AI Readiness Assessment: The 5 Dimensions to Score Before Your Next AI Project

An AI readiness assessment is a structured check of whether an organization's data, knowledge, systems, governance and people can support an AI project in production, not just in a demo. It scores each of those five dimensions, names the weakest one, and tells you what to fix, or which project to pick, before you spend a quarter building something that stalls.

Most AI projects that fail do not fail on the model. They fail on a scanned contract nobody can parse, a permission model that does not survive ingestion, a security review with no controls to point at, or a pilot with a champion but no owner. Every one of those problems was visible before the project started. An AI readiness assessment is simply the discipline of looking first.

This guide walks through the five dimensions, what "ready" looks like in each, the warning signs, and three questions to score. If you'd rather not score by hand, the free AI Readiness Scorecard asks fifteen questions across the same five dimensions and returns an instant, benchmarked result.

What is an AI readiness assessment?

An AI readiness assessment measures the conditions that decide whether AI will work in your organization, as opposed to whether AI works in general. It is not a maturity survey about strategy slides or a count of how many teams have tried a chatbot. It looks at concrete things: whether your documents are findable and current, whether your systems can be reached safely, whether governance exists before the first incident, and whether anyone will own the result.

The output should be three things: a score per dimension, the weakest dimension named plainly, and a recommended first move. A score without a next step is trivia.

Why run an AI readiness assessment before your next AI project?

Because pilot success predicts production success far less than people expect. A demo runs on clean content with a human quietly steering; production meets the real estate. We have written about the five things that stop pilots reaching production: missing context, no integration path, unresolved governance, no owner, no evaluation. An AI readiness assessment checks for all five before you commit budget, rather than in month four.

It also changes which project you pick. A team that scores well on knowledge and poorly on systems should start with a cited question-answering use case, not an agent that writes to the ERP. The assessment points you at the project your organization can actually ship.

The 5 dimensions of AI readiness

1. Data: can AI read what it needs?

What ready looks like. The structured data a use case depends on is reachable through an API or warehouse, has a known owner, and is reasonably clean. Figures that matter come from a system of record, not from a spreadsheet someone emails on Mondays.

Warning signs. Key reports are rebuilt by hand each month. Two systems disagree on customer counts and nobody knows which is right. Data reaches the warehouse days late.

Questions to score.

• Can the data for our first use case be read programmatically, today, without a new integration project?
• Does each key dataset have a named owner who can answer for its quality?
• If two systems disagree on a number, is there an agreed source of truth?

2. Knowledge: can AI answer from your documents?

What ready looks like. Policies, procedures, contracts and runbooks live in a handful of known places, current versions are identifiable, and scanned or image-only documents are the exception. Business terms like "active customer" have one definition.

Warning signs. The same procedure exists in four versions across SharePoint and Confluence. Answers depend on asking the one person who has been there twelve years. Critical documents are scans nobody has ever made searchable.

Questions to score.

• Could we point to the authoritative version of our ten most-used procedures within an hour?
• Are our most important documents machine-readable, or mostly scans and images?
• Do teams agree on the definitions of the metrics and terms our first use case relies on?

3. Systems: can AI reach and act on what you run?

What ready looks like. The systems in scope expose APIs, permissions in those systems are accurate, and you know how they depend on each other. If an agent will write back, there is a safe, scoped way to grant that.

Warning signs. Integrations run through CSV exports. Nobody can draw how an invoice actually moves. Access rights in the document store were last reviewed years ago, so "respect source permissions" would respect the wrong ones.

Questions to score.

• Do the systems in scope have APIs we are allowed to use?
• Are the access rights in those systems accurate enough that an AI applying them would be correct?
• Can we describe, with evidence, which processes and people depend on those systems?

4. Governance: will security and compliance say yes?

What ready looks like. An AI policy exists and is enforceable, not just published. You know which AI tools are in use, including the ones nobody approved. There is a way to mask personal data, require approval for sensitive actions, and produce an audit record. The NIST AI Risk Management Framework is a useful outside reference for what "govern" should cover.

Warning signs. Every AI project triggers a fresh, months-long security review. IT's list of AI tools is far shorter than what expense reports show, the classic sign of shadow AI. Nobody could answer "show me exactly what the AI did" for last week.

Questions to score.

• Do we have an inventory of AI tools in use, sanctioned and unsanctioned?
• Can we enforce policy (data scopes, approvals, PII masking) on AI requests, rather than only document it?
• Could we produce a verifiable record of what an AI system saw and did, if a regulator asked?

5. People: will anyone own it and use it?

What ready looks like. Each use case has a business owner accountable for it next quarter, not just a sponsor for the pilot. Users are involved in testing with their own questions, and there is a plan for training. The EU AI Act's AI literacy obligation has made that last point a compliance question as well as an adoption one.

Warning signs. The project team writes all the test questions. The pilot's champion is moving roles. Nobody has budgeted time for the experts whose knowledge the system depends on.

Questions to score.

• Is there a named owner who will run this use case after launch?
• Will real users, from more than one role, test it with their own questions?
• Have the subject-matter experts it depends on been given time to review answers?

How to score your AI readiness assessment

If you score by hand, rate each of the fifteen questions 0 to 3 (0 = no, 1 = partly, 2 = mostly, 3 = yes, with evidence). That gives each dimension a score out of 9 and a total out of 45. Then read the bands below, and pay more attention to your lowest dimension than to the total.

Two rules override the total. Any dimension scoring 3 or below out of 9 is a blocker for anything beyond a read-only pilot. And a governance score of 0 to 3 means stop and fix that first, because it is the dimension that can freeze a working pilot overnight.

A self-assessment is a starting point, not a measurement. When you need the evidence-based version at the level of individual systems and processes, an Enterprise Twin scores AI-readiness from 0 to 100 per team, process and system, from connected evidence rather than opinion.

How to run an AI readiness assessment in five steps

1. Pick the use case before the assessment. Readiness is always readiness for something. Score against one candidate project, or the answers drift into generalities.
2. Score with the people who know. A data owner, a security lead, a process owner and two future users will score more honestly than a project team alone.
3. Demand evidence for every 3. "Yes, with evidence" means you can show it: the API documentation, the inventory, the named owner.
4. Name the weakest dimension in one sentence. That sentence is the most useful output of the whole exercise.
5. Turn the result into a test. The best confirmation of a readiness score is a small trial on your own content. Our guide to running a proof of concept on your own documents shows how, and our AI implementation roadmap covers the practical next step of scoping the work on your own documents.

AI readiness checklist vs. AI readiness assessment

A checklist asks whether something exists. An assessment asks whether it is good enough for a specific project, and which gap to close first. Use a checklist to prepare; use an assessment to decide. The fifteen questions above work as both, as long as you insist on evidence and score against a real use case.

From readiness score to business case

A readiness score tells you whether you can ship. A business case tells you whether you should. Once the weakest dimension is known, size the value of the first use case with the formulas in how to calculate AI ROI, or put a number on the hours lost to knowledge friction with the ROI Calculator. Then pick the layers you will need, which our enterprise AI platform checklist breaks down one by one.

The fastest way to get a baseline is still the five-minute AI Readiness Scorecard: fifteen questions, five dimensions, an instant score and the modules to start with.

Frequently asked questions