Building a recurring report
A fixed pipeline can collect validated fields and use AI only for bounded narrative synthesis. Expected disposition: Proceed with controls.
Task boundary
Building a recurring report is assessed as a specific task and workflow, not as a general endorsement of AI for the surrounding job. A fixed pipeline can collect validated fields and use AI only for bounded narrative synthesis.
Assumptions
- The same approved systems provide a stable set of fields on each reporting cycle.
- Deterministic code computes totals; AI drafts only bounded narrative explanation.
- A report owner compares totals, anomalies, and claims before distribution.
Decision
- Disposition: Proceed with controls
- Pattern: Tool workflow
- Inherent risk: medium
Rules that drive the decision
- A stable sequence favors a fixed tool workflow rather than an adaptive agent.
- Calculations remain deterministic; generative output is limited to reviewable narrative.
Why alternatives were rejected
- A direct prompt cannot reliably gather inputs.
- An agent is unnecessary when the sequence is stable.
Controls
- Version the field mapping and fail closed on missing, duplicated, or stale inputs.
- Recompute totals independently and bind every narrative claim to a report value.
- Require approval before distribution and retain the source snapshot.
Acceptance threshold
These are example starting thresholds for a bounded pilot of this workflow, not universal benchmarks. For building a recurring report, the accountable owner should make them stricter when the task, consequence, or policy requires it.
- Totals exactly match the established reporting calculation for every evaluation case.
- The narrative contains no claim unsupported by the current source snapshot and flags every seeded anomaly.
Reassess when
- A source schema, calculation, audience, or reporting obligation changes.
- The workflow gains permission to distribute reports or alter source records.