AI Maturity Assessment: How to Run One [2026]

An AI maturity assessment tells you where your organisation actually stands on AI capability (infrastructure, governance, data, and people) so you can invest in the right gaps instead of guessing. Most frameworks cover the technology and process layers well. The people layer is where they break down, because the default method (asking employees to rate their own AI skills) produces numbers that feel precise but consistently overstate reality.

This guide walks through how to run an AI maturity assessment that gets the people layer right. It is written for the leader who has been asked to produce one and needs a method that will hold up in a board presentation, not just a slide deck full of self-reported confidence scores.

What Is an AI Maturity Assessment?

An AI maturity assessment is a structured evaluation of how effectively an organisation uses artificial intelligence across its operations, strategy, and workforce. It produces a current-state score, identifies gaps, and gives leadership a basis for prioritising investment.

Most AI maturity assessment frameworks break the evaluation into four to six dimensions. The exact labels vary, but the substance is consistent:

The Core Dimensions

• Strategy & Vision, Does the organisation have a stated AI strategy? Is it connected to business outcomes? According to Gallup (May 2026), only 25% of US employees say their organisation has communicated a clear plan for integrating AI. If three-quarters of your workforce cannot describe the plan, your strategy dimension is weak regardless of what the C-suite believes.
• Data & Infrastructure, Are data pipelines, governance policies, and compute resources in place to support AI workloads at scale?
• Governance & Risk, Are there AI governance frameworks covering model selection, data privacy, bias monitoring, and regulatory compliance?
• Process & Workflow, Has the organisation identified and redesigned workflows where AI creates measurable value? This includes task decomposition, breaking work into components and deciding which ones benefit from AI augmentation.
• People & Skills, Can employees actually use AI tools effectively? This is the dimension most assessments handle poorly, and the one this guide focuses on.

Where Most Assessments Stop Short

The technology dimensions are relatively straightforward to audit. You can inventory your cloud infrastructure, count your deployed models, and review your data governance policies. The people dimension is harder because the obvious measurement method (a self-assessment survey) introduces systematic error. We will cover exactly how much error in the next section.

What Should an AI Maturity Assessment Measure?

A complete AI maturity assessment framework should measure demonstrated capability, not just adoption metrics or sentiment. The difference matters: adoption tells you who has access to tools; capability tells you who can use them to produce reliable output.

Technology and Process Metrics

These are the dimensions where existing frameworks perform well. Typical AI maturity assessment questions in these areas include:

• How many production AI workloads are running, and what is their uptime?
• What percentage of business-critical decisions incorporate AI-generated analysis?
• Is there a model registry with version control and evaluation frameworks?
• Are there documented rollback procedures for AI-dependent processes?

These questions have verifiable answers. You can pull them from systems, audit logs, and architecture diagrams.

People Metrics, The Gap Most Frameworks Miss

The people side of AI maturity requires measuring what employees can actually do with AI, across multiple skill dimensions. AISA's rubric, for example, evaluates 11 criteria across five dimensions: Prompting & Communication, Critical Thinking, Technical Understanding, Workflow & Application, and Safety & Responsibility.

The critical point: you need to measure skill, not collect opinions about skill. A survey that asks "How confident are you using AI for data analysis?" on a 1–5 scale tells you about confidence. It tells you nothing about whether the person can detect a hallucinated statistic, structure a multi-step prompt, or recognise when a model's output needs human review.

What Good People-Layer Measurement Looks Like

A rigorous people-layer assessment should:

1. Test actual behaviour, not self-reported frequency of tool use
2. Cover multiple skill dimensions, not just prompting
3. Produce evidence-linked scores, every rating tied to something the person said or did
4. Resist gaming, the scoring mechanism should be separate from the interaction
5. Generate role-level and team-level aggregates that feed into the broader maturity model

This is where a team AI assessment fits into the broader maturity picture. It provides the measurement layer for the people dimension while your existing frameworks handle technology, data, and governance.

Why Do Self-Assessment Surveys Overstate AI Maturity?

Self-assessment surveys overstate AI maturity because people consistently overestimate their own AI skills. This is not a minor calibration issue, it is a structural bias large enough to shift your maturity rating by an entire tier.

The Prediction Gap in Numbers

AISA asks candidates to predict their score before the assessment begins, then compares that prediction to their measured result. Across 1,358 candidates, the average predicted score was 62.7 out of 100, while the average actual score was 44. That is an average prediction gap of 18.7 points, nearly 19 points of overestimation on a 100-point scale.

To put that in context: a predicted score of 62.7 would place someone in the Proficient tier (60–79). An actual score of 44 places them in the Developing tier (28–59). The self-rating does not just add noise, it moves the needle from "needs significant development" to "ready to contribute," which are two very different inputs for a maturity assessment.

