AI Readiness Assessment: Score Your Company in Eight Questions
An AI readiness assessment measures whether a company can get dependable work out of AI. The eight-question version scores eight dimensions from 0 to 3. Half ask what the company knows about itself, and half ask how its work runs. The knowledge dimensions (codified, current, governed and connected) build on each other, so a company starts at the first one that scores low.
An annual physical starts with the same readings every year. A nurse takes your blood pressure and draws blood, and the numbers go on a chart. You might feel fine. The chart still decides where the doctor looks, because a reading doesn't care how you feel.
Most companies judge their AI by how it feels. After six months of AI tools, marketing says they work. Sales says every draft needs rewriting, and support quietly stopped opening them. All three are describing the same tools, and nobody has taken a reading.
Two questions decide whether a company can get real work out of AI. Does the company know enough about itself, in writing, for a model to work from? And is the work ready to hand over? Eight readings answer both.
The AI readiness scorecard
Score what's true today, rather than what you plan to fix. Pick each line by its description, and ignore the number until you've chosen.
For each dimension: which of these sounds most like your company? The number beside it is your score. WHAT THE COMPANY KNOWS ABOUT ITSELF Codified 0 Nothing outside people's heads. Onboarding is shadowing. 1 Scattered across slides, docs, and a wiki nobody trusts. 2 Core documents exist and are findable, and they're incomplete. 3 The company is written down in one place, in a form a machine reads. Score: [ ] Current 0 Nobody owns it. Nobody knows what's out of date. 1 Updated when something breaks. 2 Reviewed on a cadence by somebody with three other jobs. 3 A named owner, dated documents, and drift surfaces before a customer finds it. Score: [ ] Governed 0 No register. People claim whatever sounds right. 1 Legal reviews the big things. Everything else is judgment. 2 An approved-claims list covers some material. 3 A register says what may be claimed and what may not, and the work checks against it. Score: [ ] Connected 0 Nothing connected. Every prompt starts from zero. 1 People paste context in by hand, when they remember. 2 One team has it loaded into one tool. 3 The tools people already use query it, and nobody has to remember. Score: [ ] HOW THE WORK RUNS Understood 0 Nobody has mapped how the work gets done. 1 The founder knows. Nobody else has been asked. 2 Some functions have been walked through, most haven't. 3 Every function can describe its own workflow, and the descriptions agree with each other. Score: [ ] Documented 0 Nothing written. It lives in whoever does it. 1 A few checklists, mostly stale. 2 The high-volume workflows are written down. The judgment-heavy ones aren't. 3 Somebody new could run a workflow from the document alone. Score: [ ] Adopted 0 Nobody uses AI for real work. 1 Individuals use it privately, nobody knows for what. 2 One or two functions use it for a named task. 3 AI does a defined share of the work, and the company can say which share. Score: [ ] Handoff-ready 0 A task would have to be explained from scratch every time. 1 Handing one over costs more effort than doing it. 2 A person can hand off a task after reassembling the context by hand. 3 A task goes to an agent with the context already there. Score: [ ] Total: [ ] out of 24
The first four dimensions ask whether an AI tool has anything true to work from. Together they're the four conditions of an AI operating system for a company. The last four ask whether your workflows are ready to hand over. None of the eight covers which data your AI tools may see, and AI data classification sets those rules.
Reading the score
A doctor doesn't treat every number on the chart at once. One reading usually drives the others, and treatment starts there. Add your eight numbers for a total out of 24, then look past the total:
- Knowledge, in order: each knowledge dimension depends on the one before it, since a claims register can't govern documents that don't exist. Walk from Codified to Connected and stop at the first 0 or 1. Start there.
- Work, for handoff: Handoff-ready says how close a task is to going to AI, and AI use case prioritization ranks which task goes first. At 3 the context travels with the task. At 2 a person still gathers it first, and below 2 handing a task over costs more than doing it.
- Disagreement: when two people's scores on one dimension sit two or more points apart, the company has never settled that question. The gap is a finding in its own right.
