Claims Severity & Reserve Early-Warning · Wisdom notebook
Ask the follow-up.
Agentic conversational BI · one continuous thread reading the loss-runs for severity and reserves.
Before I open the triangles — what actually moved on severity and reserves across the book this month?
RKScored all 25+ programs on four early-warning dimensions — claim severity, large-loss ratio, reserve ratio and cat-claim mix — against each one's own baseline, and shaded the whole matrix.
One program stands apart, and it's a tail story, not an attritional one — Propane & Fuel Dealers on its Energy · PA/OH/TX slice:
Read across the top row: Propane lights up on all four dimensions at the shock — severity, large-loss, reserves and cat mix all move together, the signature of a fire/explosion event. Everything else is a book-level ~$26K severity and a ~40% reserve ratio.
Take Propane. Show me how the driver-slice severity has actually tracked — is it a trend or a shock?
RKPulled average claim severity for Propane · Energy × PA/OH/TX by month and laid its calm baseline band underneath.
A shock, not a trend — it sits at the book severity most months, then a fire/explosion month drives it clean off the top:
March and June already spiked to $78K and $59K; July is the biggest yet at $120K. Each is a discrete cluster of events — a monthly loss-ratio glance smooths them away, the severity view does not.
What's actually causing those shock months on the Propane slice?
RKBroke the shock-month claims on the Propane PA/OH/TX slice down by cause of loss.
It's the fire / explosion / pollution-spill trio — the man-made-catastrophe perils — that lift the average, not the routine delivery and property claims:
Fire, explosion and pollution-spill together are ~49% of the slice's claims — and they carry nearly all the severity. The attritional half (products, delivery, property) is routine; the tail half is what breaks the reserving.