Every AI project we have seen fail, failed on data — not on models. Before we scope anything, we walk clients through five questions.
One: where does the data live, and who can grant access? If the answer involves three departments and a committee, the timeline starts there.
Two: is there a single source of truth, or five spreadsheets that disagree? Reconciliation is a project of its own.
Three: how far back does clean history go? Forecasting needs seasons, not weeks.
Four: what personal data is involved, and under which consent? PDPL compliance is designed in, not bolted on.
Five: who owns the numbers after launch? A dashboard nobody maintains is a screenshot.
If you can answer all five in one meeting, you are ready. If not, the answers are the first deliverable.