresses. Many ESPs include marketing automation features, that is message sequences that start on their own based on what the contact does.

Example. A manufacturer of chemical laboratory equipment sends its newsletter from its own mail server and half the messages land in spam. After moving to an ESP with an authenticated sender, inbox delivery goes back above ninety per cent.

It is a quantitative answer to «how engaged is my audience», and the why belongs to qualitative data.

Example. A brand community measures success on followers, likes and impressions, and the numbers climb every month. Switched to active engagement, sentiment (the tone, favourable or hostile, of what the audience writes), member retention and the quality of interactions, the same community shows where the bond is genuinely relational and where it stayed transactional.

It knows what the company owns and where it is. It descends from the material requirements planning systems of the 1960s; the acronym ERP was coined by the research firm Gartner in 1990. Its historic leader is European: SAP, founded in 1972 in Walldorf, Germany, by five former IBM employees, is still the world's main producer. It differs from a CRM, which manages the relationship with customers and knows who they are and what they prefer: the ERP says whether the product is there, the CRM says whom to offer it to, and integrating the two is the precondition of Organizational Synchrony.

Example. In a retail project the company's management system was connected to the website, the Google Maps listings and the social channels, so the stock held in the ERP becomes the availability the customer sees online. Selling becomes more dynamic: products sitting in the warehouse can be pushed at the right moment, stock clears and revenue benefits.

Liability arrives first, visibility sixteen months later. It is the world's first comprehensive law on artificial intelligence, and its logic is risk: it bans practices of unacceptable risk, such as social scoring of citizens and manipulation that exploits people's vulnerabilities (bans in force since 2 February 2025); it places strict duties on high-risk systems, such as those used to select staff or grant credit; it requires transparency from limited-risk systems, such as chatbots and generated content; it leaves the rest free. Since 2 August 2025 the duties for general-purpose AI models, those behind generative assistants, have applied. Like the GDPR, it applies to anyone offering systems on the European market, wherever they are based.

Example. A company publishes a promotional video on its own site with an artificially generated voice and discloses nothing: the Article 50 penalty reaches 15 million euro or 3% of worldwide annual turnover and has been enforceable since 2 August 2026. The duties that would oblige the same company to know which systems run inside its own walls are different ones, namely Articles 26 and 27 on whoever deploys a high-risk system listed in Annex III, personnel selection among them, and they mature on 2 December 2027.

Enrico Giubertoni describes it as a project still taking shape, one that tries to balance innovation and rights.

Example. The case below is constructed. An Italian manufacturer picks a cloud provider with servers in the European Union, requires customer data to stay there and checks how the artificial intelligence assistant it adopts declares that it handles them. These choices show the European Digital Identity at the scale of a single company: technology adopted within rules grounded in values, alongside innovation.

The subject became central after 2020 for three reasons. The first is legal: in July 2020 the Court of Justice of the European Union, in the case brought by the Austrian activist Max Schrems (known as Schrems II), struck down the agreement governing transfers of personal data to the United States, showing that European data held elsewhere are not protected in the same way. The second is industrial: that same year France and Germany launched Gaia-X, the project for a European data and cloud infrastructure, and in 2023 the Chips Act (Reg. (EU) 2023/1781) aimed to bring semiconductor production back to Europe. The third is regulatory: through the GDPR, the Digital Markets Act, the Digital Services Act and the AI Act, Europe chose to exercise sovereignty above all by writing the rules that anyone operating in its market must follow.

Example. The case below is constructed. A manufacturer of industrial plants moves the archive of its customers' machine data to a cloud whose servers and operating company are in the European Union. After the Schrems II ruling every transfer of that data to the United States required case-by-case checks; by keeping it in Europe the company reduces its dependence on a non-European provider. It is European Digital Sovereignty at the scale of a single company.

It is not an isolationist path: it is an attempt to offer the world a model of digitalisation that is economically dynamic, socially inclusive and democratically accountable.

Example. The case below is constructed. An Italian SME adopting an artificial intelligence assistant moves within choices the Union has already made: the AI Act asks it for transparency on generated texts, the Digital Single Market opens sales in other countries, the competition rules for digital markets guarantee its access to the large platforms. These are pieces of the European Digital Strategy, and together they mark out the field in which the company competes.

The company controls design, production and distribution, and store sales data go back to the designers every day: the assortment follows demand instead of anticipating it. The expression spread in 1989, when the New York Times used it to describe Zara's arrival in New York. In the European Union the model was born and grew with Inditex in Spain (Zara, Pull&Bear, Bershka, Stradivarius) and with Mango, with H&M in Sweden and with Primark in Ireland, the latter almost only in physical stores; in Italy OVS and the Teddy group, with Terranova and Calliope, follow its pace. Outside the Union, in the United Kingdom, online-born fast fashion has its main names in ASOS and Boohoo. The other side is the environmental impact: the EU Strategy for Sustainable and Circular Textiles, presented by the Commission in 2022, aims precisely at reducing throwaway fashion.

Example. A garment that sells more than expected in some stores within a week goes straight back into production, while another that stays on the shelves leaves the assortment without waiting for the end-of-season sales.

