AI in Marketing: A Cognitive Processor, Not a Gadget

A cognitive processor is something that amplifies a company’s thinking, meaning it helps the company see more, compare better and decide faster. A gadget is something that executes an isolated task and impresses in a demo. Almost every company buys AI as the second thing, then wonders why it does not get what the first one promised.

TL;DR. The data shows a clear break between adoption and value. 44% of organisations have taken AI to enterprise scale, while only 37% report any impact on operating profit, and a 2025 MIT study found that 95% of organisations get no return on their generative AI investment. The cause is not the model. It is that AI gets asked to execute tasks inside a company that has never written down its knowledge. The correct role for AI is to process the company’s context, and that context has to exist in writing first.

What does “cognitive processor” mean and how does it differ from a tool?

A tool does one thing you ask of it, once. You ask for an ad text, you receive it, you use it, and tomorrow you start from zero. Its value is consumed at each use and accumulates nowhere.

A cognitive processor does something else. It receives the company’s full context, meaning what the company knows about its clients, what it has decided so far and the rules it works by, then produces something consistent with all the rest. The practical difference becomes visible in the second month, when a second person asks for the same piece of work and receives a result compatible with the first rather than one that vaguely resembles it.

This distinction is not theoretical, because it decides where the money goes. If AI is a tool, then the investment goes into subscriptions and prompt-writing courses. If AI is a processor, then the main investment goes into writing down the company’s knowledge, and the subscription becomes a secondary cost.

Ethan Mollick, professor at Wharton, describes his own practice in a way that shows where the value sits. “I give the AI a task. I use my expertise to decide what that task is. I evaluate the results, I correct my approach, and if it still does not work, I do it myself.” Choosing the task and evaluating the result both stay with the person, and both require context rather than technical dexterity.

What does the data show about companies that have already tried?

It shows a large distance between how much AI gets used and how little of it appears in the results.

The McKinsey report published in 2026, built on 1,719 respondents from 97 countries, shows that nearly nine in ten organisations use AI regularly in at least one business function, and 44% have taken it to enterprise scale, up from 38% a year earlier. Over the same period, the share reporting an impact on operating profit stayed at 37%, practically unchanged. Companies scale faster than they capture value, and the gap between those two curves is the subject of this article.

The MIT study published in July 2025, built on 52 interviews, more than 300 public initiatives and 153 leaders, states the conclusion more harshly. Despite 30 to 40 billion dollars invested, 95% of organisations get no return, and only 5% of integrated projects extract real value. Gartner had already anticipated the pattern, estimating in 2024 that at least 30% of generative AI projects would be abandoned after the proof-of-concept stage, because of data quality, costs and unclear business value.

Rita Sallam, distinguished analyst at Gartner, describes the situation without softening it. “After last year’s hype, executives are impatient to see returns on GenAI investments, yet organisations are struggling to prove and realise value.”

The most useful figure of all comes from BCG, in the shape of a rule for allocating effort in an AI transformation. 10% of the effort goes into algorithms, 20% into technology and data, and 70% into people and processes. The companies that fail invert the proportion, putting nearly all the effort into choosing the tool.

Why, concretely, does an AI project fail inside a company?

For three reasons that almost always appear together.

The first is the absence of written context. The company asks the model to produce something without ever having told it who the client is, what has been promised so far and what it is not allowed to say. The model fills the gaps plausibly, and the result is a grammatically correct text that does not resemble the company.

The second is the absence of verification. A generated result always looks finished, whether or not it is correct, and people start accepting it because it looks good. In a company without a gate calibrated on bad cases, errors enter deliverables silently, and trust collapses all at once, at the first visible incident.

The third is choosing the wrong task. What gets automated is what is easy to automate, meaning content production at volume, instead of what hurts, meaning the time lost rebuilding context at every piece of work. The result is more material, with the same lack of clarity, produced faster.

The common pattern is that all three appear before the model, in the way the company works. This is why the BCG rule with 70% on people and processes describes reality better than any comparison between models.

Which tasks are worth giving to AI and which are not?

The rule we apply is that AI receives the processing work and the person keeps the judgement work. The difference between them is tested with a single question, whether the result can be verified by somebody who did not take part in producing it. A summary of a conversation gets verified by reading the transcript, so it can be delegated. A positioning decision cannot be verified against anything, because it chooses what you leave aside, so it stays with the person.

This split also explains why companies that start with content production at volume are the first to be disappointed. They automate the visible part, meaning the text, while the part that genuinely consumes time stays untouched, because it consists of rebuilding the context at every piece of work.

