AI maturity model explained: the levels, the frameworks, and how it differs from readiness.
Gartner has five levels, MIT has four, Accenture has four groups and Cisco has four tiers — and all of them agree that the top level is small and the money is there. An AI maturity model is a ladder: it describes how deeply AI is embedded in how a company decides and operates. Readiness is whether you can start climbing. This page explains both, compares the major models with their actual level names, and shows the four-level model we score in the AI readiness assessment.
- Gartner · MIT CISR · Microsoft · Accenture · Cisco
- Readiness vs maturity
- Four-level model used in our assessment
Level names and statistics verified against the publishers’ own pages: October 2026.
of enterprises sit at MIT CISR’s top stage, “Become AI Future Ready”; 28% are still at stage one, “Experiment and Prepare”.
Profit margin above industry average for stage-four enterprises; stage one sits 9.6 points below (MIT CISR, 2022 data).
of organisations are “AI high performers” — attributing 5% or more of EBIT to AI — in McKinsey’s 2025 survey. 88% use AI somewhere.
The year Watts Humphrey began formalising the Capability Maturity Model at Carnegie Mellon’s SEI. Every AI maturity model is its descendant.
Readiness vs maturity
The two words get used interchangeably by vendors, which is how a company ends up with a “readiness score” that measures what it has already done and a “maturity assessment” that measures what it could do. The literature keeps them apart. Readiness, in the academic definition most studies adopt, is “the preparedness of organizations to implement change involving applications and technology related to AI” — the preconditions, assessed before an adoption decision, for a specific purpose. Maturity models, in the words of a 2021 systematic review, are “action-oriented standards that consist of discrete maturity levels”: staged descriptions of how reliably an organisation’s practices produce outcomes, assessed after adoption has begun.
Our working synthesis — ours, not a quoted canon: readiness asks “can we start this use case without it joining the 42% that get abandoned?”; maturity asks “how far along the ladder are we, across the whole organisation, and what does the next rung look like?”. A startup can be highly ready for one narrow use case and at maturity level one. An enterprise can be at maturity level three overall and unready for the next use case because the data it needs lives in a system nobody owns. The six dimensions are the same in both views; what changes is the question asked of each.
This is why the AI readiness assessment reports a maturity level per dimension rather than a single readiness score. The level tells you where you are on the ladder; the gap between the levels your use case needs and the levels you have is your readiness — and the roadmap is how to close it.
The major AI maturity models, with their real level names
Taken from the publishers’ own pages in October 2026. Where a widely quoted version differs from what the publisher currently says, we use the publisher.
| Model | Publisher, year | Levels or groups | What it scores |
|---|---|---|---|
| AI Maturity Assessment | Gartner, 2025–26 | Foundational · Emerging · Operational · Scaled · Transformational | Seven pillars: strategy, data, governance, engineering, operating model, culture, AI product and value |
| Enterprise AI Maturity Model | MIT CISR (Weill, Woerner, Sebastian), Dec 2024 | 1 Experiment and Prepare (28%) · 2 Build Pilots and Capabilities (34%) · 3 Develop AI Ways of Working (31%) · 4 Become AI Future Ready (7%) | How embedded AI is in decision-making, with measured financial performance per stage |
| Enterprise AI maturity in five steps | Microsoft Inside Track, Oct 2025 | Awareness and foundation · Active pilots and skill building · Operationalize and govern · Enterprise-wide adoption · Transform with agentic AI | Microsoft’s own IT journey, written as a guide for IT leaders |
| The Art of AI Maturity | Accenture, Jun 2022 | AI Experimenters (63%) · AI Innovators (13%) · AI Builders (12%) · AI Achievers (12%) | 0–100 index across foundational and differentiating capabilities; Achievers correlated with 50% higher revenue growth |
| AI Readiness Index | Cisco, 2025 (third edition) | Laggards (3%) · Followers (48%) · Chasers (36%) · Pacesetters (13%) | Six pillars: strategy, infrastructure, data, governance, talent, culture — a readiness index reported as tiers |
| Capability Maturity Model | SEI / Carnegie Mellon, 1986–1993 | Initial · Repeatable · Defined · Managed · Optimizing | Software process reliability — the template every later maturity model copied |
An older Gartner list (Awareness, Active, Operational, Systemic, Transformational) circulates widely; Gartner’s current public page uses the names above. BCG’s “AI Maturity Matrix” ranks economies, not companies, and is omitted.