His career can be traced through a few stages: - his training in the school of Silvio D'Arco Avalle, philologist and semiotician, from whom he takes the structuralist method; - the 1990s, when in Turin he set up the first research laboratories on hypertext (a text made of linked blocks that the reader moves through by choosing the links), on multimedia and on corporate communication on the World Wide Web; - the direction of the Interdepartmental Centre for Research on Communication (CIRCe) at the University of Turin, merged in 2026 into the SERTO centre, Semiotic Research in Torino; - the presidency of the Italian Association for Semiotic Studies (AISS), the scholarly society that brings together Italian researchers in the discipline; - the proposal of a «neoclassical semiotics», which brings the tools of the discipline together in a single framework, set out in Fondamenti di teoria sociosemiotica (Aracne, 2012) and in Teorie della narrazione (Carocci). In 2019 the Department of Philosophy and Education Sciences of the University of Turin dedicated the volume Semiotica 3.0 to him.

Example. In 1997 Enrico Giubertoni defended, with Guido Ferraro as supervisor, the thesis «External corporate communication on the World Wide Web»: a company website is read as a text, which communicates through the structure of its links as well as through its words.

Example. A retail company had handed its retail marketing to an outside agency, with a polished strategy, and the results were slow to come, because the work stayed on look and messaging. The block was in Financial Velocity. After the analysis the structural friction was removed by connecting the company's ERP (the system that runs orders, stock and invoices) to the website.

It is one of the few raw materials a competitor cannot buy, alongside your customer cases and your declared position on a contested question, which is why it decides the distance between two companies running the same artificial intelligence model: the advantage moves to what only you can feed it. The recurring mistake is leaving it scattered across separate systems, where every department owns a slice of the same buyer and nobody has the whole person in front of them.

Example. A retail chain keeps sales in its management system, enquiries in the website form and calls on a switchboard: three stores of first-party data describing the same customer, with nobody lining them up. The asset already exists; the work is recomposing it before buying another tool.

It works because it combines four levers: the urgency of the deadline, the scarcity of the stock, the exclusivity of access and the perceived value of the discount. The model was born in Europe: it was invented by Jacques-Antoine Granjon, who founded Vente-privée (today Veepee) in Paris in 2001 to sell fashion brands' surplus stock to its members without devaluing their price lists in stores. For a company it serves to clear unsold stock, grow the customer database and test new products. The risk is overuse: a brand that always looks on sale struggles to sell at full price and attracts customers who only buy at a discount.

Example. A furniture manufacturer has last season's collection in its warehouse. It opens a 48-hour flash sale reserved for newsletter subscribers, with a countdown on the page: the warehouse empties, new contacts come in and the store price list stays intact.

It sits at the other end of the pain point, and the work consists of turning the pain points that surfaced into gain points, concrete ones like better qualified leads and support for the sales force, or emotional ones like the experience the audience remembers.

Example. A sales network complains about poorly qualified leads, which is the pain point; the matching gain point is better qualified leads coming in, together with the tool chosen to select them. In the analysis of a marketing team the pain points photograph the current situation, the gain points say where the team wants to arrive.

The phrase was made known in 2025 by the report The GenAI Divide from the NANDA project at MIT, the Massachusetts Institute of Technology. The cause it identifies is neither model quality nor regulation but an outdated approach: systems that do not learn from use, do not retain user feedback and do not adapt to context, and therefore stay outside critical workflows. In a second sense the phrase refers to the gap between consumers who already use AI as a personal assistant and companies that see it only as an internal tool to cut costs.

A company buys licences for a generative assistant for every employee, but no process changes, and after a year use has stalled at writing emails.

It applies today, with no phase-in, so it also covers the personal data an employee pastes into an artificial intelligence service nobody in the company chose. Article 22 gives anyone subject to an automated decision three rights: human intervention, the chance to state their case, and the right to contest the outcome. It replaced Directive 95/46/EC, which each country had transposed in its own way, with a single law valid throughout the Union, and it also applies to non-European companies processing data of people located in Europe: that is why it became a model copied around the world. It rests on a few principles: lawfulness, fairness and transparency; purpose limitation; data minimisation, collecting only the data needed; accuracy; storage limitation; security; accountability, under which the company must prove it complies. Fines reach up to 20 million euros or 4% of worldwide annual turnover (Article 83). In Italy the supervisory authority is the Garante per la protezione dei dati personali.

Example. Screening job applications with a personal tool produces effects on real people with no oversight and no audit trail. The AI Act (the European regulation on artificial intelligence, EU Reg. 2024/1689) obligations for the Annex III high-risk systems, job screening among them, arrive on 2 December 2027, while the data protection regulation has applied for eight years: the question for the board is who in the company can state the origin, the owner and the legal basis of every field a model reads.

Two measurable effects run through Enrico Giubertoni's articles: the consumer who now puts generative AI alongside search engines when looking for a product, and the company that stays at the tactical level, chasing the perfect request, while the way it decides stays exactly as it was. The recurring risk is perceived credibility: confident, authoritative prose can carry factual inconsistencies, inventions or wrong citations, and readers raise their trust and lower their checking. The difference from analytical AI: that one classifies and predicts on existing data, the generative kind creates. The difference from Agentic AI: generative AI produces content and stops there, while an agent also uses tools to act.

