How to Measure AI Readiness

Measure AI readiness by scoring five dimensions on a 1–5 scale, requiring evidence for every score above 3, and applying the whole exercise to one workflow rather than the whole company. The dimensions that matter are tooling, data readiness, process maturity, automation, and team AI literacy. Anything you cannot verify gets marked unknown and stays unknown.

That method differs from the free vendor tools in one respect that changes the answer: they score the organization, and this scores the work. A $30M company can be a 2 overall and a 4 on the one process worth automating, and a company-level score would have talked you out of the right project.

What dimensions should you score?

Score five: tooling, data readiness, process maturity, automation, and team AI literacy. Each answers a question that can block a build on its own, and each is observable rather than a matter of opinion. Together they cover the ground the six-pillar enterprise frameworks cover, minus the infrastructure weighting that does not apply to a company running on SaaS.

Dimension	The question it answers	What a 5 looks like
Tooling	Is the software running the core work capable, and what AI is already in use?	Systems of record are modern, integrated, and have APIs
Data readiness	Is the data centralized, clean, and reachable?	One source of truth, queryable without an export
Process maturity	Is the workflow defined and repeatable?	Written down, followed the same way by everyone
Automation	How much already runs without someone moving it along?	Handoffs trigger themselves; exceptions are the only manual touch
Team & AI literacy	Will people actually adopt it, and is leadership driving it?	Staff already use AI tools unprompted; a named executive owns the outcome

Governance is missing from that list on purpose. For companies below roughly $50M in revenue it is almost never the binding constraint, and including it produces a scorecard where the lowest number is a compliance item nobody was going to act on this quarter. Add it back the moment you are handling regulated data — healthcare, financial, or anything covered by a client’s own compliance obligations — because then it moves to the top.

How do you score each dimension?

Use a 1–5 scale where every level has a written, observable definition, and require evidence for any score above 3. Without written level definitions, a self-scored 4 means “we feel fine about this,” which is not a measurement. Microsoft’s AI Readiness Wizard does this well: each of its ten questions offers five concrete states rather than a numeric slider, which makes flattering yourself harder.

The evidence rule is what keeps the exercise honest. A score of 4 on data readiness needs someone to actually run the query. A 4 on process maturity needs the written procedure produced. A 4 on automation needs the integration shown working. If the evidence cannot be produced in the room, the score is a 3 at most, or unknown. In practice this single rule moves most first-pass scores down by a full point, and the drop is concentrated in process maturity, where written procedures turn out to describe a process the team stopped following two years ago.

Mark gaps unknown rather than estimating them. An industry average dropped into a gap looks like data and behaves like a guess, and it will be quoted back to you in six months as though it were measured. Unknown is a legitimate score. It tells you what to go find out.

Who should answer the questions?

The people who do the work, not only the executive who commissioned the assessment. Every free readiness tool is self-scored by one person, and that person is almost always a leader describing how they believe the process runs. Interview the operators separately and the two accounts diverge, usually in the number of manual steps and always in the direction of more of them.

The practical version is not elaborate: talk to two or three people who touch the workflow daily, ask them to walk through the last time they did it rather than describe it in general, and write down every place they switch systems or wait on someone. The switches and the waits are where the hours are. A general description will not surface them, because nobody counts steps they perform automatically.

Ask what happens when it goes wrong, too. Exception handling is usually the majority of the manual effort and almost never appears in the documented process, which is why automations built from documentation break on contact with reality.

How do you turn scores into a decision?

Treat the score as a gate, not a decision. A readiness score tells you whether a build is likely to survive contact with the organization. It says nothing about whether the build is worth doing, which is a separate calculation involving volume, hours, and what an hour of that work is actually worth.

Keeping those two apart is the discipline that matters most here, and skipping it is expensive. MIT’s NANDA initiative studied over 300 AI initiatives for The GenAI Divide: State of AI in Business 2025 and found that roughly 95% of organizations deploying generative AI saw no measurable impact on the P&L, while about 5% achieved rapid revenue acceleration. Readiness did not sort those groups. Organizations that would have scored well built things nobody used, because a readiness score answers “can we” and never answers “should we.”

Use the gate this way. Any dimension scoring 1 or 2 for the target workflow is a blocker to fix before building. A 3 is a risk to plan around. A workflow where all five sit at 3 or above is buildable, at which point the readiness question is finished and the return question starts.

And calibrate before you react to a low number. Cisco’s framework bands readiness out of 100 — above 86 fully prepared, 61–85, 31–60, and 0–30 — and its 2025 index put 13% of organizations in the top band and 48% in the second-lowest. Scoring badly is the normal outcome.

What should you do with a low score?

Fix the sequence, not the score. A low readiness score is a list of things to repair, and the temptation is to repair all of them before doing anything with AI. That is a two-year program and it is how AI initiatives quietly die. Fix only what blocks the one workflow you chose.

The order we use runs: question whether the work needs to exist at all, delete what shouldn’t be there, simplify and standardize what remains, accelerate the handoffs, and automate last — only where a person doesn’t need to decide. AI is the seventh option in that sequence, not the first. Roughly half the time the honest outcome of scoring a workflow is that a step gets deleted or a system you already pay for gets configured properly, and the AI spend is zero. That is a successful assessment, not a failed one.

Here is the stance, plainly: we would not build an AI workflow on top of a process scoring 2 on maturity. Automating an undefined process encodes whichever version of it the loudest person described, and every exception then becomes an escalation to the people who built it. Standardize first. It is unglamorous, it takes weeks rather than quarters, and it is the difference between an automation that runs and one that gets switched off.

How often should you re-measure?

Annually, or after a change that actually moves a dimension. System migrations, process rewrites, and training rounds move scores. Ordinary quarters do not. Re-scoring more often than that measures noise and produces a trend line that looks like progress without any underlying change.

Re-score the same workflow with the same definitions when you do it, or the comparison is meaningless. Changing the rubric between measurements is the most common way readiness reporting becomes decorative — the number goes up, and nobody can say whether the business did.

Where to go next

What Is an AI Readiness Assessment? covers the category and how readiness differs from maturity. AI Readiness Assessment Tools Ranked compares the free vendor tools if you want a fast score first. The rest of the series sits on the AI assessment guides hub. To have someone else run this against your workflow, the AI Opportunity Assessment is the paid version at a fixed $1,500 on our pricing page, with a free fifteen-minute single-workflow version ahead of it.

Sources

• Cisco, AI Readiness Assessment Tool — 0–100 scoring bands and the 2025 global readiness distribution. Page dated 10 February 2026.
• Microsoft Adoption, AI Readiness Wizard — ten questions with five written maturity states each. Last modified 2 April 2026.
• Eide Bailly, AI Readiness Assessment — data-foundation scoring across quality, governance, and accessibility. Last modified 18 March 2026.
• MIT NANDA, The GenAI Divide: State of AI in Business 2025, reported by Sheryl Estrada, Fortune, 18 August 2025.

Method note. The five dimensions and the intervention sequence are the ones published on our own assessment page and used in client engagements. Third-party framework details were read from the vendors’ live pages in August 2026. Claims about how executive and operator process descriptions diverge, and about how often a scored workflow ends in no AI spend, are our own observations from running interview-based assessments — they are not measured statistics, and are stated as observations rather than findings.