The four levels we score, per dimension
Four rather than five, because for a company of 5 to 500 people the difference between Gartner’s “Scaled” and “Transformational” is theoretical. Each of the six dimensions — strategy, data, infrastructure, governance, skills, operations — gets its own level; the report shows them as a heatmap.
- 1
Exploring
Roughly MIT CISR stage 1 · Gartner Foundational
Individuals use AI tools; the organisation has no stated use case, no data owner, no policy, no budget line. In this dimension the next step is always cheap and always unglamorous: pick a use case, name an owner, write a page.
- 2
Foundations
Roughly MIT CISR stage 2 · Gartner Emerging
A use case is chosen and measured, data for it is located and owned, a policy exists, someone has been trained. Pilots are possible but fragile. Most SMBs we assess sit here in two or three dimensions and at Exploring in the rest — the heatmap is rarely flat.
- 3
Pilot-ready
Roughly MIT CISR stage 3 · Gartner Operational
The dimension can support a production pilot: data is accessible by API, infrastructure can deploy and monitor, governance reviews new use cases, people can operate and correct the system, operations has an owner for month two. This is the level the chosen use case needs in every dimension it depends on.
- 4
Scaling
Roughly MIT CISR stage 4 · Gartner Scaled
AI is embedded in how this part of the business works and the next use case is a reuse, not a project. MIT CISR puts 7% of enterprises at this stage overall; Cisco’s Pacesetters are 13%. For an SMB, reaching Scaling in one or two dimensions is a competitive position, not a vanity badge.
The evidence that maturity pays — and that most never get there
Correlations, not proofs; the better studies say so themselves. They point the same way.
MIT CISR: the top two stages outperform
Using 2022 survey data from 721 enterprises, stages one and two had growth and profit below industry average (stage 1: −12.6 and −9.6 points); stages three and four were “well above” (stage 4: +17.1 growth, +10.4 profit). The authors note the data mostly reflects analytical AI, not generative.
Accenture: Achievers grow faster
Among 1,176 AI-using firms, the 12% “AI Achievers” had maturity scores “correlating with 50% higher revenue growth than their peers”. Accenture forecast the share would reach 27% by 2024; no later confirmation was published.
Cisco: readiness has stalled at 13%
Pacesetters have made up 13–14% of organisations for three consecutive years. 97% of them say they deployed AI at the scale and speed needed to realise ROI, against 41% overall; only 19% of all organisations have fully centralised data infrastructure.
McKinsey: use is universal, impact is rare
88% of organisations use AI in at least one function, 39% report any enterprise-level EBIT impact — most under 5% — and about 6% qualify as high performers. “Meaningful enterprise-wide bottom-line impact from the use of AI continues to be rare.”
IBM: one in four initiatives met its ROI
Two thousand CEOs in 2025: 25% of AI initiatives delivered the expected return, 16% scaled enterprise-wide, and half admitted recent investments left them with “disconnected, piecemeal technology”. Infrastructure and operations are the dimensions that fail quietly.
Gartner and S&P: the abandonment rate
Gartner predicted at least 30% of generative AI projects would be abandoned after proof of concept by end-2025, and over 40% of agentic AI projects cancelled by end-2027; S&P found 42% of companies abandoned most of their AI initiatives in 2025, up from 17%. Maturity is partly the ability to stop early.
What moves a dimension one level, in practice
From the assessments we have run. Each step is smaller than the frameworks make it sound; the trick is doing it in the weakest dimension first.
Strategy: write down the use case you rejected.
Exploring to Foundations is one page: the use case, its number, its sponsor, and the use cases you decided against. The rejected list is what separates a strategy from a wish.
Data: name an owner before you clean anything.
Most data problems are ownership problems wearing a quality costume. A named owner, a completeness estimate and read access for one engineer move a dimension from Exploring to Pilot-ready faster than any platform purchase.
Infrastructure: one API, one deploy, one log.
