Start with the decision, not the demonstration.
A compelling AI demonstration can create momentum. It cannot tell you whether a company is ready to operate the same idea inside a real workflow. That takes a different conversation: what decision needs to improve, who owns it, which information supports it, and what happens when the answer is incomplete.
Start with a recent example of the work. Ask someone to show the documents they open, the systems they check, the exceptions they handle, and the approvals they wait for. A walkthrough often reveals requirements that a high-level use-case description leaves out.
Assess the foundations together.
People, operations, data, and governance are connected. A reliable data source is useful only when a team can access it appropriately and knows how to interpret it. A trained team still needs a clear workflow. A documented workflow still needs a person who can decide when the proposed solution is ready.
The assessment should record evidence rather than impressions. Useful inputs include process maps, representative cases, source-owner details, access rules, existing performance measures, and relevant policies. Where the evidence is missing, the finding is a preparation task—not a reason to invent a maturity score.
Turn gaps into owned actions.
Consider an inventory-planning idea. The company may already have stock and order records, but product identifiers may not match across systems. Planners may agree on the pain point while using different exception thresholds. Those findings change the first phase of work: reconcile the pilot item set, agree the rules, and name the reviewers.
Each action needs an owner, a dependency, and an acceptance condition. “Improve data quality” is too broad. “Reconcile the in-scope product identifiers and have the data owner approve the result” gives the team something concrete to complete and review.
Build a roadmap that can change.
A useful roadmap is a sequence of decisions supported by evidence. Preparation comes first when prerequisites are unresolved. A bounded pilot follows when the team can test a useful workflow safely. Expansion is a later decision, informed by quality, adoption, operational benefit, and running cost.
The output is more than a readiness document. It is a shared account of what is ready, what needs work, and what the company will do next. That gives an AI strategy a practical starting point—and gives the people delivering it a basis for saying yes, revise, or stop.