When to Model
Not every question needs modeling. But repeated ambiguity is a signal that business meaning should move out of conversation text and into a shared semantic definition.
Model when repeated questions are drifting
Section titled “Model when repeated questions are drifting”You should model a metric or dimension when:
- different users ask for the same thing in different words
- the same metric keeps needing manual exclusions
- one term can map to multiple tables or fields
- the team keeps arguing about “what counts”
- a result is important enough to appear in repeated reviews
Do not model everything on day one
Section titled “Do not model everything on day one”Premature modeling creates overhead without improving trust. Start with the metrics that:
- drive recurring business decisions
- appear in weekly or monthly reviews
- need consistent interpretation across people
- are painful to restate each time
Good candidates for early modeling
Section titled “Good candidates for early modeling”- revenue
- active accounts
- conversion rate
- retention
- churn
- gross margin
- key funnel stages
Signs that prompting is still enough
Section titled “Signs that prompting is still enough”You may not need modeling yet if:
- the question is exploratory and unlikely to repeat
- the metric is obvious and low risk
- the answer is used by one person for one-off investigation
- the source schema already expresses the meaning clearly
The mental model
Section titled “The mental model”Use prompting to express a question.
Use semantic modeling to standardize meaning.
If the team is still negotiating meaning in every conversation, modeling is usually the next step.