This pattern is explored in depth in 5 AI Skills Professionals Overestimate.

The Gap Varies by Role

The overestimation is not uniform. Engineering roles show an average gap of 15.3 points (predicted 69, actual 53.7). Product roles show a gap of 7.4 points. Founders show 8.7 points. Design roles show 15 points. The variance means a self-assessment survey will not just inflate your overall number, it will distort the relative picture across teams.

What This Means for Your Maturity Assessment

If your AI maturity assessment relies on self-reported skill levels for the people dimension, you are likely overstating that dimension by roughly one tier. Since the people dimension typically carries 20–30% of the overall maturity score, this cascades into your composite rating.

The fix is straightforward: replace self-assessment with measured assessment. Use a tool that evaluates what people can actually do, scores independently of the conversation, and produces evidence you can audit. An AI readiness assessment built on this principle gives you a people-layer score you can defend.

How Do You Run an AI Maturity Assessment Step by Step?

Running an AI maturity assessment requires sequencing the work so that each phase feeds the next. Here is a practical framework that treats the people layer with the same rigour as the technology audit.

Step 1: Define Scope and Stakeholders

Before you assess anything, answer three questions:

• What is the assessment for? Board reporting, investment prioritisation, M&A due diligence, and regulatory compliance each require different depth and framing.
• What is the boundary? Whole organisation, a single business unit, or a specific function?
• Who owns the output? The assessment needs an executive sponsor who will act on the results, not just file them.

Document these decisions. They determine which dimensions you weight most heavily and how granular your people-layer measurement needs to be.

Step 2: Audit Technology, Data, and Governance

This is the part most teams know how to do. Use your preferred framework, whether that is a vendor maturity model, an internal rubric, or a standards-based approach. Key activities:

• Inventory deployed AI tools and models
• Review data governance policies and access controls
• Assess infrastructure capacity and scalability
• Evaluate AI governance frameworks against regulatory requirements (EU AI Act, NIST AI RMF, internal policies)
• Document integration points and dependencies

The output is a scored assessment of your technology and process dimensions, with evidence for each rating.

Step 3: Measure People Skills (Not Self-Ratings)

This is the step most assessments get wrong. Here is how to get it right:

Do not send a survey. Instead, deploy a measured assessment that evaluates demonstrated AI fluency. AISA's AI readiness assessment works by having each candidate complete a conversational assessment with an AI facilitator, while a separate AI evaluator scores independently against a published rubric. Every score is linked to evidence in the candidate's own words.

The output gives you:

• Individual scores across multiple dimensions (prompting, critical thinking, technical understanding, workflow application, safety)
• Team-level aggregates and distributions
• Role-level comparisons
• A confidence calibration signal, how accurately each person estimated their own ability

From AISA's data across 2,268 completed assessments, the tier distribution shows where most people actually land:

Nearly 70% of people score in the Emerging or Developing tiers. If your self-assessment survey is telling you most of your workforce is Proficient or above, the prediction gap data explains why.

For a deeper look at how organisations have used measured assessment to baseline their teams, see AI Skills Assessment Accuracy, Wingie Case Study.

Step 4: Synthesise Across Dimensions

Now combine your technology/process scores with your people-layer scores into a single maturity view. Weight the dimensions according to your organisation's context:

• Technology-heavy organisations (SaaS, fintech) may weight infrastructure and data higher
• Knowledge-work organisations (consulting, legal, media) should weight the people dimension higher, because AI value creation depends almost entirely on how well individuals use the tools
• Regulated industries should weight governance proportionally to their compliance exposure

The synthesis should produce both a composite score and a dimension-level breakdown. The composite gives the board a headline; the breakdown tells the operating team where to invest.

Step 5: Identify Gaps and Prioritise

Compare your current-state scores against your target state. The gaps become your investment roadmap. Common patterns we observe:

• Technology ahead of people: The tools are deployed, but adoption is shallow because employees lack the skills to use them effectively. This is the most common pattern. Microsoft and LinkedIn's Work Trend Index (May 2024) found that only 39% of people who use AI at work have had AI training from their company.
• People ahead of governance: Individual contributors are experimenting with AI, but there are no guardrails, no approved tool list, and no data classification policy.
• Strategy ahead of everything: Leadership has a vision, but neither the infrastructure nor the skills are in place to execute it.

Each pattern implies a different investment sequence. The maturity assessment's job is to make the pattern visible. For a broader look at building the enablement programme that follows, see AI Enablement: A Playbook.

What Should the Output Look Like for the Board?

The board needs a one-page maturity summary, a dimension-level breakdown, a gap analysis with prioritised recommendations, and a timeline. Anything more gets filed; anything less gets questioned.

