AI Maturity Model: 5 Stages and Where You Stand

Most companies rate their own position on an AI maturity model higher than the evidence supports. Ask a leadership team where they sit on the five-stage scale and the answer usually skews one stage ahead of what a neutral outsider would say.

That gap matters because the stage you're actually at determines what to fix next, not the stage you'd like to be at. Only around a third of organisations report having begun scaling AI across the enterprise at all, according to McKinsey, even with adoption now close to universal. Most of the gap between "using AI" and "AI maturity" is exactly this: nobody checked the stage before setting the target.

This article is for the CIO or data leader who needs an honest read on where the organisation actually sits, not the version of the answer that looks good in a board deck.

Key Takeaways

• Gartner's widely cited AI maturity model places organisations into five stages, running from early ad hoc experiments to AI reshaping the operating model.
• Most leadership teams place their own organisation one stage higher than an outside, evidence-based read would.
• Moving between stages depends on specific, checkable conditions, not on enthusiasm or headcount spent on AI projects.
• An AI maturity model tracks capability over years and multiple projects, which makes it a different tool from a one-time readiness assessment taken before a single investment decision.
• Skipping a stage is the most common way an AI programme stalls, since scaled AI depends on the governance and data foundations built at earlier stages.

What is an AI maturity model?

An AI maturity model is a staged framework that rates how deeply and effectively an organisation uses AI, based on evidence rather than intent. It sorts organisations into levels, from isolated experiments to AI embedded across the business, using signals like governance, data quality, and measured value rather than how many people mention AI in meetings.

That's different from an AI readiness assessment, which scores whether you're prepared to start a specific AI investment right now. A maturity model answers a longer-running question: how has your organisation's AI capability changed, and where does it sit relative to where it was a year ago. Most organisations need the assessment first, for the investment decision in front of them, and the maturity model second, once more than one AI initiative is running and worth comparing over time.

Several named frameworks exist, including MITRE's and CMU's own AI Adoption Maturity Model, but Gartner's five-stage version is the most widely referenced across industry and research.

This overall model folds governance in as one dimension among several. If governance is the specific piece you're evaluating, an AI governance maturity model narrows the same five stages down to just that one axis, and some organisations track technical depth separately as an AI capability maturity model rather than scoring the whole business at once. Both are useful for a narrower question. Neither replaces the organisation-wide read this piece is built around.

The five stages of AI maturity

According to Gartner's AI maturity model, organisations typically fall into one of five stages. The Gartner AI maturity model is the version most enterprises reference when comparing their own stage against a published standard. A note on the framework: stage names and boundaries vary slightly across published versions of Gartner's model; this piece uses Gartner's current published terminology.

Foundational

What it looks like: AI conversations happen informally, with no coordinated strategy or governance in place. Any AI use is ad hoc, driven by individual curiosity rather than a funded initiative.

The signal you're here: nobody in the organisation can name a specific AI project with a budget and an owner attached to it.

What moves you forward: a named use case with an executive sponsor, even a small one, moves an organisation out of Foundational faster than any amount of internal AI literacy training.

Emerging

What it looks like: early pilots exist, with growing executive interest but no consistent pattern for turning a pilot into a production system.

The signal you're here: at least one funded proof of concept is running, but nobody can say with confidence whether it will reach production.

What moves you forward: a governance owner assigned before the pilot ships, not after it stalls.

Operational

What it looks like: AI is embedded in at least one real workflow, with defined ownership and a measurable outcome attached to it.

The signal you're here: one AI system is running in production, generating a number someone tracks weekly.

What moves you forward: proving the pattern once, then deliberately repeating the same governance and data practices on a second use case instead of starting from scratch.

Scaled

What it looks like: AI capabilities are deployed across multiple functions, with measurable ROI tracked at the organisational level rather than per project.

The signal you're here: more than one business unit runs a production AI system, and leadership can point to the value each one generates.

What moves you forward: treating the data and governance foundation as shared infrastructure, not something rebuilt for each new use case.

Transformational

What it looks like: AI reshapes decision-making, operating models, and competitive position rather than sitting inside individual workflows.

The signal you're here: AI is inseparable from how the business actually runs, including newer patterns like autonomous agents making decisions within defined limits.

What moves you forward: at this stage the work shifts from building capability to governing it responsibly at scale, since the risk profile of an AI-native operating model is different from a company running a handful of production models. Getting there is really the practical form of AI readiness for enterprises, applied one stage at a time rather than all at once.

How to tell which stage you're actually in

The honest version of this exercise starts with data, not with strategy documents. If your organisation can't say clearly whether its production data is clean enough for a second AI use case, that's usually a sharper signal than any statement about ambition. Data readiness for AI is worth checking directly rather than inferring from how confident the room sounds.

When Classic Informatics built a full suite of clinical and analytics systems for Chris O'Brien Lifehouse, a specialist cancer centre with no unified systems across nine departments, the honest starting stage was closer to Foundational than anyone on the client side had assumed going in. The systems that got built afterward moved the organisation through Emerging and into Operational within a defined programme, not by accident.

Three questions place most organisations accurately: is there a production AI system with a named owner, does a second use case reuse the same governance pattern as the first, and can leadership name the metric that AI moved last quarter. Two "no" answers put you at Foundational or Emerging regardless of how many pilots are technically running.

Why most organisations overestimate their stage

Three patterns explain almost every case of an inflated self-rating.

1. Counting Pilots as Production

What it looks like: leadership cites three or four AI pilots as evidence of Operational-stage maturity, when none has reached a stable production state with a named owner.

Why it happens: a pilot generates visible activity and internal excitement, and activity gets mistaken for capability.

How to fix it: only count a use case once it has survived a full quarter in production with someone accountable for its output. If nobody internally can call that judgment without bias, external AI readiness services can validate the stage before it goes into a board deck.

2. Skipping Straight to Scaled Language

What it looks like: an organisation with one working AI system describes itself as "scaling AI," borrowing Scaled-stage language for an Operational-stage reality.

Why it happens: the vocabulary of ambition is easier to adopt than the governance work the next stage actually requires.

How to fix it: describe the current stage using this piece's stage-specific signals, not the stage you're aiming for next.

3. Treating One Success as the Whole Organisation's Maturity

What it looks like: one business unit's AI success gets reported as the company's overall maturity level, while most of the organisation hasn't started.

Why it happens: the success story is the easiest one to tell, and it travels further than the quieter truth about the rest of the business.

How to fix it: rate maturity at the business-unit level first, then report the honest average, not the best example.

In our experience, the gap between a leadership team's self-rating and an outside assessment is rarely one stage. It's more often two, and the second stage is usually the one where the actual governance work still needs doing.

Let's Sum Up!

A maturity model is only useful if the placement is honest. Most organisations sit a stage lower than they'd like to claim, and that's not a failure. It's the starting point for knowing what to fix before the next stage becomes possible.

Most organisations don't need a more ambitious roadmap. They need an honest placement on the curve they're already on. Classic Informatics has helped enterprise teams get that placement right, then closed the specific gap it revealed. Reach out if your own placement could use a second set of eyes.