How Mature Is Your Company at AI, Really?

A company's AI maturity is not read from what the committee says. It is read from the work its operation could no longer do without it. Three questions with verifiable answers place it, with no outside maturity model needed.

Carlos Andrés Ramírez · 2026-10-12

Start with how the question is answered today. In a committee meeting someone says the company is at an intermediate stage: there are licences, a couple of pilots and a named initiative. Nobody disputes it, because nobody has anything to dispute it with. It is an opinion about the organisation, given by the people with the most interest in it sounding good.

The short answer is that the real stage is located by dependence, not by declaration. A company is where the work already done with AI is, with an owner and a measurement. Everything else is intent, which is a respectable stage, but a different one.

The underlying problem

Why do AI maturity models tell you what you want to hear?

Most maturity models in circulation are published by someone who sells the next stage. They describe steps with attractive names, and whoever fills them in picks the most generous reading available, because every answer can be interpreted. "We have use cases in production" is true if there is one and also if there are twenty.

Picture a twenty-statement questionnaire the committee answers in an hour. It produces a score that looks like a diagnosis, and nobody has had to open a single process to get it. A serious diagnosis does the opposite: it asks things you can check by watching the operation on any ordinary Tuesday.

The framework

Three tests separate real maturity from declared maturity

The three tests look at different things: what work depends on AI, who answers for it, and what evidence shows it works. If all three hold for a process, that process is mature. The company's stage is the sum of the processes that pass.

1. Dependence test: what work would stop if we switched the AI off?: If switching the tools off for a week changes nothing in lead times, costs or queues, AI is being used but the operation does not depend on it. That is an exploration stage, useful and still reversible. The work that would slow down, get dearer or fall back to people carrying a load they no longer have is the work that counts.
2. Ownership test: who answers, by name, for each use?: Every process with AI needs one person who decides what the agent does, what gets reviewed and what happens when it is wrong. If the answer is "the technology team" or "the committee", there is no ownership. A process with no owner depends on nothing going wrong.
3. Evidence test: what was measured before, and what is measured now?: Maturity is being able to show with your own numbers that the process improved: cycle time, errors, cost per case. Without a prior measurement there is nothing to compare, and without a comparison the success story is an anecdote. This is the test fewest companies pass.

The mechanism

A company's real stage is usually below its average and above its worst function

Imagine a company with a customer service team where an agent resolves queries, with a named owner, a resolution metric and a comparison with last year. That process passes all three tests. In finance, an assistant is used by some people to draft emails, with no owner and no measurement. In procurement, a pilot has been stalled for six months.

Averaged, the company declares an intermediate stage. Looked at process by process, it has one mature process, one in exploration and one stalled. Those are three situations calling for three different decisions: scale the first, tidy up the second and decide whether the third deserves to continue. The average hides exactly the difference the committee needs to see before allocating budget.

A company is not at the stage it says. It is at the stage of the processes that depend on AI, have an owner and can be measured.
BECOME

The consequence

Committing budget without this reading funds the wrong stage

The decision a diagnosis usually supports is moving to the next phase: more licences, more teams, more use cases. If the company placed itself by declaration, that investment lands in an organisation that still cannot supervise what it already has. Ownerless processes multiply and measurement stays absent, so spending grows faster than the ability to know whether it pays off.

There is a second consequence, less visible. When leadership finds the real stage was lower than the declared one, it tends to read that as AI failing. It was a diagnostic error, and it is fixed with a week of work on each process, not with a change of strategy.

List every process where AI is used today, however lightly. For each one, answer the three tests in writing: what would stop without it, who answers by name, and what measurement existed before. Any process that cannot answer all three is not mature, however widely it is used.

Frequently asked questions

How do I know what stage of AI maturity my company is at?

List the processes where AI is used and check three things in each: whether the work would depend on it if it were gone, whether a named person answers for it, and whether a before-and-after measurement exists. The company's stage is the set of processes that meet all three. What does not is exploration, even if it has licences and users.

Are the AI maturity models published by consultancies and vendors useful?

They are useful as a shared vocabulary, not as a diagnosis. They tend to measure what is declared in a questionnaire rather than what can be checked in the operation, and whoever fills them in leans toward the most favourable answer. Use them to structure the conversation and verify each answer against a specific process.

Can a company be at different stages in different functions?

Yes, and that is the norm. One function may have a process with an owner and a measurement while another uses AI informally and a third has a stalled pilot. Averaging them hides which decision belongs to each, so place the stage process by process and decide separately where to scale, where to tidy up and where to stop.

What if the diagnosis comes out lower than leadership expected?

Present it as a process diagnosis, not a verdict on strategy. A low result shows where an owner or a measurement is missing, which are concrete tasks of a few weeks. It is better to find that out before committing budget to the next phase than after spending it.

Let's place your company's real stage together

From the idea to the operation

Turning this thesis into something operable starts by deciding where the value sits in your company and what must change to capture it.

About the author

Carlos Andrés Ramírez — Transformation Director

Specialist in business transformation and reinvention. Director of Specialised Programmes and lecturer in Artificial Intelligence at UPC's Graduate School.