AI Maturity Model: What the Levels Mean for Smaller Firms

An AI maturity model is a ladder of stages, usually five, that describes how far an organization has come in using AI, from first experiments at the bottom to AI running through most of its decisions at the top. It's a useful map of direction and a common language for comparing companies. It's a poor guide to what a 10-to-200-person business should build next, because most models were written for large organizations or inherited their design from ones that were, and the evidence that climbing the levels causes better results is thinner than the charts suggest.

In September 2026 we read the 18 pages ranking for "ai maturity model" and close variants. Of those, 16 described the same one-way climb from "aware" or "experimenting" to "transformational." Only 3 gave criteria for each level that you could actually check against your business. And only 1 of the 18 mentioned a company with fewer than 200 employees by headcount, which is odd, since that's the size band where AI use is lowest and the advice is thinnest.

This page explains what the levels mean, compares the best-known models, looks at where the ladder shape came from and what the research says about it, and then shows how to use a maturity model at your size without letting it set your priorities. It's the level-based companion to our guide to AI readiness assessment services, which covers the other half of the question: whether a specific change is ready to make now.

Key Takeaways

• An AI maturity model ranks an organization on a ladder of stages, usually five, from experimenting with AI to running on it.
• Of 18 ranking pages we reviewed in September 2026, only 1 mentioned a company under 200 employees, and only 3 gave level criteria you could check.
• The ladder comes from software engineering. Its 1993 source, the Capability Maturity Model, said it described "large, government contracting organizations."
• MIT CISR found firms in its top two AI stages above industry average on growth and profit, but it assigned stages by how effective AI already was.
• A study of 12 UK firms with 10 to 99 employees found "scant support" for a stages model. They adopted technology where it supported the business.
• At 10 to 200 people, measure maturity by what you've shipped and measured, not by a level you rate yourself at.

Table of Contents

• What is an AI maturity model?
• The best-known AI maturity models compared
• Where the levels came from
• Does moving up the levels pay off?
• Why maturity models misfire for smaller businesses
• How to use an AI maturity model at 10 to 200 people
• A maturity scale you can count
• Where Avolis fits
• Frequently Asked Questions
• Continue Learning

What Is an AI Maturity Model?

An AI maturity model is a framework that sorts organizations into ordered levels according to how widely and how well they use AI. Each level comes with a description of what a company at that stage typically has in place (a strategy, data systems, skilled people, rules for safe use), and the model implies that you move up by adding what the next level describes. Most use five levels, a few use three, four, or six, and nearly all of them score the same handful of areas: strategy, data, technology, people, and governance.

Stripped of each vendor's labels, the typical five levels look like this:

People often use maturity and readiness as if they mean the same thing, and they don't. Maturity describes how advanced a business is with AI overall. Readiness describes whether one specific change can be made now, which is why a company can sit at the bottom of every maturity scale and still be ready to change one workflow tomorrow. The pillar guide covers that distinction in more depth. This page stays with the levels.

The Best-Known AI Maturity Models Compared

The most widely cited AI maturity models come from an analyst firm, three cloud vendors, a federally funded research center, a business school, and a large consultancy, and they disagree on the number of levels while agreeing on the direction. The table below uses each publisher's own level names wherever we could read them on its own pages.

Gartner's is the name most people search for. Its research on the model is published to clients, so the level names above are as reproduced by vendors that cite it, such as BMC (BMC, AI Maturity Models). MITRE's model is the most detailed of the free ones, with five levels described as "hierarchical and scalable progress throughout AI adoption" (MITRE, AI Maturity Model). Microsoft's newest version, written for AI agents rather than AI in general, gives the most observable criteria of any we read, such as whether there's an executive sponsor and separate environments for testing and production (Microsoft, Agentic AI adoption maturity model).

Look at the last column and a pattern shows up. Accenture's groups come from "1,176 of the world's largest companies" (Accenture, The Art of AI Maturity, 2022). MITRE's pillars assume an organization large enough to have separate strategy, data, and technology functions. The cloud vendors' models describe the path to scaling on their own platforms, which is a reasonable thing for a cloud vendor to describe. None of this makes the models wrong. It means each one describes the path its authors studied or sell into, and where a model reports who it studied, those were large organizations.

