Companies in the AI era: from tax entity to cognitive subject

In short. Every company is a tax entity. In the AI era it needs an upgrade and must also become a cognitive subject: it has sensors that pick up the market, an algorithm that turns signals into actions, and people who separate operational actions from strategic ones. The decisive point is tailoring. A system copied from a recipe makes a company like every other, which is to say a commodity (a product chosen on price alone). A system built on the company’s identity sharpens its values and its proactive strength.

From tax entity to cognitive subject

To the state, a company is a VAT number, a balance sheet and a calendar of deadlines. The description is accurate and incomplete. A company watches the market, remembers what it has learned, decides and corrects itself: it behaves like an organism with a nervous system.

Until recently that nervous system lived in a few people’s heads. The seasoned salesperson “felt” that a customer was about to leave; the owner “knew” a season would be hard. It was precious and fragile knowledge: unmeasured, untransferable, and gone when its owner left.

AI changes the economics of that knowledge. Reading thousands of signals a day, comparing them with a threshold and triggering an action used to cost too much; today it costs little. That is why the company must become a cognitive subject, and a dynamic one: its nervous system changes as the market changes.

The idea has a long history. In Adaptive Enterprise (1999), Stephan Haeckel described the “sense-and-respond” company as opposed to the “make-and-sell” one. David Teece (2007) called a firm’s ability to sense opportunities, seize them and transform itself dynamic capabilities. What was then theory for large groups is now within reach of a hundred-person company.

Three layers: sensors, algorithm, people

A cognitive subject rests on three layers, which most companies keep tangled together.

1. Sensors pick up the market. They are the figures that move: inbound requests, lost quotes, response times, customer questions, payments.
2. The algorithm is the system of values, decisions and automatic actions the company gives itself. It sets which threshold matters, what happens when it is crossed and which action starts without waiting for anyone. The word sounds more intimidating than it is: an algorithm is a written rule, and every company already has many, only implicit.
3. People read the outcomes and sort actions into two families: operational ones, which can be automated, and strategic ones, which stay with whoever decides.

A complete sensor has four parts: a data point, a threshold, an automatic action and a person who decides. Remove one and what is left is a report: people read it, discuss it, and the market moves on.

Smarketing: where sensors matter most

In my glossary, Smarketing (sales plus marketing) is the practice of making sales and marketing work as a team toward the same revenue goal, with the same data, the same definition of a qualified lead and a constant flow of outcomes back.

It is where a company either hears the market or stays deaf to it. Marketing sees the weak signals, sales sees the strong ones, and they usually look at them at different times and with different numbers. Every marketing action and every sales action should have its own radar.

An example. New requests in the CRM (the system that tracks contacts and deals) start falling. Without sensors, the drop surfaces in the quarterly report and the response comes months later. With a sensor, when requests stay below the threshold for two weeks in a row, the algorithm alerts sales and marketing together, launches a pre-written, pre-approved reactivation campaign, and asks a person whether the drop is an accident or a sign that the market is shifting.

Every data point is a sign

There is a deeper way to say why every action in a cognitive company has a precise reason, and it comes from semiotics (the study of signs and how they produce meaning).

The philosopher Charles Sanders Peirce defined a sign as something which stands to somebody for something in some respect or capacity. It is a three-part definition, and it fits the company almost literally:

• the sign is the sensor’s data point: the lost quote, the silent customer, the recurring question;
• the object is what lies behind it: the customer’s need, the market shifting;
• the interpretant is the meaning that data point takes on for that company, and this is where the algorithm and people do their work.

Peirce called indices the signs tied to their cause, as smoke is to fire. Sensor data are indices: a lost quote is the smoke of a need we failed to grasp. The algorithm is how the company reads those signs, and every action comes from that reading. That is why it has a precise reason, like a tax ID: a unique identifier stating which signal it comes from and which rule set it off.

The consequence matters for the rest of this article. The same sign means different things to different companies, because the interpretant changes. A quote lost on price says one thing to a company that sells quality and another to one that sells volume. That is why a copied algorithm reads the signs through someone else’s eyes.

