Generate business insights (KPIs, anomalies, trends) with causal explanations and narrative dashboards in conversational form.
Detects notable patterns in business data — movements, anomalies, trends — and narrates them. In practice two distinct things: statistical detection, and language generation explaining what was detected. Conflating them is the standard error.
Recurring reporting where the analyst's time goes into writing up findings rather than finding them.
Where causal attribution matters and the data cannot support it. Correlation narrated fluently reads exactly like causation.
Detection must be statistical and reproducible; explanation may be generated. The figures in the narrative must come from queries, not from the model.
Asserting cause from correlation. 'Revenue fell because of the pricing change' is a claim the data usually cannot support, and it is the kind of sentence these systems produce most readily.
A period-over-period movement analysis where thresholds and anomaly scoring surface the material changes, and generation drafts the explanation for a reviewer to correct and sign.