The AI Maturity Model in Four Stages, With One Question for Each

Ask a room of business owners whether their company is using AI, and every hand goes up. Ask whether AI has changed a number they report to their board, and most of the hands come down.

The gap between those two answers is a question of stage. This post lays out the four-stage AI maturity model we use with the companies we work with, and a single question that tells you which stage you are in.

Who are the existing AI maturity models built for?

The well-known AI maturity models are built for large enterprises. MIT, Microsoft, MITRE and Gartner each publish a serious one, and each assumes a CIO, a transformation budget and time for a formal assessment. If you run a company of 10 to 200 people, none of them was written with you in mind.

That is not a knock on them. Each is careful work for the audience it serves:

• MIT: four stages drawn from a survey of 721 companies, where firms in the first two stages performed below their industry average and firms in the last two above it.^[1]
• Microsoft: an agentic adoption model that grades five capability areas across five levels.^[2]
• MITRE: a maturity model paired with an organizational assessment tool.^[3]
• Gartner: a maturity toolkit with a roadmap attached.

Those are good instruments for their audience. The operator running a real business on a lean team needs something different: a model you can sort yourself into after one read, named for what you would actually see in the office on a Tuesday.

What are the four stages of AI maturity?

The four stages are Stalled, Scattered, Structured and Strategic. Stalled means AI has no owner. Scattered means real value in pockets with no map. Structured means AI runs in real workflows, measured against a baseline. Strategic means AI shapes what you sell. Each stage has one question that sorts you into it.

The model is ours, drawn from client engagements rather than from a survey. The 2026 research we cite below does not produce the stages. It describes, in measured terms, the symptoms we see at each one.

What does the Stalled stage look like?

A stalled company has no owner for AI, no agenda and no roadmap. The tell is simple: AI is nobody's actual job. Individual people may use a chat assistant on their own, but nothing the company does has changed, and nobody has been asked to change it.

Stalled companies are rarely against AI. They are busy, or waiting for the dust to settle, or they tried one pilot that went nowhere and did not try again.

Tools are easy to catch up on later. What a stalled company loses is months of learning how AI behaves inside your operation, with your data and your people.

Why is Scattered the most dangerous stage?

Scattered is dangerous because it feels like progress. Real value exists in pockets, such as one person's prompt workflow that saves a day a week or a vibe-coded internal tool, so the company believes it has adopted AI. What it lacks is a map: nobody can list where the value is.

Scattered is also the default state of AI adoption across the whole economy. Google's ATLAS study of almost 15 million Gemini interactions found AI in use across more than 68 percent of occupations, yet it touched only 21 percent of the tasks in the median occupation that used it.^[4]

That is broad but shallow, which is Scattered measured across the economy.

We made the longer version of that argument in why AI reaches most jobs but only a fifth of the work. The figures come from one vendor's usage data, so treat the direction as solid and the exact percentages as Gemini's.

Confidence runs ahead of the evidence

The feeling of progress is itself measurable. CloudBees surveyed more than 200 enterprise technology leaders in 2026: 92 percent were confident their AI-generated code was production-ready, while 81 percent reported more production issues tied to it.^[5] The same companies could attribute only a third of their AI spend to specific business outcomes.

CloudBees sells software delivery tooling, so read it as a vendor's survey. The pattern still matches what we see: a Scattered company usually believes it is Structured. Google's DORA research team described the same stage in plain words.

What stops companies moving from Scattered to Structured?

Ownership stops them, not technology. Moving to Structured takes a named owner, an agreed definition of success and a number from before AI to compare against. None of those is a technical problem. They are decisions somebody in the company has to be given the authority to make.

Survey data points the same way. Aggregated 2026 figures from Forrester, Anaconda and others put the share of agent pilots that never reach production at 88 percent.^[7] Among shipped deployments with negative returns after a year, the reported causes were unclear success criteria (41 percent), missing tool or data access (33 percent) and drifting evaluation (26 percent).

The sample sizes behind those aggregates are not disclosed, so hold the exact figures loosely.

Deciding what success means and granting access are both decisions an owner makes.

One of the largest measured looks at company AI use reaches the same place. Researchers studied 1,764 organizations and more than 17 million ChatGPT Enterprise messages.^[8] Three of the five authors work at OpenAI and the other two were paid contractors for it, which makes the conclusion more telling, since it runs against selling seats.

ResearchResearcher

How Organizations Use AI: Evidence from ChatGPT

Aaron Chatterji et al. · 12 Aug 2026

Even with these limitations, the patterns documented in this paper point to a central feature of enterprise AI diffusion: adoption is only the beginning of deployment. The rapid adoption of generative AI by firms should therefore not be equated with immediate productivity transformation.

Source: arxiv.org. Reproduced verbatim; punctuation is the source's. Photo: Cornell Watson, CC BY-SA 4.0, via Wikimedia Commons.

What does the Structured stage look like?

A structured company runs AI inside real workflows, on shared tooling, and measures the results against a baseline. The baseline is the defining act. You wrote down the number from before (hours, turnaround, error rate, cost), and you compare against it. Without that, you have Scattered with nicer tools.

DORA's 2026 report on the return from AI puts the question bluntly: "Do you have a baseline?"^[6] Its worked example for a 500-person engineering organization models a 39 percent first-year return. It also shows why an average like that can mislead you.

• Gains run 35 to 40 percent on simple new work and 10 percent or less on complex older systems.
• The sample calculator raises the rate of failed releases from 5 to 6 percent after AI adoption.
• DORA frames all of it as a high-uncertainty estimate meant to start a conversation.

You only know which side of those splits you are on if you measured before you started. That is why the Structured question is about a number, not a tool.

What separates Structured from Strategic?

Structured means AI runs your operation better. Strategic means AI changes what you sell. The test is subtraction: if AI disappeared tomorrow, would your offer change? A structured company would lose efficiency. A strategic company would lose part of its product, because the thing customers buy now has AI built into it.

At the Strategic stage, capability that started inside the operation becomes something customers pay for. The service ships with agents in it, a task that used to be a cost becomes part of the offer, or pricing rests on outcomes the AI makes possible.

Not every business needs to get here. For plenty of companies, Structured is the right place to stay, and the model does not grade that as a failure.

Skipping a stage is the classic failure

The stages run in order. The company that jumps from Stalled straight to "let's launch an AI product" has never run AI in its own operation. That is how pilots end up in the 88 percent.

Your own operation is the cheapest place to learn how agents behave before customers depend on them, which is why the first thing to automate is usually inside the building.

How do you move up a stage?

You move up one stage at a time, and each move is a single decision. Stalled to Scattered takes permission. Scattered to Structured takes ownership. Structured to Strategic takes productizing: building what already runs reliably inside the business into what you sell. None of them requires a new platform first.

1. Stalled to Scattered: give one person permission, one workflow and one before-number to beat.
2. Scattered to Structured: name an owner, pick the baseline and put everyone on shared tooling.
3. Structured to Strategic: take the workflow that already runs reliably and put it into the offer.

The later stages run on humans and agents working the same workflows, with the judgment kept where your customers are paying for it. If you want to see where we would start inside an operation, the automation use cases are the short version.

So which stage are you in, going by the four questions rather than the last demo? That honest answer is worth more than a roadmap toolkit, because the next move from every stage is one decision, and it is yours to make.