Where the Levels Came From

The five-level ladder in most AI maturity models wasn't discovered by studying AI adoption. It was inherited from software engineering, where the Capability Maturity Model set out five levels in the late 1980s and early 1990s, and it came with limits that its own authors stated plainly.

The Capability Maturity Model, or CMM, was developed at Carnegie Mellon's Software Engineering Institute, which the US Department of Defense funds, originally as a way to assess the software process of contractors. Its 1993 version named the levels Initial, Repeatable, Defined, Managed, and Optimizing, and described the bottom level as "ad hoc, and occasionally even chaotic" (Paulk, Curtis, Chrissis, and Weber, Capability Maturity Model for Software, Version 1.1, Software Engineering Institute, 1993, p. 8). Put the CMM's names next to MITRE's and Microsoft's in the table above and the family resemblance is obvious.

Three passages from that report matter for anyone applying its descendants to a small business:

• It was built for big organizations. The authors wrote that the model "is expressed in terms of the normative practices of large, government contracting organizations," and named a version "for small projects and/or small organizations" as future work (p. 52).
• The top levels were barely observed. "Maturity Levels 4 and 5 are relatively unknown territory for the software industry," the report says, with only a few examples of either (p. 16).
• It insisted on order. "Skipping levels is counterproductive because each level forms a necessary foundation from which to achieve the next level" (p. 25).

That last claim, that organizations move through fixed stages in a fixed order, is older still, and it has been tested. In 1973 and 1979 Richard Nolan proposed that a company's use of computing passes through set stages. In 1984, two reviews in the same issue of Communications of the ACM went through the evidence. John King and Kenneth Kraemer found that the main claims of empirical support "come from Nolan himself." Nobody else had corroborated them.

In their own study of 56 cities in 10 countries, the organizations using the most "mature" computing-management policies were also "suffering from the highest levels of problems with computing" (King and Kraemer, "Evolution and Organizational Information Systems", Communications of the ACM, 1984, p. 7). Izak Benbasat and three co-authors reviewed seven empirical studies and concluded that together they cast "considerable doubt on the validity of the stage hypothesis" (Benbasat, Dexter, Drury, and Goldstein, "A Critique of the Stage Hypothesis", Communications of the ACM, 1984).

None of that stopped maturity models from multiplying. By 2005, researchers counted "more than 150 maturity models" across IT management alone (de Bruin, Rosemann, Freeze, and Kulkarni, "Understanding the Main Phases of Developing a Maturity Assessment Model", ACIS, 2005, p. 3). A widely cited 2011 paper summed up the standing criticism: maturity models have been called "step-by-step recipes" that oversimplify reality and lack empirical foundation, and they "tend to neglect the potential existence of multiple equally advantageous paths" (Pöppelbuß and Röglinger, "What Makes a Useful Maturity Model?", ECIS, 2011, p. 4).

Our reading of the history: the ladder is a design choice, not a finding. It was drawn for large software contractors, its top rungs were described by analogy before anyone had seen many companies on them, and the idea that every organization climbs the same rungs in the same order failed its first serious tests. AI maturity models carry all three traits forward. That doesn't make them useless, but it does mean the ladder's shape tells you how the model was built, not how your business has to grow.

Does Moving Up the Levels Pay Off?

The best-known evidence shows that companies at higher AI maturity stages perform better financially, but it's correlational, the samples that report firm size are large companies, and in the most-cited case the stages were assigned partly by how well AI was already working. Better-controlled research finds much of the advantage belongs to firms that were already stronger before they adopted AI.

The headline study is MIT's Center for Information Systems Research. Using a 2022 survey of 721 companies, its researchers found that "enterprises in the first two stages of AI maturity had financial performance below industry average, while enterprises in stages 3 and 4 had financial performance well above industry average" (Weill, Woerner, and Sebastian, "Building Enterprise AI Maturity", MIT CISR Research Briefing, December 2024). Most firms sat in the lower stages: 28% in Stage 1, 34% in Stage 2, 31% in Stage 3, and 7% in Stage 4.