Example. The Marketing department of a machine tool manufacturer has three hundred product sheets generated in a fortnight. Management sees the saved hours at once and sees much later that the safety instructions were never checked: the work moved from writing to verification, and verification had been assigned to nobody.

In the articles the concept arrives before the name: the fight for visibility moves to grounding, the anchoring of the answer to the data and content the model leans on, and the goal becomes how reliable the model finds your brand, so that the company's data and content become the basis of what the assistant tells the customer.

Example. A manufacturer discovers that the assistant recommends three competitors to anybody asking who supplies a given component, and that its own technical sheet exists only as an image. Rewritten as text, with structured data and quotable sources, the same information becomes citable and the company name enters the answer.

It was founded in 1994 by David Bohnett and John Rezner, initially under the name Beverly Hills Internet. Sites were arranged in thematic «neighbourhoods», and the address combined neighbourhood, street and house number, producing long and odd addresses. Yahoo! bought it in 1999 and closed it in the United States in 2009; part of the archive was saved by the volunteers of the Archive Team. For marketing it is an ancestor of user-generated content: millions of people learned to publish there, long before blogs and social networks.

Example. In 1997 Enrico Giubertoni and Erica Borgaro published their first personal website on GeoCities. The free space was 1 megabyte, later raised to 5, and at the time it seemed immense.

In photos the coordinates are written into the metadata (the hidden data that come with the file) following the EXIF standard, born in Japan in the 1990s; with GPS in phones and with platforms such as Flickr, which from 2006 let users put photos on a map, it became an automatic gesture. It helps find and sort content by place, and in marketing it makes content appear in local searches; it is also personal data, because it reveals where a person is, and in Europe it falls under the GDPR.

A smartphone with location turned on records the exact spot where the photo was taken in its metadata.

Inspiring organisations and leaders communicate from the inside out, starting from the Why, because decisions, from a purchase to loyalty to a brand, have a strong emotional and identity component.

Instead of "we sell management software", a company says "we want people who run a business to spend their time deciding, not filling in forms".

Example. The case below is constructed. A financial services firm decides who may upload customer data into an artificial intelligence tool, who approves a new use and who checks every quarter that the rules are followed. The document that holds these answers, with a person's name next to every role, is its governance: the names are what turn it from a statement of intent into a system that works.

It grants everyone four freedoms: to use the program, to study its source code (the instructions written by programmers), to modify it and to redistribute it. The condition is copyleft: whoever distributes the program, modified or not, must do so under the same licence and make the code available. Software licences fall into two families, and this is the distinction that matters for a company: - copyleft, requiring derivative versions to stay free as well: the GPL; the lighter LGPL (Lesser GPL), which allows a library to be used inside a closed program; the AGPL (Affero GPL), which extends the obligation to anyone offering the program as a network service; the MPL (Mozilla Public License), with copyleft limited to individual files; - permissive, asking only for credit to the author and for the licence notice to be kept: MIT, BSD and Apache 2.0, the latter with an explicit patent clause. The GPL applies to software the same principle Creative Commons expresses with the SA clause for text and images; Creative Commons itself, however, advises against using its licences for code, because the GPL covers aspects CC licences leave aside, such as the distribution of source code.

Example. The case below is constructed, because the term does not yet appear in the articles. A manufacturer embeds a GPL-licensed control program in a machine it sells, and modifies it: by delivering the machine to customers it distributes the program, so it must make the code of its changes available too. Had it chosen an MIT-licensed library, it could have kept that code closed.

It is the best-known form of washing, and the most regulated one: in Europe the directive on empowering consumers for the green transition (EU 2024/825) bans generic environmental claims that cannot be proven, such as «eco-friendly» or «sustainable» without recognised evidence. The word was coined in 1986 by the American environmentalist Jay Westerveld, struck by hotel signs asking guests to reuse towels "for the environment" when the real aim was to save on laundry. The general mechanism, with its variants, is described under washing.

Example. A company writes «green» on a product's pack because it changed the colour of the cap, while formula and supply chain stay the same. The claim does not survive a check, and from then on even the brand's true claims are read with suspicion.

Tools, however advanced, stay at the service of a human strategy that steers them; the point is to give equal weight to technological innovation and to understanding people. Its roots are in Don Norman's user-centred design (The Design of Everyday Things, 1988) and in design thinking; in Europe it has taken the form of Industry 5.0, through which since 2021 the European Commission has called for an industry that is human-centric, sustainable and resilient, a trait of the European Digital Identity.

Before bringing an AI assistant into customer service, a company observes which requests really frustrate customers and which ones the agents want to hand over.

Example. The case below is constructed. At a component manufacturer artificial intelligence drafts the reply to every request for quotation, with prices and delivery times. A technical sales engineer reads it, corrects whatever does not add up, and only then signs and sends it: the person remains the last step, and the responsibility for the quotation remains theirs.

