A reporting readiness checklist for client-facing automation

Fast is not the same as ready

Automation can assemble a report before anyone has checked whether every source arrived, whether the period aligns, or whether a platform changed a metric definition. The operational risk is not the time saved; it is the quality failure that reaches a client.

Put readiness checks before narrative generation and a named human approval before release.

Check the inputs first

Every report path should know which sources are expected. The workflow should wait, retry within a sensible limit, or visibly mark a missing source rather than silently substituting zero.

• All expected sources are present.
• Each source covers the requested period and timezone.
• Metric definitions and naming match the approved reporting guide.
• No source is stale beyond its agreed freshness threshold.
• Currency, attribution, or campaign identifiers are aligned.
• Material exceptions have an owner and a note.

Draft observations from evidence

AI can turn a validated table into a first-pass narrative, but it should not invent causes. Instruct the layer to separate measured facts from hypotheses and to cite the source fields used for each observation.

If a trend is ambiguous, the draft should say that the data is insufficient rather than confidently explaining why performance changed.

Require release approval

The account manager should see source status, any changed definitions, and the final narrative before the report is sent or published. Approval is an explicit state, with the reviewer and time recorded.

The goal is not to make every report manual again. It is to move human attention to the judgement that remains necessary: interpretation, relationship context, and release responsibility.