The footnote matters more than the chart. The researchers didn't place firms in stages by what they had built. They grouped them by "Total AI Effectiveness," a combined measure of how well AI was improving operations, customer experience, and the firm's wider ecosystem, with Stage 4 reserved for firms scoring 100%. In other words, a company lands in the top stage partly because AI is already paying off for it, so it's not surprising that top-stage companies are doing well. The study is careful and honest about its method. It's just measuring something closer to "AI is working here" than "this company has climbed a ladder."

Accenture's version has the same shape. Its top group, the AI Achievers, made up 12% of the large firms it studied. Their maturity score correlated "with 50% higher revenue growth than their peers." Deloitte's fourth State of AI in the Enterprise survey, of 2,875 executives in 11 countries, adds a useful twist: 17% of respondents were "Underachievers," reporting "a high number of deployments" but low outcomes (Deloitte AI Institute, State of AI in the Enterprise, 4th edition, 2021). Lots of AI activity, in other words, isn't the same as AI paying off.

The sharpest test of the "more AI, better results" idea comes from Statistics Canada, which linked firms' survey answers about AI to their administrative records. A simple comparison showed AI adopters with 16.8% higher labor productivity than non-adopters. Controlling for how productive each firm was before it adopted AI cut that to 10.2%, and adding controls for complementary capabilities (the other things strong firms tend to have, such as research and development, cloud computing, data analytics, and technology training for staff) cut it to 5.1%, which was no longer statistically significant. "Overall, there is no statistically significant direct association between AI adoption and productivity," the agency concluded (Statistics Canada, "Artificial intelligence adoption and productivity in Canadian firms", Economic and Social Reports, April 2026).

Our reading is that none of this says AI doesn't pay. It suggests the payoff sits in specific uses that fit a specific business, and the firms that find those uses were often well run to begin with. That's a different message from "get to Level 3," and for a small operation it's the more useful one.

Citation-ready summary: MIT CISR's 2024 Enterprise AI Maturity Model found firms in its top two stages above industry average on growth and profit, but it assigned stages by how effective AI already was, so the link is partly built in. Statistics Canada found an AI productivity premium of 16.8% that fell to an insignificant 5.1% once prior productivity and complementary capabilities were controlled for.

Why Maturity Models Misfire for Smaller Businesses

Maturity models misfire at 10 to 200 people for three reasons: they assume resources a small firm doesn't have, they reward sophistication rather than fit, and small firms rarely adopt technology in the neat order a ladder describes. The gap in adoption by size is real, but it isn't the kind of gap a ladder fixes.

Start with the numbers. In the European Union's 2025 survey of enterprise technology use, 17.0% of small firms (10 to 49 employees) used at least one AI technology, against 30.4% of medium firms and 55.0% of large ones with 250 or more (Eurostat, Artificial intelligence by size class of enterprise, dataset isoc_eb_ai, updated June 2026). Small firms nearly tripled their rate from 6.1% in 2021, so the gap isn't about small firms standing still. The spread by industry matters too: in 2025, 24.8% of enterprises in real estate used AI, and 10.8% in construction.

What holds small firms back isn't a missing level. Among EU firms that had considered AI but weren't using it in 2025, 70.9% cited a lack of relevant expertise, while only 20.7% said AI wasn't useful to them (Eurostat, Use of artificial intelligence in enterprises). In Canada, 34.3% of businesses with 20 to 99 employees said AI isn't relevant to their business, against 21.3% of those with 100 or more (Statistics Canada, Canadian Survey on Business Conditions, second quarter 2026). A maturity model can't tell those two groups apart. It will place both at the bottom, and it will prescribe the same next steps to a firm that doesn't know how to start and a firm that has looked carefully and found nothing worth doing yet.

