Automated Executive Commentary
Built Claude workflows that read structured query outputs and generate insight summaries in the voice of a senior analyst. Now embedded in the weekly reporting pack for multiple enterprise clients — not a gimmick, a genuinely better product than hand-written commentary.
Outcomes
30–60 min writing → 8 seconds
Live in multiple client packs
Analyst time shifted to strategy
The Problem
Analytics teams were spending 30–60 minutes per report writing narrative commentary — paraphrasing numbers the dashboard already showed. The writing was often inconsistent and rarely added genuine insight. Senior analysts were doing work that shouldn't require a senior analyst.
The Approach
Designed a structured prompt architecture where the SQL model output is passed as context. Claude is instructed to write in a specific analyst voice — with instructions to focus on what changed, why it matters, and what action it implies.
Added guardrails so the model only comments on statistically significant changes and flags anomalies rather than narrating every data point.
The commentary generation is automated as part of the weekly data refresh pipeline — it runs alongside the SQL, not after it.
Every output is reviewed before it goes to the client. The value isn't removing human judgement — it's compressing the time between having the data and having something worth saying about it.
The Result
Commentary generation takes ~8 seconds. Multiple enterprise clients now receive this as the default format for their weekly exec packs. Senior analysts spend their time on the thinking, not the formatting. The key insight: AI commentary is only as good as the data model underneath it. Clean, tested, well-documented models are what make the AI output trustworthy.
Full Stack
Interested in something similar?
Let's talk →