AI readiness assessment: the five checks that predict success

Around 70% of firms now use AI, and close to 90% of them report no measurable productivity impact, according to a study of nearly 6,000 executives from the National Bureau of Economic Research.

Run properly, an AI readiness assessment doubles as a check on your wider strategy and foundations. The same questions that reveal whether you're ready for AI also expose where priorities are fuzzy, where goals lack owners, and where KPIs were never really agreed, and it's far cheaper to close those gaps now than to discover them mid-execution with budget already committed. So treat what follows as two exercises in one: the questions and answers below will tell you whether your department or business is set up for AI operations and analytics, and along the way they'll force the kind of clarity on priorities and measurement that most teams only reach after something has already gone wrong.

The five quiz, explained

1. Data foundation. Can your key business data be brought together in one place, and would two (or more!) departments agree on the numbers once it was? If your revenue figure differs between the CRM and the finance system, no AI initiative built on top can be trusted, because its outputs inherit this disagreement. Judge yourself honestly: consolidated and reconciled data scores high, a warehouse with known gaps scores just OK, and reporting stitched together from exports scores low. If you’re already running into problems here, you need to fix your data foundation first. Our framework for choosing a data stack is a good starting point.

2. Use case clarity. You need to have clear goals for AI implementation in order to get clear results. “We want to use AI” is too vague to judge fairly. What you need is a business decision that would be made differently with a reliable prediction or a faster answer, for example:

• Which products to reorder for Q4?
• Which customers are about to churn and what saving them is worth?
  The litmus test is whether or not you can give the use case as a question with a financial answer. If "leadership wants an AI story," score low, because projects without a specific decision/outcome attached don't just fail, they never end up mattering.

3. Team and context. Two different resources often get confused here. Technical capacity is whether anyone can build and maintain models, and it's the one companies often fixate on. Business context is whether the people who understand how your business works have time to teach that context. Context is the scarcer resource and the one AI cannot work without (if you want useful results). A team with no ML engineers but deep, available business knowledge is more ready than a team with data scientists and no time from the business, and this is a variable where partnering with someone else changes the outcome fastest, because an embedded deployment team supplies the technical half while your people supply the context.

4. Workflow integration. Where will the AI's output actually land? If the answer is a dashboard someone might check, score yourself low. If the answer is a specific meeting, a specific decision, a specific system that will consume the output automatically, score high. AI that lives beside the work gets admired and ignored. The readiness question is whether you've identified the exact point in an existing process where a prediction or an answer changes what happens next, and whether the owner of that process has agreed to change it.

If your team already runs everything in Claude, Cursor, or ChatGPT, then the workflow you're integrating into is that conversation, and the readiness check now becomes whether your data platform seamlessly join it. That's what MCP makes possible, an open standard that connects AI clients directly to your pipelines and warehouse. We've written about what that looks like for data teams in practice, and it changes this dimension's scoring a little: a team that works in Claude all day, connected to their actual data through MCP, is far better integrated than a team with a beautiful looking dashboard nobody uses.

5. Measurement. Before anything is built, can you state the baseline you're improving on and the number that will prove it worked? Companies that skip this end up in the 90% that can't demonstrate impact, not necessarily because there wasn't any, but because nobody measured the before. A readiness assessment that doesn't force a baseline is setting up next year's "AI isn't delivering" conversation.

Scoring it

Having a strong data foundation with a weak everything else means you're likely closer than you think; the remaining dimensions are decisions and process, which move faster than the underlying infrastructure. A weak foundation but strong everything else is the more dangerous profile, because enthusiasm will push you into building on sand, and the first untrusted outputs will burn credibility you can’t easily win back. Uniformly average scores mean nobody has figured out a real use case yet, so start with dimension two (finding a real use case) and the rest should start to fall into place.

One more thing to check

We’ve got a lot of experience doing this work, not just writing about it online! So here’s one more thing that you should be thinking about while assessing your data readiness. AI readiness isn't a state you reach before starting, it's built fastest by starting small with the right support. The companies that get to production quickest are the once who pick one use case with a clear owner and a measurable baseline, get the data foundation right for that slice, ship it, and let the win fund the next one. The assessment's job is to pick that first slice.

That's also a big part of what we do: Kleene.ai deploys the data platform and the embedded analyst team together, which means the two dimensions companies most often fail on, the foundation and the context transfer, arrive as part of the engagement rather than prerequisites for it. Typical time from kickoff to AI in production is weeks, not quarters, precisely because the readiness gaps are handled inside the project rather than before it.

If you've run the assessment and want a second opinion on the scores, bring them to us.