In everyday English the word means insistent, exaggerated promotion. On this site hype is read in two ways: - as a signal: "behind all the hype there is a paradigm shift, a change that rewrites the rules of work". The noise, therefore, shows where to look; - as an expectation that deflates: it happened to the metaverse, still discussed but with far less hype than in previous years. It follows that a company's choice comes from analysing its own process and data, while the volume of the buzz remains information about the market. That is the invitation behind the strategic vision of AI: look beyond the hype.

Example. A bank reads every week that a new platform will transform business lending and wonders whether to adopt it at once. Looking beyond the hype means asking which process would change, which data it needs and how the result will be measured, before signing the contract.

Example. The case below is constructed. A bank offers each customer a savings plan calculated on their actual income and spending over recent months, with the instalment and the debit date chosen from that data. Classic personalisation speaks to a segment of similar customers; hyper-personalisation speaks to one person.

It describes companies, while the buyer persona describes the people who decide inside them. It is the first step of account-based marketing: the target account list is drawn from the ICP, and without a shared ICP marketing and sales chase different customers.

Example. A quality software supplier analyses its twenty most profitable customers and finds they are almost all manufacturers between fifty and two hundred people, certified and with one external audit a year. That profile becomes the filter for deciding where sellers spend their time.

It is also the first indicator a board demotes, because it reports how many people saw and leaves open the question of what came in.

Example. A company rebuilds its list of marketing indicators and starts from what it stops counting: raw impression counts, follower growth, un-contextualised likes. In their place come depth of engagement and the shift in brand recall. Impressions stay useful to the people running the campaigns, who decide where to act tomorrow, and matter little to the people deciding whether that budget exists next year.

The useful comparison is with multi-company courses, open to participants from different companies with the same syllabus for everyone. In-company training costs more per person, but it brings two advantages: - the material is real: exercises run on the company's own documents, data and workflows, so what people learn applies the next day; - people speak openly: with no competitors in the room, they bring their real problems, which are where change starts. This is why, when introducing artificial intelligence, the article pairs it with consulting: training gives people the method, consulting helps choose where to apply it.

Example. A manufacturing company that wants to use artificial intelligence to plan production brings the course into its own plant: exercises run on its orders and warehouse data, and managers discuss their problems without worrying that a competitor is listening.

The journey is measured by stage, from discovery to decision (the TOFU, MOFU, BOFU model), and a contact ready for sales is called a marketing qualified lead. It works well when the market is broad and buyers research on their own before talking to a supplier. With only a few dozen possible customers it sits alongside account-based marketing, which picks companies by name instead of waiting for them.

Example. A financial advisory firm publishes a monthly analysis of business lending rates and offers the full version to anyone who leaves a name and email. Within a year the list becomes the main source of meetings, and cost per lead falls as the content climbs the search results.

In the articles the choice is made through social listening, the analysis of what gets said online about a brand, which identifies who is already compatible with the brand and measures how much their community actually engages. The risk lives inside the same mechanism: for a fee, a voice can promote a company's environmental image without having checked the implications, widening the reach of a claim nobody has yet proved.

Example. A garden tools manufacturer compares two creators with the same follower count: one has talked about gardening for years, the other promotes any sector going. Listening shows that only the first has a community that discusses the products, and the contract goes there.

In 2011 Google's Jim Lecinski named this step the Zero Moment of Truth (ZMOT), adding it to the two moments described by Procter & Gamble: the first at the shelf, the second when using the product. The variant where research is online and purchase is in-store is called ROPO (research online, purchase offline). Today InfoCommerce is evolving into smart buying, with an AI assistant in place of search results.

A person reads reviews and compares prices on their phone, then goes to the shop already decided on the model.

Moving from separate applications to an ecosystem changes the value of the product: customers stay because everything is connected, and each extra service strengthens the others. Google with mail, calendar and documents, Apple with devices that recognise each other, and in Asia super-apps such as WeChat are its most complete forms. The flip side is lock-in: leaving is costly, which is one reason why Europe, through the Digital Markets Act, requires large platforms to let users take their data with them.

Mail, calendar and documents that refer to one another: an invitation in the inbox creates the calendar event with the document attached.

Example. The case below is constructed. A valve manufacturer posts a short TikTok video on how to spot a worn seal. The account has few followers, yet the video reaches maintenance technicians across the country, because the algorithm shows it to people who watch maintenance content: what drives distribution is the viewer's interest.

Together with Digiland it was the Italian alternative to the large US services such as GeoCities and Xoom. It belongs to the first season of the Italian web, when publishing a site was a personal act before a commercial one.

Example. A university student publishes his lecture notes on Interfree, and within a few months the page becomes a reference point for his fellow students.