The second problem is what the upper levels reward. A 2026 paper on AI maturity in small and medium-sized firms puts it directly: "maturity models that equate progress with scale and sophistication risk penalizing SMEs despite effective, context-appropriate AI use" (Sawang and Sornlertlamvanich, "Artificial Intelligence (AI) Maturity in Small and Medium-Sized Enterprises", arXiv preprint, February 2026). Even the assessment process can be out of scale. AI Sweden's national maturity assessment, a well-run program, takes about 80 hours of participation, "with two individuals attending meetings and around 60 people contributing to the evaluation" (AI Sweden, AI Maturity Assessment). A 45-person company doesn't have 60 people to contribute.

The third problem is order. When Tamara Levy and Philip Powell studied 12 established UK firms with 10 to 99 employees as they adopted the internet, the stage idea didn't hold up. They found "scant support from these cases for a stages model of development." The firms treated the internet "as they would other technology investments: if it supports the business then the investment will be made," and the authors concluded that "SMEs are unlikely to follow a stages model" (Levy and Powell, "Exploring SME Internet Adoption: Towards a Contingent Model", Electronic Markets, 2003, pp. 180–181). That was a different technology, two decades ago, so treat it as a caution rather than proof. Still, it describes something a ladder can't: small firms adopt a piece at a time, where it supports the business, and skip the rungs that don't apply.

Picture a 45-person fire-protection contractor that inspects, tests, and repairs sprinkler and alarm systems for commercial buildings. (It's a composite we've put together for illustration, not a client.) Its eight field technicians file around 60 inspection reports a week from a field-service app, and each report lists the deficiencies found, from a corroded sprinkler head to a failed tamper switch. An office coordinator turns every deficiency into a repair quote by hand, reading the report, looking up parts and labour, and typing the quote into the accounting system. The quotes run about a week behind, and some deficiencies never get quoted at all.

Run this company through any model in the table and it lands at Level 1. There's no AI strategy, no data team, no governance policy, and the only AI in use is the office manager drafting customer emails in a chat tool. The model's next steps follow from that: write a strategy, form a steering group, build data foundations, run pilots. Every one of those is reasonable for a 5,000-person enterprise. For this company, they'd take months and still not touch the deficiency quotes, which are high-volume, repetitive, already captured in one system, and tied directly to revenue the business is currently leaving on the table.

How to Use an AI Maturity Model at 10 to 200 People

Use an AI maturity model as a description of where you are and what larger organizations tend to add over time, not as a to-do list. It's worth consulting when someone outside the business asks for it, when you want to track change over several years, or when you need a shared vocabulary with a partner. It's the wrong tool for choosing your first or next AI project.

Even the vendors who publish these models leave room for this. AWS's generative AI maturity model says that "no single level is inherently superior or inferior," and that "the appropriate maturity level is contextual to the organization's goals and readiness" (AWS Prescriptive Guidance, Generative AI maturity model). That's the right way to read all of them.

A few practical rules help:

• Place yourself with evidence, not a feeling. For each area, write down what you could show someone: the policy, the system, the named owner, the number. If a level can only be claimed by saying so, treat it as unclaimed. Our page on readiness assessment methodology sets out the evidence levels in detail.
• Rate the workflow, not just the company. A business can be at Level 3 for dispatch and Level 1 for estimating, and that split is more useful than any average (our methodology page explains why).
• Skip what doesn't apply. A steering committee makes sense when many teams run AI projects at once. At 45 people, the owner and the office lead talking every Friday is the steering committee.
• Read the top level as the model's authors' goal, not yours. A plumbing company doesn't need AI to reshape its business model. It needs its dispatch board to stop losing jobs.

The pillar guide covers what else changes at this size, from governance to data.

There's one case where a scored maturity model is exactly the right tool: when a lender, an investor, a board, or a large customer wants to see how you compare with peers. A benchmark answers that question and nothing else can. Our page on AI readiness assessment consulting firms covers which kinds of firm sell a benchmark and what to ask for.

A Maturity Scale You Can Count

If you want a way to track AI maturity at 10 to 200 people, count what you've shipped and measured instead of rating yourself on a ladder. Each step below needs something you could hand to an outsider, a number or a name, so two people looking at the same business will place it in the same spot.