If your CRM has an API, your site is server-rendered and you can deploy a small service with logging, you are Pilot-ready. If every integration needs the original vendor, you are Exploring regardless of your cloud bill.
Governance: policy, inventory, risk check.
Three artefacts, each a page. Foundations is the policy; Pilot-ready adds the inventory and a written risk check for new use cases; Scaling is when nobody has to be reminded. Our template covers the first.
Skills: one internal teacher beats one external course.
The dimension turns on whether someone inside uses the tools for real work and can show others — Deloitte’s respondents name skills the biggest integration barrier. A 45-minute induction with one hallucination and one injection example is the Foundations threshold.
Operations: decide the metric before launch.
The difference between Foundations and Pilot-ready in operations is a sentence written before go-live: “we will know this works when X, and Y owns it after launch”. IBM’s 25% ROI figure is mostly companies that skipped the sentence.
Six dimensions, four levels, one heatmap.
The free AI readiness assessment scores your maturity level per dimension from evidence, adds the website side, and returns a 90-day roadmap for the dimension that would stop your use case first.
Get the free AI readiness assessmentAI readiness checklist·AI use policy template·AI agents we build
Fair questions
- What is an AI maturity model?
- A staged description — usually four or five levels — of how deeply AI is embedded in an organisation’s decisions, processes and culture, from ad-hoc experiments to AI shaping the operating model. The form descends from the Software Engineering Institute’s Capability Maturity Model of the late 1980s. Gartner, MIT CISR, Microsoft, Accenture and others publish their own; the level names differ, the shape does not.
- What is an AI maturity assessment?
- The act of placing an organisation on one of those ladders, dimension by dimension, from evidence: documents, data samples, system access, interviews. Vendor self-assessment quizzes place you from your own ticks and flatter accordingly — Cisco measures 13% of organisations as fully ready while 58% report a well-defined strategy. Our assessment scores six dimensions on four levels and shows the spread as a heatmap.
- AI readiness vs AI maturity — which do we need?
- Readiness if you are deciding whether to start a specific use case; maturity if you want to know where the organisation stands overall and what the next rung is. In practice they are one exercise: the maturity level per dimension, compared with the level your use case needs, is your readiness gap. That is why our readiness assessment reports maturity levels.
- Which maturity model should a small business use?
- None of the enterprise ones literally — they assume data offices and multi-year programmes. Use their dimensions and their honesty about base rates (7% at the top stage, 6% high performers, 13% fully ready), and score yourself on a four-level scale per dimension. That is the model on this page, and it is deliberately boring: the value is in the per-dimension spread, not the label.
- What level is “good enough”?
- Pilot-ready in every dimension your chosen use case depends on. Scaling everywhere is neither necessary nor, for most SMBs, worth the cost. A company that is Pilot-ready in data, infrastructure and operations and at Foundations in governance can run a safe internal pilot; the same company should not launch a customer-facing agent until governance catches up.
- How long does it take to move up a level?
- Exploring to Foundations in a dimension is days to weeks — one page, one owner, one session. Foundations to Pilot-ready is weeks to a quarter, usually gated by data access or a missing integration. Pilot-ready to Scaling is the long one, measured in successful use cases rather than months. The assessment’s 90-day roadmap targets the first two transitions in the weakest dimension.
Sources: Gartner: AI Maturity Model toolkit · MIT CISR: Enterprise AI Maturity Model (Dec 2024) · MIT CISR press release (Dec 2024) · Microsoft Inside Track: enterprise AI maturity in five steps (Oct 2025) · Accenture: The Art of AI Maturity (Jun 2022) · Cisco AI Readiness Index 2025 · McKinsey: The state of AI in 2025 · IBM Institute for Business Value: 2025 CEO Study · Gartner: 30% of GenAI projects abandoned after PoC (Jul 2024) · Gartner: 40% of agentic AI projects cancelled by 2027 (Jun 2025) · S&P Global: AI abandonment rate (via CIO Dive, Mar 2025) · Deloitte: State of AI in the Enterprise 2026 · Sadiq et al.: AI maturity model, a systematic literature review (PeerJ CS, 2021) · Jöhnk, Weißert & Wyrtki: Ready or Not, AI Comes (BISE, 2021) · Capability Maturity Model — history