Technical due diligence before an AI build

AI due diligence should start before a company commits the roadmap. The question is not whether a model can produce a useful answer in a demo. The question is whether the product, data, infrastructure, and team can operate the feature after customers depend on it.

That review has to look below the prompt.

Product boundary

The review should name the user action the AI feature changes. "Add AI to onboarding" is too broad. "Read an uploaded policy document and draft the first compliance checklist" is narrow enough to design, test, and price.

A narrow product boundary exposes the real risks: wrong source data, missing permissions, unreviewed answers, slow response time, unclear ownership, or a customer expectation the product cannot meet.

Data path

The data path shows whether the system can answer the question it is being asked. A due diligence review should trace where source records live, who can read them, how often they change, how deleted records are removed from search, and what the model sees at request time.

This is where many AI plans break. The model is ready, but the company has no reliable path from the production database to the prompt.

Operating path

AI features need the same operational discipline as the rest of the product. The review should check logs, traces, retries, rate limits, spend limits, human review queues, incident ownership, and rollback behavior.

The most useful question is blunt: when a customer reports a wrong answer, can the team reconstruct the input, retrieved context, model output, review state, and product action?

If the answer is no, the feature is not ready for broad release.

Team path

The team path matters because AI work crosses product, application code, infrastructure, data, security, and support. A due diligence review should say who owns each layer and which hires or vendors are needed next.

That output gives founders a better decision than "hire AI engineers." It says which skill is missing, why the current team cannot cover it, and what the first production milestone should be.