Data becomes a product when someone can decide through it

Observations and measures first need provenance and a quality state. Comparison and uncertainty determine whether a change is meaningful; a signal may then become an alert or recommendation.

A responsible person inspects the evidence, makes an authorised decision and follows the action through completion, failure or reversal. A later outcome informs evaluation but does not, by itself, prove causation.

The information architecture definition explains why objects, labels and relationships matter before a team chooses another visualisation.

The ConnectX case shows an adaptive approach. Eight situations linked user intent to required data, interface modules, digital-twin behaviour and action. The product did not display every energy metric to every role.

The Doodle case shows why a computed result needs product meaning. Availability alone could not represent meeting priority, so tentative schedule and organisational priority informed a bookability indication rather than an unsupported promise.

These cases establish product modelling and decision-interface work. They do not establish data-platform engineering, model performance or business outcomes.

Natural-language querying and generated explanation can lower the cost of exploring data. The product still needs to answer:

• Which metric or field did the system use?
• Which filters and time period were applied?
• Is the source current and authorised?
• What is observed, derived, forecast or inferred?
• Which alternative explanation remains plausible?
• Can the person inspect or modify the query?
• What action is being recommended, and against which objective?

For consequential decisions, evaluation should include query correctness, source coverage, explanation fidelity, action quality and recovery from missing or conflicting data. The article on generative AI product patterns separates retrieval, synthesis and recommendation so each can be evaluated on its own terms.

A named client-side business or domain owner approves each consequential metric definition, source and permitted interpretation. Tcules can make that contract usable and implement it, but does not invent institutional truth from ambiguous data.

When data is incomplete or late, the product distinguishes missing from zero, exposes freshness and coverage, and follows the client’s policy for suppressing or qualifying a decision.

Data engineering can be included when it is explicitly scoped and the relevant systems are accessible. The expertise claim does not imply that every engagement includes warehouse, pipeline, governance or source-remediation work.