Insurance Analytics Platform

The problem

The operations team lived on ad-hoc data pulls. Every question — “how’s claims throughput this week?”, “where’s churn risk concentrated?” — turned into a manual SQL request and a one-off chart. Analysts were a bottleneck, decisions lagged the data, and nobody agreed on what the numbers even meant.

How I approached it

I treated this as a data product, where the real deliverable wasn’t a chart — it was a shared definition of truth.

• KPI framework first. I worked with stakeholders to define 8 operational KPIs — claims throughput, risk exposure, churn signal, and more — with explicit formulas. Agreeing on definitions was 80% of the value.
• Self-serve over bespoke. Instead of answering questions, I built a Power BI dashboard (DAX measures over a SQL model) that let the team answer their own — killing the request queue.
• Unstructured signal, too. I added an NLP sentiment pipeline (TF-IDF) across 500+ free-text records to surface operational risk signals that structured fields missed.

Outcome

• 40% reduction in ad-hoc reporting requests — analysts got their time back.
• 8 KPIs with agreed definitions became the team’s shared operating language.
• Real-time risk signals surfaced from text that previously went unread.

What I learned

The dashboard was the easy part. The durable win was forcing a definition for each metric — once “churn signal” meant one specific thing, every downstream conversation got faster. Data products succeed on alignment before visualization.