Operators’ Corner · v3
Data is not just code: why reliable AI answers require context, evidence and disciplined delivery
By Richard White ·
Free, no form.
Abstract
A worked product-construction supplement follows a late invoice payment through impact analysis, contract tests, data correction and consumer recovery.
Ask an AI assistant how many claims were open at quarter-end and it can return a precise number that answers the wrong question. Connecting a model to data provides access; reliable answers also need the applicable meaning, explicit authority, controlled execution and evidence that the result is fit for its use.
Part one is for senior leaders: what data products share with software engineering and the additional evidence they carry, context as an operational asset, the different risks of AI that builds, answers and acts, and what to fund. Part two is for engineering and data leads: discovery, agreed meaning, a release pipeline whose gates match the consequences, layered testing, migration, operation, a worked example, and how coding agents change delivery without removing its controls.
In the paper
Three passages, as published
Access is not context
Connecting a model to data gives it access. Access alone doesn’t give it that context.
Shared disciplines
So the lesson is to apply software discipline to the whole data product, its meaning and evidence included.
The goal
A system that answers quickly when the meaning is settled, clarifies when it’s ambiguous and stops when the evidence falls short.
Next step
Discuss this paper against your own program.
A free sixty-minute working session with Richard on your top gaps. You leave with a one-page summary within three business days, and three next questions.