It has many authors. The Briton Tim Berners-Lee, who invented the World Wide Web at CERN in Geneva and in 1993 had it released free of patents, so that it would belong to everyone. The American Lawrence Lessig, who in 2001 founded Creative Commons, the licences through which authors allow others to reuse and share their work. The Italian philosopher Luciano Floridi, whose concepts of the infosphere and onlife describe human life as now interwoven with the digital. In Italy one of the first avant-gardes was the Faculty of Communication Sciences at the University of Turin, in the 1990s, when the internet was not yet fully understood or accepted in the country. There the semiologist Guido Ferraro, founder of the Centre for Semiotic Research, set up from 1992, together with Enrico Giubertoni, then a student, several research laboratories on hypertext, multimedia and companies' external communication on the World Wide Web. The Turin sociologist Luciano Gallino, who after his years at Adriano Olivetti's Olivetti studied the relationship between technology, work and society and helped found CSI Piemonte, the public IT consortium of the Piedmont region. The Italian jurist Stefano Rodotà, who in 2015 chaired the commission of the Italian Chamber of Deputies that wrote the Declaration of Internet Rights. The European contribution is clear: putting the person, their rights and their dignity at the centre of technology is the thread linking these authors to the European Digital Identity.

Two professionals who have never met work together for years through an online group, exchange work under a Creative Commons licence and consider each other friends.

It is the first standard in the world on artificial intelligence management systems. It specifies the requirements through which an organisation that develops, supplies or uses AI systems establishes, implements, maintains and improves an AI management system over time: policy and objectives, roles and responsibilities, assessment of the risks and of the impact of the systems on people and society, controls on data, transparency and human oversight, monitoring and continuous improvement. It has the same structure as the other management system standards, such as ISO 9001 for quality and ISO/IEC 27001 for information security, and like these it can be certified by a third party. It does not replace the AI Act, but it helps to demonstrate that the risks have been managed in a structured way.

Example. The case below is constructed. A financial services firm that uses artificial intelligence to assess credit applications appoints a manager for its AI management system, assesses the risks of each model for customers, documents human oversight and seeks certification from an accredited body. To the business customer who asks how those algorithms are governed, it answers with a certificate.

In corporate communication it is what lets a website and a catalogue sound like the same company. When it breaks, readers struggle to say why, and meanwhile they stop trusting.

Example. An engineering company talks about precision on every page of its website, then sends out a promotion written like a supermarket flyer. Each text is correct on its own, and together they break the isotopy the reputation rested on.

The term covers both the discipline and the department in charge of it. On this site IT almost always appears in relation to other functions: "Marketing cannot operate in a silo from Sales, nor Sales from IT or Customer Service." The reason is that inefficiencies arise in the handovers between departments, and when IT finds out months later about tools adopted on their own by marketing and sales, the result is Shadow AI. For this reason systems integration and Organizational Synchrony between IT, marketing and sales weigh as much as the technology chosen.

Example. At a bank, marketing wants a new campaign tool: IT checks that it connects to the CRM, that data stay protected and that the vendor contract holds up, before the first email goes out.

For companies using AI, article 98 matters most: confidential information is protected as a trade secret only if the company has protected it with measures reasonably adequate to keep it secret. It follows that protection depends on the rules the company has set itself and can prove.

Example. A chemical company guards the formulas of its additives. If a technician pastes one into a public AI chat, and the company has never set a rule on these tools, it will be harder to argue in court that the formula was protected.

The word comes from Old French jargon, the twittering of birds, and Geoffrey Chaucer already uses it in that sense in the fourteenth century. The risk is a bias of the internal community: people who understand each other instantly believe they share a common language. The remedy is a definition that arrives where it is needed, at the moment the term appears.

Example. In a bank the risk department talks about «LGD» and «stage 2», and marketing nods without understanding. The resulting campaign promises customers something the risk department will never approve.

The code sits in a block invisible to the reader and uses the schema.org vocabulary, launched jointly by Google, Bing, Yahoo! and Yandex in 2011. The W3C (the body that sets web standards) published it as a recommendation in 2014; Manu Sporny is among its authors. It matters for marketing because it is how search engines and artificial intelligence read a site without having to interpret it: whoever describes their data precisely is understood, and cited, better than others.

Example. A chemical company adds a JSON-LD block to each product page with the name, safety data sheet and production site. When a buyer asks an AI assistant who makes that additive in Europe, the answer names the company.

Masaaki Imai introduced it to the West with «Kaizen: The Key to Japan's Competitive Success» (1986), and it is the daily engine of lean: lean says what to remove, kaizen says how to do it one step at a time. The principle holds well outside production. In a sales team kaizen is the short meeting after every lost deal, where one single thing to change in the next one is chosen. In marketing it is the single change to a page or an email, measured before moving on to the next.

Example. A chemical company selling through distributors asks every salesperson, on Friday, for the most frequent objection of the week. The answer prepared for that objection goes into the sales material on Monday: one improvement a week, compounding.

People buy products there, and for this reason it has become the country's flagship social network: anyone exporting cosmetics to Korea has to pass through it.

Example. The case below is constructed. An Italian manufacturer of coffee machines entering South Korea opens an official KakaoTalk channel before it even has a Korean website. There customers ask questions, book the demonstration and pay, and anyone who wants to give a gift sends the machine to a colleague straight from the chat.

Google introduced it in 2012, and since then search engines have reasoned more and more about things and less and less about strings of text. Its most visible output is the panel that appears beside the results when someone searches for a name. The site's articles put it this way: structured data «can contribute to Knowledge Graph panels and enhance brand visibility». The graph belongs to Google, so nobody writes it directly and no markup inserts it into a page: it is fed by sources that agree with each other, namely schema.org structured data on your own site, Wikidata, Wikipedia and official profiles. More broadly, the term names any database organised as nodes and relations, and AI platforms too use structured data to «build knowledge graphs» of their own.