How do you measure whether it works?

Not through volume produced, which rises anyway and says nothing. The useful measurement is the time elapsed to a result a person accepts without rewriting it, tracked on the same type of work, before and after.

The second measurement, harder to accept, is the rate at which results get rejected at verification. A rejection rate that falls steadily shows the system is learning the company’s rules, while a flat rate shows people have started accepting whatever they receive, meaning the gate has stopped working.

The third, and the one that matters long term, is consistency. Two different people ask for the same piece of work on two different days, and the results get compared. If they differ substantially, the system has insufficient written context, however good the model behind it.

What is different in a market that started later?

The starting point is lower, and that is a risk and a window at the same time. Eurostat data for 2025 puts AI use among European companies with at least ten employees at an average of 20.0%, while the member states at the bottom of the range sit around 5%, meaning a quarter of the European level.

The same data shows where the effort goes wherever AI actually is used, because 34.7% of European companies using it apply it in marketing or sales, meaning exactly the area where it gets decided who is chosen. McKinsey measured the regional gap separately, showing that 12% of companies in Central Europe have implemented AI at organisational scale, against 28% in Western Europe, and estimates a regional potential of more than 700 billion euros.

The useful reading of these figures is not that some markets are behind, because that much is known. The useful reading is that the space for differentiating through thinking, rather than through volume, is still open in a market where almost nobody occupies it. A mid-sized company that writes down its knowledge now arrives, within two years, at a position its competitors can no longer reach with a subscription.

What does AI used as a processor look like in practice?

The best example I can give is our own system, because I measure it daily. All the knowledge about clients, delivery work and decisions is written in text files, versioned, and above them sit the procedures by which each type of work gets done. When a work session on a client begins, that client’s context loads automatically, so nobody explains it any more.

What passes through the system is research, client profiles, conversation analysis, drafts and verification. What does not pass through it is the decision and the relationship with the person, and that boundary is not an image precaution. It is the practical conclusion of the fact that a model produces good options and cannot carry the consequence of the choice. I described the layered structure and its figures in the article on the AI Brain, and how it connects to an actual marketing strategy in the article on strategy that uses AI.

The effect that matters to a founder shows up in consistency. Two people asking for the same work receive compatible results, a new deliverable starts from its written procedure, and the reason behind a decision stays written next to the decision. None of that comes from the model, it comes from what was written around it, and that is precisely why changing the model breaks nothing.

Where does a company that has done nothing so far start?

With three moves, in this order, and none of them requires a software budget.

The first move is the knowledge inventory. You look at what the company knows and where that information currently sits. At most companies of the size we are discussing, the answer is that it sits in three or four people’s heads, in inboxes and in files nobody can find. As long as that is the situation, no tool has anything to process.

The second move is choosing a single piece of work. Not the department, not processes in general, but one type of deliverable you produce often and that visibly consumes time. You write its procedure all the way through, with everything that needs to be known for it to come out well, then you test it by having somebody else follow it. The procedure is good when the second person produces a result the first one accepts.

The third move is the verification gate, calibrated on bad cases. You take a deliverable you know is weak and run it through the check. If it passes, the gate does not work, however convincing its report looks. Calibrating on good examples is the most frequent reason companies believe they have control when they do not.

Only after those three moves does it make sense to put a model on top, and from that point things move much faster than anyone expects, because the hard work had already been done.

Frequently asked questions

Where do I start if I have nothing written down? By gathering in one place, in text format, what the company knows about its clients. A model can only read what exists in writing, so this step cannot be skipped, however good the tool you buy.

How much of marketing can AI do? The processing part, meaning research, structuring, drafts and mechanical checks. The decision about what you say and to whom, meaning positioning, stays human work, because it requires choosing what to leave aside.

Do we risk sounding like everybody else? Yes, if you use the model without your own context, because then it writes the average of the internet. The risk disappears to the extent that the company writes down its voice, its examples and its rules, and the model receives them every time.

How do I know whether an AI project is worth continuing? By asking whether the time to a result a person accepts has fallen, not by the volume produced. A system that produces ten times more material somebody has to rewrite anyway has moved the work rather than reduced it.

Do we need a new department? Not at the size we are discussing. What is needed is one person who knows the operation and has allocated time to write the procedures, plus a clear decision about what the system is not allowed to touch.

If you want to see what AI would process in your company’s case, meaning what knowledge already exists and what is missing for it to be usable, you can request a positioning analysis. It shows you how you look from the outside now, and from there it becomes clear whether your problem calls for tools or calls for clarity first.