500 submissions scored · prepared by Plumbline Risk · August 4, 2026 · decision-support (assist) output
0.698 AUC (score vs. loss) · Gini 0.40 | 14.8% Book loss frequency | 23% of losses sit in the RED band (13% of book) | 67% RED-band loss ratio vs. 62% book average |
| Band | Risks | % of book | Loss rate | % of all losses | Loss ratio | Loss rate, to scale |
|---|---|---|---|---|---|---|
| GREEN | 235 | 47% | 7% | 23% | 35% | |
| AMBER | 199 | 40% | 20% | 54% | 87% | |
| RED | 66 | 13% | 26% | 23% | 67% |
The worst loss ratio on this book is AMBER, not RED. AMBER runs at 87% against a 62% book average and carries 54% of all losses, because it is 40% of the book. RED is a smaller, sharper group. On your own data that ordering may differ, and it is worth checking first: the band carrying the most loss is where a rate or deductible change earns the most, and it is not always the one at the top.
The RED band is 13% of the book and carries 23% of the losses, running at a 67% loss ratio against a book average of 62%.
It also carries 16% of the premium ($3,232,295). That is the part a headline number usually hides: removing the band removes its premium too, so blanket declinature is rarely the right answer and rarely available in practice.
A more useful read: bringing the RED band alone to a 65% target loss ratio implies roughly a 3% rate increase on those risks, achievable through pricing, deductibles, sub-limits on the flagged peril, or selective declinature. The value of the score is knowing which risks warrant that attention before binding, not declining a fifth of your submissions.
For reference, the theoretical upper bound, removing the RED band entirely, takes the book from 62% to 61%. Treat that as a ceiling on the opportunity, not a recommendation.
| Decile | Risks | Loss rate | Lift vs. book | Loss rate, to scale |
|---|---|---|---|---|
| 1 | 50 | 32% | 2.2x | |
| 2 | 50 | 22% | 1.5x | |
| 3 | 50 | 28% | 1.9x | |
| 4 | 50 | 22% | 1.5x | |
| 5 | 50 | 6% | 0.4x | |
| 6 | 50 | 10% | 0.7x | |
| 7 | 50 | 12% | 0.8x | |
| 8 | 50 | 6% | 0.4x | |
| 9 | 50 | 6% | 0.4x | |
| 10 | 50 | 4% | 0.3x |
Risks carrying at least one high-severity flag had a 20% loss rate vs. 14% for unflagged risks (17% of the book was flagged).
| Flag | Risks | % of book | Loss rate | Lift vs. book |
|---|---|---|---|---|
| CAT_WILDFIRE | 44 | 9% | 25% | 1.7x |
| AGED_ROOF | 156 | 31% | 21% | 1.4x |
| FIRE_LOAD_MISMATCH | 31 | 6% | 19% | 1.3x |
| CAT_EARTHQUAKE | 16 | 3% | 6% | 0.4x |
Lift above 1.0x means the flag predicts loss frequency, those risks lost more often than the book average. Flags sitting below 1.0x should be retuned on your loss data.
43% straight through, 7% loss rate against 15% book | 4% held on catastrophe exposure despite a clean frequency score |
The 18 held risks have a loss frequency of 11%, below the 15% book average. That is the point of holding them rather than a failure of the score: they are not risks that lose often, they are risks that would lose badly and together. A frequency ranker cannot see that, which is exactly why it should not be allowed to auto-quote them.
0 risks below 65% of modelled rebuild cost | 0% of the book | $0.0M total insured-value gap |
This is a gap in insured VALUES, not in premium. Premium at stake is this figure multiplied by your rate, so it is far smaller, and we do not know your rate. Our rebuild estimate is modelled from a footprint proxy and covers the building only, while a declared TIV often includes contents and business interruption, so treat each one as a prompt to check the statement of values rather than as a finding.
Illustrative run on synthetic data using published hazard layers where available and offline baselines elsewhere, as set out at the top. Property-level enrichment from vendors such as Opta or iClarify and Ecopia is not licensed today, and nothing here depends on it. Weights are unfitted priors until a partner's own loss experience replaces them. Output is decision-support (assist), not a bind decision.