Why count instead of rate: yes, it's still a ladder. The difference is that every step on this scale describes something that already happened, so you can't reach it by writing a document. That's the difference from most ladders, where a strategy or a policy can move you up a level before any work changes. It also removes the pressure to skip ahead. A business at step 1 with one workflow that saves the office lead, say, six hours a week is further along, in any sense that matters to the owner, than a business with a strategy deck and no workflow in use.

Back at the fire-protection contractor, the path is short. Automating the deficiency quotes, with the coordinator reviewing each draft before it goes out, moves the company to step 1. Measuring how long quotes take and how many deficiencies get quoted, before and after, moves it to step 2. On a Gartner-style scale, it's still at Level 1 or 2. By its own numbers, it has more working AI than a company that rates itself higher on the strength of a strategy document.

If you'd like to take stock of the AI tools already in use before you start counting, our guide to AI readiness assessments for SMBs includes a keep, standardize, or stop inventory built for that.

From our own diagnostics: owners often place their business near the bottom of a maturity scale and assume that means they aren't ready to start. When we work through the workflows with the people who run them, there's often at least one that's ready now, and the reason it hasn't been touched is that nobody had measured it. This is our own client work, not independent research.

Where Avolis Fits

Our two-week diagnostic is how we find which workflow should become your step 1. We work through each workflow in scope with the people who run it, measure volumes, time per task, and handoffs, map where the work stalls, and rank what's worth building by impact and effort. Every engagement with us starts with that diagnostic, and the ranked result is yours whatever you decide to do next.

If you'd rather see how the scoring works first, our page on readiness assessment methodology explains the rules behind it. If you're weighing who to work with, start with which AI consulting company to choose. And if an earlier AI effort stalled after a maturity assessment, our analysis of why AI projects fail covers the most common reasons.

Frequently Asked Questions

What is an AI maturity model?

An AI maturity model is a framework that sorts organizations into ordered levels, usually five, by how widely and how well they use AI. Each level describes what a company at that stage typically has in place, such as a strategy, data systems, skills, and governance. Most models were built from studies of large enterprises, so they describe a path to scale.

What are the five levels of AI maturity?

Most five-level models run from aware or initial, through experimenting, operational, and systemic, to transformational or optimized. Gartner's model uses Awareness, Active, Operational, Systemic, and Transformational. MITRE's uses Initial, Adopted, Defined, Managed, and Optimized. The shape comes from software engineering's Capability Maturity Model, whose 1993 version set the five names.

What is Gartner's AI maturity model?

Gartner's AI maturity model is a five-level scale, as reproduced by vendors that cite it: Awareness, Active, Operational, Systemic, and Transformational. It describes a progression from talking about AI, through pilots and production use, to AI shaping the business model. Gartner's own materials are for clients, and it doesn't publish the sample the model was built from.

Does a higher AI maturity level improve financial performance?

Firms at higher stages tend to perform better, but the evidence doesn't show that climbing causes it. MIT CISR found its top two stages above industry average on growth and profit, yet it assigned stages by how well AI already worked. Statistics Canada found a 16.8% productivity premium for AI adopters that fell to an insignificant 5.1% after controls.

Is an AI maturity model useful for a small business?

It's useful as a description and as a benchmark when a lender, board, or large customer asks for one. It's a weak guide to what to build next, because most models assume resources small firms lack and reward sophistication over fit. A study of 12 UK firms with 10 to 99 employees found "scant support" for adopting technology in stages.

Continue Learning

A maturity model tells you where large organizations have tended to go. At 10 to 200 people, the more useful question is what's running, who owns it, and what it has changed.

Before you accept a maturity level, ask what it counted.

The wider assessment:

• AI readiness assessment services
• AI readiness assessment methodology
• AI readiness assessment for SMBs
• AI readiness assessment checklist and PDF
• Should we implement AI?

Choosing who does it:

• AI readiness assessment consulting firms
• Which AI consulting company should I choose?
• Why do AI projects fail?

