Data programmes have a reputation problem because they are usually scoped as estate-wide clean-ups. That framing guarantees a multi-year programme with a benefit that arrives after the current leadership team has moved on.
The alternative is decision-led data. Start from the twenty or thirty decisions that carry the most value in the next four quarters — credit approval, assortment, crew rostering, claims triage, supplier risk — and treat the data each one needs as a product with an owner, a contract, a freshness guarantee, and a published quality score.
“Twenty owned data products beat a pristine estate you will finish building in 2029.”
This reframing changes the economics. Instead of a two-year platform build, you get a sequence of data products that each unlock measurable value within a quarter, and a platform that grows underneath them because it is being pulled rather than pushed.
Three disciplines make it stick. First, publish quality scores where the decision makers can see them, because a data product with no visible score is a data product nobody trusts. Second, give every data product a single accountable owner with the authority to refuse requests that would degrade it. Third, retire data products aggressively; estates grow by accretion and shrink only by decision.
The payoff is not elegance. It is speed: teams that have decision-led data products ship AI use cases roughly three times faster, because the first six weeks of every project stop being an archaeological dig.