Example. An industrial paint manufacturer describes its name, headquarters, founder and official profiles on its website with structured data, and makes them match its Wikidata entry: Google links the sources to the same entity and, when people search for the brand, shows the panel with the correct information.

In the published articles the lesson is a sibling of this one: a quantitative indicator tells you how much an audience moves, and the reasons it moves live in qualitative data.

Example. A hundred leads can be worth less than twenty. An indicator that counts leads generated leaves that difference invisible; the average value per signed contract shows it, and that is the number the finance department recognises as its own.

Generative answers change its job: the customer asks an assistant to summarize it, so the page now has to be written to be read by a machine that will quote it.

Example. A campaign sends a thousand visitors to a page listing twelve products with the full menu on top. Rewritten around one offer, with the form visible without scrolling and the navigation cut to the essentials, the same spend produces quotation requests instead of visits.

It is the engine beneath conversational assistants such as ChatGPT and Gemini, and it recurs in Enrico Giubertoni's articles for one commercial consequence: a growing share of people now take to those assistants the informational searches they used to run on a search engine. One limit stays in view: the model learns from the data it was trained on, so it inherits that data's bias along with its knowledge.

Example. A components manufacturer discovers that its technical buyers arrive from a conversation with an assistant that names three suppliers, and this one is absent from the list. The question for management changes shape: from how to climb the results page, to how to put reliable company material where the language model can quote it.

Example. A business-to-business supplier replaces its downloadable brochure with a spreadsheet that estimates what unworked leads cost. Whoever fills it in arrives at the phone call holding their own number, and the conversation starts one step further along.

It was born in the Toyota Production System, built by Taiichi Ohno between the 1950s and the 1970s. John Krafcik coined the word «lean» in 1988, and James Womack and Daniel Jones made it famous with «The Machine That Changed the World» (1990) and «Lean Thinking» (1996). From the factory the method spread to every business function, and each extension took its own name: Lean Startup for new products, Lean UX for design, Lean Marketing and Lean Sales for customer acquisition, Lean AI for growth driven by artificial intelligence. The basic tools are still two Japanese words: muda, the waste to remove, and kaizen, improvement in small steps. Enrico Giubertoni uses lean as an example of a word that splits departments: on the factory floor it names a method, and in marketing one letter is enough to hear lead.

Example. An industrial paint manufacturer maps the flow of an order, from the customer's request to delivery. It finds that the product sits in the warehouse longer than it takes to make it: the first waste to remove is that wait, before the production line itself.

The term comes from Lomit Patel, author of «Lean AI: How Innovative Startups Use Artificial Intelligence to Grow» (O'Reilly, 2020), who applies it to growth marketing (marketing driven by measured growth): artificial intelligence runs in a few hours the tests a team would run in weeks. The lean point stays the same as ever: automation amplifies the process it finds. Applied to a process full of waste, artificial intelligence produces waste faster; that is why the first step is removing the muda, and the second is automating.

Example. A laboratory equipment manufacturer uses an artificial intelligence system to test dozens of ad variants on an audience of technicians. Before switching it on it checks that the leads generated get called back: otherwise the system multiplies the leads, and sales works the same share of them as before.

Alistair Croll and Benjamin Yoskovitz set it out in «Lean Analytics» (2013), with the principle of the One Metric That Matters. The book separates the metrics that drive a decision from vanity metrics, the numbers that grow and feel good while saying nothing about what to do. For marketing and sales it is an antidote to dashboards crowded with indicators: the metric changes with the stage, and at every stage there is one that decides.

Example. An accounting software company, having just launched a product, measures for one quarter only the share of customers still using it after thirty days. As long as that number stays low, sending more traffic to the site means sending more people to a product that loses them.

Ash Maurya created it in 2010 by adapting Alexander Osterwalder's Business Model Canvas, and describes it in «Running Lean». Compared with its parent model it moves the focus from the company to the market: the boxes on partners, activities, resources and relationships give way to problem, solution, metrics and advantage. That is why it works well as an alignment tool between marketing and sales, since it forces you to write in one line the problem being solved and the metric that says whether it is being solved.

Example. The sales director of a packaging manufacturer asks marketing and sales to fill in the Lean Canvas for a new line separately. The two sheets agree on channels and disagree on the customer's problem: that is where the discussion belongs, before the campaign.

The tool is the factory's, following the value stream and removing waste, applied to the customer's journey instead of the product's. The most recent reference is Allan Dib's «Lean Marketing» (2024). In practice it means a few channels run well instead of many run badly, content measured on its contribution to revenue, small and frequent experiments instead of large and rare campaigns. The step that makes all the rest possible is measurement: a waste becomes visible only when you know which activity brings the customers.

Example. A machinery manufacturer runs many social channels, two trade fairs and a newsletter. Measuring where last year's signed contracts came from, it finds that almost all customers came through two channels: it concentrates the budget there, and the rest becomes an experiment to justify.

