AI Maturity Model: Four Stages You Can Verify
Table of contents
An AI maturity model is a staged scale that places an organization by how it runs AI today, from first experiments to AI built into business operations. In Iron Goo's model, the weakest observable marker limits the company's stage, so spending more does not move it up. Gartner's toolkit lists Foundational, Emerging, Operational, Scaled and Transformational. Many third-party pages give a different set: Awareness, Active, Operational, Systemic and Transformational. We could not find that set on any Gartner page we could read, so check a list against Gartner's own page before using it.
The distinction between buying AI and running it shows up in ordinary office work. A company can buy assistant licenses for every employee and subscribe to a chatbot platform while its only demonstrable use remains individuals drafting text. The subscriptions tell you what the company bought; the work records tell you what changed.
Four published AI maturity models measure different things
Published AI maturity models place organizations on stages using different combinations of business practices, capabilities and technical infrastructure. Their stage names are useful only after you know what each model scores and who it was designed to assess.
| Model | Stages | What it scores | Built on or for | How you get it |
|---|---|---|---|---|
| Gartner | Five: Foundational, Emerging, Operational, Scaled, Transformational | Capabilities across seven AI pillars | Organizations assessing AI capabilities; its survey was conducted in the fourth quarter of 2024 and published in a June 2025 press release, with 432 respondents from organizations in the US, UK, France, Germany, India and Japan | The toolkit describes a guided questionnaire; the detailed report appears to be a Gartner client product |
| MIT CISR | Four: Experiment and Prepare, 28%; Build Pilots and Capabilities, 34%; Develop AI Ways of Working, 31%; Become AI Future Ready, 7% | AI effectiveness in operations, customer experience and the ecosystem | A 2022 survey of 721 enterprises, plus 16 executive interviews in 2024; percentages are shares of surveyed enterprises | The research briefing describes the model and findings |
| MITRE | Five: initial, adopted, defined, managed, optimized | 20 dimensions across six pillars, with one multiple-choice question per dimension | Organizational assessment; the guide lets each organization set its own target level by mission and business practices | The guide is public; request the assessment tool from MITRE |
| Google Cloud AI Adoption Framework | Three: tactical, strategic, transformational | Six themes: Learn, Lead, Access, Scale, Automate and Secure | Organizations building AI and machine learning capabilities; published in June 2020 | The framework and whitepaper are public |
In Gartner's survey, a seven-question assessment rated each area from Level 1, described as planning or beginning, to Level 5, described as leadership. MIT CISR's stages use an equally weighted measure of AI effectiveness across three areas. MITRE and Google score separate dimensions or themes, so a company can see where its capabilities differ rather than relying on a single average.
Microsoft's Cloud Adoption Framework uses the phrase AI maturity level for a different measure: skills and data readiness for building AI solutions. Level 1 pairs a basic understanding of AI concepts with feasible use cases such as any Copilot solution; the table describes what a team can build, not how the whole business runs AI.
The published performance numbers deserve a careful reading. In the MIT CISR enterprise survey, stage 1 enterprises had growth 12.6 percentage points and profit 9.6 points below their industry averages; stage 4 enterprises had growth 17.1 points and profit 10.4 points above those averages. The figures are from the 2022 Future Ready Survey of 721 enterprises, which primarily concerned analytical AI, given when it was conducted.
In Gartner's survey, 45% of leaders in high-maturity organizations said their AI initiatives stayed in production for three years or more, compared with 20% in low-maturity organizations. The MIT CISR figures come from a survey of 721 enterprises and the Gartner figures from leaders' self-reports at organizations in six countries; both are associations, and neither proves that moving up a stage raises profit. Neither source reports results for companies of 10 to 200 people, so applying the findings to a company that size is our inference.
How an AI maturity assessment places a company
An AI maturity assessment places a company by comparing evidence of its current practices with the markers for each stage. Iron Goo's four stages and placement rule are our model, not an industry standard: the company's stage is the highest stage whose markers all pass, together with every stage below it.
A company with no AI use at work has not reached stage 1. For a company already using AI, the checks work as follows:
- →Stage 1: Individual tool use
Name the people who use AI at work and what they use it for. The stage 1 marker passes when you can identify both.
- →Stage 2: One automated process
Find one recurring job where AI produces output that a named person approves. The marker is a recorded baseline from before the change and an after-number for that process.
- →Stage 3: Connected processes
Find at least two automations that hand data to each other or read the same source of truth. The marker is a written list of the systems each automation reads and writes, with a named owner for each flow.
- →Stage 4: Managed operations
Find one inventory covering every automation, with an owner, monthly cost, scheduled re-tests on a fixed set of past cases, alerts reaching a person and model retirement dates on a calendar. The marker is an inventory reviewed within the last 30 days.
