How to run a data-readiness assessment before an AI pilot

Most teams skip the readiness assessment because it feels like delay. Then they spend a quarter building on data that could never have supported the feature, which is a much longer delay. A good assessment is short, structured, and honest, and it pays for itself the first time it stops a doomed pilot.

Start at the problem, not the data lake

A product and operations leader who now consults on AI adoption for mid-sized companies keeps coming back to one rule: start from the real business problem, not the technology. He is willing to tell a client that AI is not the right fit, because a consultant who cannot say that never proves any return. The assessment inherits that discipline. You do not assess “our data” in general. You assess the specific data that one specific outcome requires. Everything else is out of scope for now.

Do the discovery before anyone builds

A CTO whose firm has shipped software across healthcare and fintech told us quality starts at the pre-sale stage, not after the code is written. His team runs discovery sessions to understand the actual problem, documents each step, and defines what done means before a single sprint begins. The same order works for data readiness. Before a pilot, sit down and write out, for this use case, exactly which records the feature reads, how fresh they need to be, and who is allowed to use them. The teams that spend six months explaining their business logic after the build has started are the ones that skipped this.

Check five dimensions, not just “is it clean”

Run the data for the use case against five questions. Quality: is it accurate at the record level the feature needs, not the aggregate level a report needs. Governance: are you allowed to use it this way. Architecture: can a production service reach it inside the latency budget. Discoverability: can you actually find every relevant record, or are some siloed in a system nobody mentioned. Compliance: does any field carry obligations, like personal data, that reshape the design. A pilot can pass four of these and still be blocked by the fifth.

Timebox it

A focused assessment scoped to one use case takes about two weeks. An enterprise-wide readiness effort runs six to ten. That gap tells you something important: the way to keep an assessment fast is to keep it narrow. Do not try to certify the whole company as “AI ready,” because there is no such state. Certify that this data can support this feature, ship the pilot, and assess the next use case when it comes.

Write the verdict down

The output is not a feeling, it is a short document: for this use case, here is what is ready, here is what is not, and here is the specific work required to close each gap. That document is what lets a leadership team decide to proceed, to fix first, or to drop the idea, without the decision turning political later.

How we approach it at Density Labs

The AI Readiness Assessment is exactly this assessment, run as a fixed two week engagement priced at $2,500. We scope to one use case, walk the five dimensions, and hand you a written verdict with a prioritized fix list. It is deliberately narrow and deliberately fast, because a readiness assessment that takes a quarter has become the pilot.

Two weeks of honest assessment is cheaper than three months of hopeful building. Do the boring pass first.