Seven types of sensors

Sensors group by what they measure. The seven types below apply to almost any company that sells to other companies; the thresholds are examples, and their right value depends on the company.

Seven types of sensors

Type	What it measures	When it fires (example)	What a person decides
Demand	Inbound requests in the CRM	Below threshold for two weeks	Whether the drop is seasonal or structural
Outcome	Lost deals and their reasons	Same reason three times in a month	Whether to change offer, price or story
Speed	Time to first reply	A request idle for 24 hours	Whether the delay is a one-off or a process flaw
Language	Questions and objections on sales calls	A new question that keeps recurring	Which answer to give, in which voice
Loyalty	Order frequency of active customers	Silence beyond twice the usual interval	Whether to call, with what proposal
Cash	Payment timing	A punctual customer paying late	Whether it means difficulty or dissatisfaction
Relationship	Contacts' role changes	A contact moves company	Whether and how to reopen the relationship

Two observations. First, some types sit in finance (cash), others in customer service (language), and all of them speak about revenue. A nervous system crosses functions, which is why it has to be designed by leadership rather than by a single department. Second, the most important column is the last one. No tool can buy it for you.

Why a generic recipe turns a company into a commodity

This is the point I care about most, and I say it from experience: if you take this article literally, it will make your company generic.

A table of seven sensors with ready-made thresholds and actions looks like a recipe. Copy it, install it, it works. The trouble is that every competitor can copy the same recipe, and a company that senses the same things, with the same thresholds, and reacts with the same actions ends up behaving like all the others.

Organisational sociology described this long ago. Paul DiMaggio and Walter Powell (1983) called it mimetic isomorphism: under uncertainty, organisations imitate those that look successful, and a whole field grows ever more alike. In Different (2010), Youngme Moon showed the same phenomenon from the market side: when competitors chase each other on the same features, differences fade and customers choose on price. That is the definition of a commodity.

AI raises the risk, because the tools are the same for everyone and so are the starting templates. If your algorithm comes from a generic template, it thinks the way everyone thinks.

Something very different happens when the work starts from the company’s identity. Which signals matter to you depends on what you promise customers. A company that sells reliability sets its speed threshold far lower than one that sells customisation. A company built on long relationships weighs loyalty above demand. Thresholds, actions and above all the decisions left to people are the operational translation of values. Built this way, they sharpen what makes the company recognisable and strengthen its proactive strength: the ability to go to a customer with a proposal instead of waiting for a request.

Every car has more than two wheels. Nobody chooses a car for its wheels.

People: talent counts exponentially

One might ask whether, with sensors and an algorithm, people’s talent matters less. That is a myth. In reality talent is fundamental, and its effect is exponential, because it acts twice and the second time multiplies the first.

The first time is in designing the algorithm. Deciding which sensors are needed, where to set thresholds and what to automate requires deep knowledge of the company: its sales cycle, customers, margins and values. An algorithm designed with talent works for that specific company; a generic one is worth little.

The second time is in correction, and there are two kinds:

• automatic corrections, which the algorithm applies on its own within rules already written;
• manual corrections, where a person rewrites the rules: changes the threshold, the offer, the algorithm itself.

The first make a company react. The second make it learn. Chris Argyris and Donald Schön (1978) called them single-loop and double-loop learning: in the first you correct the action, in the second you correct the criterion that produced it.

That is why the effect is exponential. Good design without good correction ages; good correction on a generic design improves a copy. When talent enters both moments, every rewrite starts from a base already shaped for that company, and the advantage compounds round after round. AI multiplies whatever it finds: give it talent that designs and corrects, and it multiplies talent.

Working in depth

With companies I always start from the same place, which is also the hardest: mapping internal processes. Where an opportunity is born, where it stalls, who decides and with what information, and which values guide those decisions even when nobody writes them down.

Only then do we choose which types of sensors are needed, set the thresholds and decide what to automate. That is the moment AI stops working for a generic company and starts working for yours.

Working in depth takes longer than following a recipe. In return it produces a system no competitor can copy, because it is made of your identity.

If you want to understand which sensors your company needs and how to write an algorithm that is truly yours, tell me about your priorities.

Share your priorities