Typical wastes are the waits between one contact and the next, proposals written from scratch every time, weak leads worked as if they were qualified, information asked of the customer twice. Lean Sales grew out of the practice of sales teams, without a recognised founding text like those of the other entries in the lean family: that is why the term circulates with close and sometimes different meanings. The fixed point is the yardstick, which stays the lean one: every step is worth the value it has for the buyer.

Example. The sales team of a construction chemicals company measures the time between a request for a quote and sending it. Most of it goes into searching price lists and data sheets across different folders: a single quote template, with the data already linked, shortens the deal without touching the price.

Six Sigma was born at Motorola in the mid-1980s, and the name refers to a process that fails fewer than 3.4 times in a million. Michael George made the combination systematic with «Lean Six Sigma» (2002). The method runs in five phases, summed up by the acronym DMAIC (Define, Measure, Analyze, Improve, Control). In marketing and sales it is useful when a result swings for no visible reason: conversion rates that change from month to month, response times to enquiries that vary from one salesperson to another. Lean makes the process shorter, Six Sigma makes it predictable.

Example. A mechanical components manufacturer notices that quotes go out sometimes in a day, sometimes in three weeks. Using DMAIC it measures the path of every quote, finds that the delays come from requests with incomplete technical drawings, and introduces a form that makes them arrive complete.

Eric Ries set it out in «The Lean Startup» (2011), taking lean from the factory to innovation. Its core is the build-measure-learn loop: you build an MVP, measure how real customers react, and learn whether to persevere or change direction, with what Ries calls a pivot. For marketing and sales it changes the order of things: assumptions about the customer are tested in the market before investing, and every campaign becomes an experiment with a precise question. Enrico Giubertoni brought the method to the room in 2015 with a talk at Lean Startup Machine Milan, the workshop format born in New York, and presented it on the OkRadio programme «Aperitivo Aziendale».

Example. A financial advisory firm is considering a subscription service for small businesses. Before building the platform it publishes a page describing the service, with the price and a sign-up button, and measures how many people press it: the market's answer, before any investment.

Instead of long design documents approved once and for all, the work runs on hypotheses: you state an assumption about user behaviour, build the lightest prototype that tests it, and observe. Jeff Gothelf and Josh Seiden set it out in «Lean UX» (2013). In marketing it is the method for designing landing pages, forms and purchase paths: every design choice is a hypothesis about conversion, and it is tested with real people before it becomes final.

Example. A regional bank wants to simplify its online loan application. Instead of redesigning the whole path it states a hypothesis (people drop out when asked for documents), prepares two versions of that single step and watches where people stop.

He belongs to the line of thought that set out to define an ethical and legal framework for the emerging information society. Two ideas make him recognisable: - «Code is law», from the 1999 book «Code and Other Laws of Cyberspace»: online, people's behaviour is regulated by software architecture as much as by laws, so whoever writes the code also writes the rules; - «Free Culture», from the 2004 book: copyright that is too rigid holds back creativity, which feeds on reuse. It is the same idea Lessig had already put into practice in 2001 by founding Creative Commons, with its «some rights reserved» licences: the book gives it its theory three years later. In March 2010 Lessig gave the keynote lecture of the «Internet è Libertà» (Internet is Freedom) conference at the Italian Chamber of Deputies, recounted in an article on the site's Italian blog: you cannot understand what the internet is by thinking about single applications.

Example. A bank decides its app accepts transfers only after a double identity check: no internal rule imposes it, yet the code makes it impossible to bypass. This is Lessig's «code is law», the rule written into the program.

Between two departments the next step is needed: alignment rests on written boundaries and on the metrics each function is judged by, and that is where the work happens.

Example. Marketing and Sales blame each other: the first is measured on lead volume, the second on closed revenue, and the stretch that joins the two numbers appears on no objectives sheet. The meeting between the two managers produces cordiality and each one goes back to their own piece of the process; changing the shared metric changes the behaviour of both departments.

The company listens before it speaks: it collects questions, complaints and wishes, and uses them to improve products and services. Its root is the Cluetrain Manifesto (1999) by Rick Levine, Christopher Locke, Doc Searls and David Weinberger, whose first thesis reads "markets are conversations". In Italy the idea took the form of a manifesto of companies practising listening and relationship marketing, promoted by the blog Aziende con le Orecchie (Companies with Ears).

Every week a company reads the questions that reached customer service and turns them into website topics and product changes.

It was born in China, where Alibaba launched Taobao Live in 2016, the same year Jack Ma announced New Retail; from there it spread with TikTok Shop and live streams on social networks. It is an onlife form of the old TV shopping: the experience is that of a store with a sales assistant, but it happens online and leaves measurable data on who watches, who asks and who buys.

Example. A cookware manufacturer runs a live stream every Thursday in which a chef cooks with products from the catalogue. During the stream a purchase link appears, and the products used in the recipe sell within the following half hour.

