LLM commentary for executive reporting
A production reporting workflow that generates first-pass explanations of monthly KPI movements while analysts retain responsibility for validation.
Outcome Integrated into monthly production reporting, reducing first-pass manual discovery.
- AI/LLM
- Analytics Engineering
- Production Reporting
- Snowflake
- dbt
- SQL
- Power BI
- LLM commentary
Context
Monthly KPI reports showed leadership where figures had moved, but explanations required reporting teams to investigate the underlying data separately. The opportunity was to bring an initial explanation into the reporting workflow so analysts could spend more time validating findings and investigating genuine anomalies.
My role and ownership
I took the commentary workflow from proof of concept into production at RMIT. I connected the reporting data to an LLM workflow and integrated its explanations into monthly reporting, working with the analytics environment and its existing Snowflake, dbt, SQL and Power BI stack.
Technical approach
- Connected reporting data to the commentary generation workflow.
- Generated explanations of internal drivers behind KPI movements, including projects and funders.
- Made commentary available at report generation time rather than as a separate first-pass research task.
- Kept analysts responsible for validating the output and investigating the movements that required deeper analysis.
Tradeoffs
The useful boundary was between a first-pass explanation and a validated conclusion. Automatically generated commentary could reduce repeated discovery work, but it did not remove the need for analyst judgment. Keeping review in the workflow made the output useful as a starting point for investigation.
Outcome
The workflow became part of monthly production reporting. It standardised initial explanations and reduced first-pass manual investigation while analysts continued to validate commentary and investigate genuine anomalies.
Reflection
This project showed how AI can improve an existing reporting process through a focused addition. The valuable change was making context available alongside the numbers, with a clear place for human review.