---

Sources

All sources retrieved 2026-09-24.

• Mark Paulk, Bill Curtis, Mary Beth Chrissis, and Charles Weber, Capability Maturity Model for Software, Version 1.1, CMU/SEI-93-TR-024, Software Engineering Institute, Carnegie Mellon University, February 1993, retrieved 2026-09-24 — https://www.sei.cmu.edu/documents/1092/1993_005_001_16211.pdf
• John Leslie King and Kenneth L. Kraemer, "Evolution and Organizational Information Systems: An Assessment of Nolan's Stage Model," Communications of the ACM 27(5), 466–475, 1984 (UC Irvine technical report version), retrieved 2026-09-24 — https://escholarship.org/content/qt9ws8n556/qt9ws8n556.pdf
• Izak Benbasat, Albert S. Dexter, Donald H. Drury, and Robert C. Goldstein, "A Critique of the Stage Hypothesis: Theory and Empirical Evidence," Communications of the ACM 27(5), 476–485, 1984, retrieved 2026-09-24 — https://doi.org/10.1145/358189.358076
• Tanja de Bruin, Michael Rosemann, Ronald Freeze, and Uday Kulkarni, "Understanding the Main Phases of Developing a Maturity Assessment Model," Australasian Conference on Information Systems, 2005, retrieved 2026-09-24 — https://eprints.qut.edu.au/25152/1/Understanding_the_Main_Phases_of_Developing_a_Maturity_Assessment_Model.pdf
• Jens Pöppelbuß and Maximilian Röglinger, "What Makes a Useful Maturity Model? A Framework of General Design Principles for Maturity Models and Its Demonstration in Business Process Management," European Conference on Information Systems, 2011, retrieved 2026-09-24 — http://fim-rc.de/Paperbibliothek/Veroeffentlicht/327/wi-327.pdf
• Peter Weill, Stephanie L. Woerner, and Ina M. Sebastian, "Building Enterprise AI Maturity," MIT CISR Research Briefing, December 19, 2024, retrieved 2026-09-24 — https://cisr.mit.edu/publication/2024_1201_EnterpriseAIMaturityModel_WeillWoernerSebastian
• Statistics Canada, "Artificial intelligence adoption and productivity in Canadian firms," Economic and Social Reports 36-28-0001, April 2026, retrieved 2026-09-24 — https://www150.statcan.gc.ca/n1/pub/36-28-0001/2026004/article/00002-eng.htm
• Statistics Canada, Canadian Survey on Business Conditions, second quarter 2026 (use of AI and relevance by employment size), June 2026, retrieved 2026-09-24 — https://www150.statcan.gc.ca/n1/pub/11-621-m/11-621-m2026010-eng.htm
• Eurostat, Artificial intelligence by size class of enterprise (isoc_eb_ai), EU27, updated 15 June 2026, and news release of 11 December 2025, retrieved 2026-09-24 — https://ec.europa.eu/eurostat/web/products-eurostat-news/w/ddn-20251211-2
• Eurostat, Use of artificial intelligence in enterprises, Statistics Explained, retrieved 2026-09-24 — https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises
• Tamara Levy and Philip Powell, "Exploring SME Internet Adoption: Towards a Contingent Model," Electronic Markets 13(2), 173–181, 2003, retrieved 2026-09-24 — https://electronicmarkets.org/fileadmin/user_upload/doc/Issues/Volume_13/Issue_01/V13I2_Exploring_SME_Internet_Adoption__Towards_a_Contingent_Model.pdf
• Sukanlaya Sawang and Virach Sornlertlamvanich, "Artificial Intelligence (AI) Maturity in Small and Medium-Sized Enterprises: A Framework of Internalized and Ecosystem-Embedded Capabilities," arXiv:2603.08728, February 2026 (preprint), retrieved 2026-09-24 — https://arxiv.org/abs/2603.08728
• Accenture, "More Than 60% of Companies Are Only Experimenting with AI" (on The Art of AI Maturity: Advancing from Practice to Performance), June 8, 2022, retrieved 2026-09-24 — https://newsroom.accenture.com/news/2022/more-than-60-percent-of-companies-are-only-experimenting-with-ai-creating-significant-opportunities-for-value-on-their-journey-to-ai-maturity-accenture-research-finds
• Deloitte AI Institute, State of AI in the Enterprise, 4th edition, press release, October 2021, retrieved 2026-09-24 — https://www.prnewswire.com/news-releases/state-of-ai-in-the-enterprise-fourth-edition-becoming-an-ai-fueled-organization-a-survey-from-the-deloitte-ai-institute-301405642.html
• MITRE, AI Maturity Model, retrieved 2026-09-24 — https://aimaturitymodel.mitre.org/
• Microsoft, Agentic AI adoption maturity model, Microsoft Learn, retrieved 2026-09-24 — https://learn.microsoft.com/en-us/agents/adoption-maturity-model/
• AWS Prescriptive Guidance, Generative AI maturity model: maturity levels, retrieved 2026-09-24 — https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-gen-ai-maturity-model/overview-levels.html
• Google Cloud, AI Adoption Framework (white paper), retrieved 2026-09-24 — https://services.google.com/fh/files/misc/ai_adoption_framework_whitepaper.pdf
• BMC, AI Maturity Models (reproducing Gartner's five levels), retrieved 2026-09-24 — https://www.bmc.com/blogs/ai-maturity-models/
• AI Sweden, AI Maturity Assessment, retrieved 2026-09-24 — https://www.ai.se/en/adoption/ai-maturity-assessment