It is a proposal published in 2024 by Jeremy Howard, and it remains a voluntary convention with no body governing it. The file is written in Markdown (a way of formatting text with a few symbols, readable by a person too): a heading with the site name, a short summary and then lists of the pages that matter, each with one line of explanation. HTML pages are often published alongside Markdown twins, free of menus, ads and code, so the model reads the content and spends fewer resources cleaning it up. It complements the other files at the root: robots.txt tells crawlers what they may visit, the sitemap lists every address, and llms.txt points to the pages that explain the site best and the order in which to read them.

Example. The case below is constructed, because the term does not yet appear in the articles. A financial advisory firm puts its name, two lines on what it does and the list of its service pages in /llms.txt, each linked to its Markdown version: an assistant answering a question about its sector finds clean text to quote right away.

It is a subset of artificial intelligence: artificial intelligence is the ambition of making machines display human cognitive faculties, machine learning is the most widely travelled road towards it. Where traditional programming asks for precise rules for every action, here the system is handed large volumes of data and algorithms that find the regularities inside them.

Example. A retail chain gives the system three years of till receipts, and the system isolates micro-segments nobody had ever defined: those who buy only during the sales, those who come back every six weeks, those who stopped after the rebranding. The rule «customer who returns every six weeks» was written by the system reading the data, and far sharper offers follow from it.

Enrico Giubertoni turns it into a rule of calculation: payback is computed on margin rather than on revenue, because a project worth a hundred thousand euro of revenue at a thirty per cent margin returns thirty thousand a year, and on payback time the difference is a factor of three. In the published articles the term appears as marginality, where the loss of marginality is the price of waiting.

Example. A sales team closes a hundred thousand euro order by granting a discount that eats the contribution margin. The number lands in the turnover everyone reads in the management meeting, and the cash available to hire one more salesperson is exactly what it was last month.

Enrico Giubertoni measures it at the hand-offs rather than in meetings: marketing answers for lead volume, sales for closed revenue, and the stretch that joins those two numbers sits outside both objective sheets, so nobody guards it.

Example. Marketing hands sales a contact who has already asked for a quote, and the context stays behind: what they had read, how urgently they had written. The rep calls back three days later and starts again from the first question, so the same qualification gets paid for twice. Neither department got its own piece wrong: what the piece in the middle lacks is a name who answers for it, a deadline for picking it up and evidence that it happened.

The outcome depends on how the company designed its own flows before buying the tool: two companies running the same platform get different results, because each poured its own definition of a lead and of a conversion into it.

Example. The platform records a lead when somebody downloads a document, while the sales network only calls it a lead once a quotation has been requested. In the boardroom the two numbers contradict each other and the argument about attribution, meaning which channel takes credit for a sale, turns into one opinion against another. Redraw the boundary between Marketing and Sales first, then configure the platform.

It is measured on the whole company: several people from the same business reading content, attending an event or visiting pricing pages. It is the account-level version of the marketing qualified lead, which looks at a single person. In business-to-business sales a group decides, and one interested contact says little about a purchase that involves five or six roles.

Example. In the same month three people from a banking group, in finance, IT and purchasing, download different documents from the same supplier. None of them alone passes the scoring threshold, the company as a whole does, and the seller is alerted.

The judgement usually rests on a score that adds who the person is (role, company) and what they did (downloads, visits, replies). It comes from inbound marketing and is the first rung of a ladder: sales accepts it (sales accepted lead), checks it and qualifies it (sales qualified lead). The criterion has to be written together with sales, otherwise marketing delivers volumes that sales judges empty.

Example. A company providing financial services to businesses gives ten points to anyone who downloads the tax credit guide and twenty to anyone who visits the pricing page. Above fifty points, with a finance director role, the contact becomes an MQL and goes to sales.

In the articles it is the benchmark for every own shop: its strength is logistical efficiency, and a badly served customer moves there accepting even a non-original product, as long as the purchase runs smoothly. The power of the largest platforms is now regulated in Europe: the Digital Markets Act imposes obligations on the platforms that control access to the market, the so-called gatekeepers, to keep those markets fair and contestable.

Example. Amazon's arrival in Italy sped up the take-up of e-commerce, raised service standards and opened selling through its marketplace to many European businesses. The balance is bittersweet: merchants went online, but the market they sell on depends on a platform that is not European, and what is missing is the digital sovereignty that means European digital identity, not nationalism. In the Netherlands and Belgium the European platform bol.com is the leading online store and holds its own against Amazon, which only arrived with a Dutch site in 2020; Zalando, born in Berlin, did the same in fashion. China, with Alibaba's New Retail, gave commerce its own paradigm: now it is Europe's turn.

It mirrors the physical world, with streets, shops and meeting places, but removes its limits of space. The word was born in 1992 in Snow Crash, a science-fiction novel by the American author Neal Stephenson, which imagines an endless street where people live a second, digital life; the first mass attempt was Second Life (2003). The term came back to the centre in 2021, when Facebook renamed itself Meta. Among the countries that invested most is South Korea, which made the metaverse a national plan and opened virtual public services, such as the capital's Metaverse Seoul. Where social networks tell physical life as it happens or afterwards, the metaverse is a place where the experience itself happens. Outside gaming, however, this kind of experience has not yet managed to become a stable part of companies' marketing processes.

A fashion brand opens a shop in a virtual world: avatars try on digital garments and can order the real ones, delivered to the door.