The One-Page Summary

Structure it as:

1. Overall maturity tier, a single label (e.g., "Developing") with a composite score
2. Dimension scores, a radar chart or simple table showing each dimension's rating
3. Biggest gap, the single dimension furthest from target, with a one-sentence explanation of business impact
4. Top three recommendations, specific, costed, with expected timeline to move one tier

Supporting Detail

Behind the one-pager, provide:

• People-layer data with measured scores, not self-ratings. Show the prediction gap to explain why previous survey-based assessments may have been optimistic.
• Role-level breakdowns so the board can see which functions are strongest and which need the most development. AISA data shows meaningful variation: Product roles average 55.8, Engineering 54.2, Design 48.1. These differences matter for prioritisation.
• Benchmark context, how your scores compare to external benchmarks. AISA's data, cross-referenced against Anthropic's AI Fluency Index (93% marker coverage) and the U.S. DOL AI Literacy Framework (100% coverage), provides a defensible reference point.

What to Avoid in Board Reporting

Do not present self-assessment survey results as skill data. If you have both self-rated and measured scores, show them side by side, the gap itself is a finding. Do not present tool adoption rates ("82% of employees have used ChatGPT") as evidence of capability. Usage frequency and skill level are different constructs.

For guidance on whether your organisation is ready to act on the assessment results, Is My Company AI Ready? provides a useful diagnostic.

How Often Should You Reassess?

Reassess your AI maturity every six months for the people layer and annually for the full framework. The people layer moves faster because skills change with training, tool updates, and practice. The technology and governance layers move on infrastructure and policy cycles, which are slower.

Why Six Months for People

AI tools and models are changing on a weekly basis, Gemini 4 Argon launched this month, and model capabilities shift with each release. Employee skills need to keep pace. A six-month cadence lets you:

• Measure the impact of any training or AI enablement programmes you have deployed
• Detect skill regression in teams that have not been actively developing
• Update your maturity score with current data before the next planning cycle

AISA's assessment produces stable results (98% of people who retake land in the same or an adjacent tier) so score changes between assessments reflect genuine skill movement, not measurement noise.

Why Annually for the Full Framework

The technology, data, and governance dimensions change on longer cycles. Infrastructure upgrades, policy revisions, and data platform migrations happen quarterly at most. An annual full assessment, supplemented by the six-month people-layer refresh, keeps your maturity picture current without creating assessment fatigue.

Tracking Progress Over Time

Build a simple dashboard that tracks:

• Composite maturity score (quarterly or semi-annual)
• Dimension-level scores with trend lines
• Prediction gap trend (is the gap between self-rated and measured skill narrowing?)
• Tier distribution shifts (are more people moving from Developing to Proficient?)

This gives you a longitudinal view that shows whether your AI investments are translating into actual capability gains. When you are ready to operationalise this measurement, an AI readiness assessment provides the people-layer data feed, while your existing technology audit covers the remaining dimensions.

For a comparison of tools that can support this ongoing measurement, see AI Training for Employees: Measure First.

---

Related reading: AI Maturity Model: 5 Stages & the People Layer, how maturity stages map to skill tiers and what each stage requires.

Related reading: Measure AI Literacy in Employees, a detailed guide to choosing measurement methods for the people dimension.

Related reading: AI Fluency Measurement Tools, an enterprise buyer's comparison of measurement platforms.

Frequently Asked Questions

How do you conduct an AI maturity assessment?

You conduct an AI maturity assessment by evaluating your organisation across four to six dimensions: strategy, data and infrastructure, governance, process, and people. The technology and process dimensions are audited through system inventories and policy reviews. The people dimension should be measured through a skills assessment that evaluates demonstrated capability rather than self-reported confidence. Combine the dimension scores into a composite maturity rating, identify the largest gaps, and prioritise investment accordingly.

What questions are in an AI maturity assessment?

AI maturity assessment questions vary by dimension. Technology questions cover deployed models, infrastructure capacity, and integration architecture. Governance questions address data privacy policies, model risk management, and regulatory compliance. People-layer questions (when done properly) are not survey questions at all but conversational assessments that evaluate prompting skill, critical thinking about AI output, technical understanding, workflow application, and responsible use. The goal is to measure what people can do, not what they believe they can do.

How long does an AI maturity assessment take?

A full AI maturity assessment typically takes four to eight weeks from scoping to board-ready output. The technology and governance audit takes two to four weeks depending on organisational complexity. The people-layer assessment can run in parallel, individual AISA assessments take approximately 20 minutes per person, and team-level results are available as soon as assessments are complete. The synthesis and reporting phase adds one to two weeks. Plan for six months between reassessments of the people layer and twelve months for the full framework.