When Governed comes first: if AI drafts already reach customers, a 0 on Governed moves to the front, wherever it falls in the order. Every one of those drafts can carry a claim nobody approved.
A physical also backs your answers with a lab test, and one cheap test checks the knowledge half. Inside your team's AI workspace, with web search turned off, ask what your company charges and who it sells to. An answer drawn from an outdated document points at Current. A confident guess with no document behind it points at Connected.
A twelve-person company's score
Picture a hypothetical twelve-person software company. It scores Codified 2, Current 1, Governed 0, Connected 1, Understood 2, Documented 1, Adopted 2 and Handoff-ready 1, for 10 of 24.
Two of its teams use AI for internal work, and nothing the tools draft reaches a customer yet. So Current, its first low knowledge score, is where it starts. The documents exist, and nobody owns keeping them true. The company needs an owner before it needs another seat.
The day its AI starts drafting for customers, the 0 on Governed jumps the line. How to build an AI-native operating system covers the order of the fixes from there.
Cost of low scores
A low reading costs the company something every week, usually without a line on any budget.
| Dimension | Cost |
|---|---|
| Codified | Each hire learns the business by asking around, and each new document starts from a blank page |
| Current | A price or policy change reaches the documents only if someone remembers to update them |
| Governed | Marketing copy can promise a feature or a result nobody signed off on |
| Connected | People paste in the same background before every task, and the effort never carries over |
| Understood | A few people hold the real process in their heads, and the work stalls when they take a week off |
| Documented | A departure takes a workflow with it, and the next hire rebuilds it from scratch |
| Adopted | Nobody can say how much of the week AI saves, or on which work |
| Handoff-ready | An agent can't take a task until someone gathers the files and rules it needs |
Scoring mistakes
Most bad readings come from how the scorecard gets filled in, rather than from the company itself.
- Scoring the plan: the wiki you mean to write scores 0 today. Score what exists, and keep the plan for the next pass.
- Scoring alone: "The founder knows. Nobody else has been asked." is the description of a 1 on Understood. Ask the people who do the work.
- Buying seats first: more licenses can lift Adopted a point and leave the knowledge half where it was.
- Calling the wiki done: a wiki can earn a good score on Codified and still sit at 0 on Current and Governed. Knowledge base vs. AI operating system explains why it meets one condition of four.
Deciding where to start
A physical is worth taking when you feel fine, and so is this one. If the knowledge half scores low, fix the first weak condition before you buy more seats. If all four knowledge dimensions score 3, the company doesn't need a build. It may need somebody to keep it true.
My free Cadence consultation scores your company against the same four conditions and shows where your AI performs and where it fails.
Take the readings before you pick the treatment.
Common questions
What's a good AI readiness score?
A good AI readiness score has no passing grade, because a high total can hide the one low reading that matters. The useful number is the first knowledge dimension, in order, that scores 0 or 1. A company with every knowledge dimension at 3 doesn't need a build, only someone to keep it true.
Who in the company should take the AI readiness assessment?
The AI readiness assessment works best when the founder takes it alongside the people who lead each function and use AI for real work. Four to six respondents give a truer picture than one, because each sees a different part of how the company runs.
Is an AI readiness assessment the same as an AI maturity model?
An AI readiness assessment and an AI maturity model overlap without being the same thing. A maturity model usually rates how far a company has adopted AI. In the eight-question assessment, adoption is one dimension of eight, and half the questions go to what the company knows about itself.
Why doesn't the assessment score infrastructure or talent?
The eight-question assessment leaves out infrastructure because a startup with traction buys its AI as subscriptions, and those tools are only as good as what they know about the business. Talent shows up in the work half, where Understood, Documented and Handoff-ready measure whether the team can hand work to AI.
How often should a company retake the AI readiness assessment?
A company should retake the AI readiness assessment after fixing one of its low dimensions, so the score visibly moves. A large change is another reason to retake it, such as new pricing, a new buyer or a new person running a function.