The problem.
BT's advice business sat across a decade of acquisitions, several platforms, and fragmented client data. The wind-down required a single accurate record per client, assembled across all of it.
The clock was external. A board-mandated deadline meant the timeline didn't move. Only the scope and the approach could.
The approach.
Run it like a product, not a data project. Customer first (the downstream teams who would use the data), then spec, then build.
- Mapped every source system that held a piece of the client truth, and the gaps between them.
- Designed reconciliation rules with the ops and compliance teams that would eventually rely on them.
- Shipped iteratively, one client cohort working end-to-end, then scale, rather than waiting for the perfect schema.
- Quality bar set by the strictest consumer (regulator reporting), not the most lenient.
The outcome.
Single source of truth delivered on the board-mandated timeline.
400k+ clients with reconciled records, used as the basis for ongoing servicing and external reporting.
What I learned.
Boring products with board deadlines are some of the highest-stakes product work you can do. Treat them with the same discipline as anything customer-facing.
The people who eventually consume the data are the only useful spec reviewers. Design with them in the room, not for them at a distance.
Data ownership without product discipline becomes a backlog of one-offs. Run it as a product or it runs you.