Start by listing every AI use in the company, including individual assistants and subscriptions bought by separate departments. Check the markers against documents or logs, then flag spending associated with higher stages as a mismatch to review, fix or cancel.
The rule prevents one polished demonstration from standing in for the company's operating practice. A flow can exist without a before-and-after measure; two flows can exist without a written account of the systems they share. In either case, the missing record tells you which claim you cannot yet make.
Our approximate reading is that Gartner's Foundational and Emerging stages correspond to our stage 1, Operational to stage 2, and Scaled to stages 3 or 4. MIT CISR's first three stages roughly correspond to our stage 1, stage 2, and stages 3 to 4, respectively; its fourth stage extends beyond our scale.
The highest maturity level in all dimensions may not be practicable or relevant.
I think stage 3 or 4 in Iron Goo's model is a realistic ceiling for most companies with 10 to 200 people. Gartner's Transformational stage and MIT CISR's stage 4, whose characteristics include creating and selling AI-augmented business services, usually need not be targets for a company whose business is property management, manufacturing or another non-AI service. A target should describe work the company needs to run, with enough ownership and measurement to keep that work dependable.
An 85-person property manager at stage 1
A hypothetical property management company has 85 employees and takes an inventory of its AI use. Every number in this example is hypothetical. The inventory finds three purchases or uses:
- An AI assistant: The company bought licenses for all 60 office staff, and 30 use the assistant to draft emails and summarize leases. That is stage 1 activity because the record shows individual use.
- A website chatbot platform: The subscription answers frequently asked questions about tenant portal hours, but nobody reviews the answers. The chatbot does not meet the stage 2 requirement for a named person approving AI output, and there is no before-and-after figure.
- A maintenance triage flow: AI tags the urgency and trade for requests in a shared inbox, and a coordinator approves each tag. The approval is present, but nobody timed the work before the flow started, so the company cannot show the required before-and-after figure.
The company therefore sits at stage 1 under our model. Its licenses and chatbot platform are spending associated with the broader deployments a buyer might expect at stages 3 or 4, yet the records do not show a measured stage 2 process. The purchase of a platform does not prove that multiple processes are connected or that operations are managed.
Most activity sits in stage one. The maintenance flow is close to stage two, while spending sits above stages whose markers have not been met.
The maintenance flow already has a named approver, making measurement the missing stage 2 marker. Suppose the company pauses the AI step or times a representative sample of the manual work: at 120 requests a month and four minutes per request, the baseline is 480 minutes, or eight staff hours a month. Those are hypothetical inputs and our arithmetic, not a measured result.
The company can record the same measure after the flow runs and retain both figures with the coordinator's name. Iron Goo's operations service builds a first automation with a person approving its output and keeps the baseline for the before-and-after number.
Public evidence shows why a license count is a weak placement marker: the UK Department for Business and Trade's Copilot pilot issued 1,000 licenses from October to December 2024, averaged daily active users equal to 30% of license holders on working days, and found no evidence that reported time savings led to improved productivity; an unreviewed Census working paper found that 64% of AI-using US businesses in its November 2025 to January 2026 survey reported no organizational adjustments to use AI.
How does AI maturity differ from AI readiness and digital maturity?
AI maturity describes how the company currently runs AI across its uses; AI readiness asks whether one particular job can be automated next. Gartner's own definition connects the concepts by saying a maturity model informs actions to increase AI readiness. The deciding difference in this practical use is the unit being assessed: the company today versus the conditions for one job.
Use the ten-minute AI readiness test when you have selected a process and need to assess that job. The AI readiness guide gives the full per-job standard.
Digital maturity describes the systems and data a company has; AI maturity, in this model, counts processes where AI output is used under an owner and measured. Cloud accounting and a CRM can give staff organized digital records while every AI assistant remains an individual writing tool. That company can have established digital systems and still sit at AI stage 1.
An implementation roadmap answers a separate question: the model says where you are, while the AI upgrade order sets out what to do in what order. A stage label can identify the gap, but the label alone does not assign a person to close it.
The marker that moves a company to stage 2
The stage 2 marker is one recurring process with a timed baseline, a named person approving AI output and a recorded after-number. Among AI-using US employer firms with 1 to 499 employees in the Federal Reserve Banks' 2025 Small Business Credit Survey, 49% said they were experimenting, 44% said AI was partially integrated and 7% said it was fully integrated. The survey was fielded from September to November 2025 using a convenience sample, not a random sample.
The first jobs for AI can help you choose a candidate from work your staff already perform. Pick the one process closest to stage 2 and give it a timed baseline and a named approver this month.
Put your next steps in order
Build on what you have just read with a free ebook on semantic SEO and AI visibility. Get worked examples, decision rules, checklists and a do-this-week action in every chapter, plus a six-week plan for your site.