On the page review. "The 18 pages" means the distinct pages ranking for "ai maturity model," "ai maturity model levels," "ai maturity framework," "gartner ai maturity model," "ai maturity model small business," and "ai maturity assessment" on 2026-09-24, fetched and read in full. They included pages from BMC, MITRE, Microsoft, WitnessAI, The Decision Lab, MIT Sloan, Credo AI, the US General Services Administration, Helpware, Ekipa, State of AI for Small Business, Fieldguide, KPMG, Berkeley Partnership, AI Sweden, an open-source GitHub model, and two academic papers. Gartner's own pages were blocked to automated retrieval and aren't counted. "Criteria you could check" means per-level traits observable in the business (Microsoft, KPMG, and the GitHub matrix). The one page that named a headcount under 200 addressed businesses of 1 to 50 employees. It's a snapshot of one day's results, not a market survey.

On Gartner's level names. Gartner's own model is published to clients. The five names used here appear consistently across the vendor pages that reproduce it, including BMC's, and we've attributed them that way.

On the MIT CISR figures. Growth and profit are percentage points above or below industry average in 2022. Stages were assigned by the survey's Total AI Effectiveness measure (Stage 1 at 0–49%, Stage 4 at 100%), so the relationship between stage and performance is partly built into the grouping. The briefing doesn't report the size of responding firms, and we don't present it as evidence about smaller businesses.

On the Statistics Canada productivity study. The figures are from the agency's summary of Li and Liu's forthcoming paper, using firm-level survey data from 2019 and 2021 linked to administrative records. It measures AI adoption in general, before the spread of generative AI tools.

On the Eurostat figures. The indicator is the share of enterprises with 10 or more employees using at least one AI technology, EU27, excluding agriculture, mining, and the financial sector. The barrier shares apply to firms that had considered AI but didn't use it. The industry shares (real estate 24.8%, construction 10.8%) are from Eurostat's 2025 Statistics Explained article. The 2025 figures come from the survey collected in early 2025.

On the fire-protection contractor. It's an illustrative composite, not a client. Its figures (8 technicians, about 60 inspection reports a week) are representative, not measured.

On first-party claims. Statements such as "from our own diagnostics" describe Avolis's own engagements. They are not independent research and are not offered as benchmarks.

---

About Avolis Research Group

Avolis Research Group is Avolis's in-house research practice, focused on how operations-heavy small and mid-sized businesses actually adopt AI. It synthesizes primary economic research, government survey data, and results from real implementations into practical, vendor-neutral guidance.

More about Avolis and how we work · Get in touch