Methodology
How the OPAL Score is built.
Twelve questions today. Transparent weights. SME-realistic SGD exposures. Public Zurich Singapore product lines. Address plus official PUB/OneMap links are on the product. Satellites, OCR, and bank login are not.
The door vs the next signals
The live product is a two-minute 12-question door, plus extra signals when you add them. Address can sit on the score. We do not rasterize flood maps or fly satellites.
Bank/Plaid login, Stripe SGD 99 checkout (still off for the free pilot), and Henry on-camera demo film are still off. OCR for scans is on the wording PDF path. The insurer desk lists OPAL-scored SMEs — not Zurich’s book.
Five pillars
Each category is scored 1–10 from your 12 intake signals, then blended with deterministic baselines. Weighted aggregate → OPAL Score 0–100 vs industry benchmark.
- cyber30%
- liability25%
- property20%
- employer15%
- business interruption10%
Exposure banding
LLM-suggested SGD exposures are capped by revenue tier (e.g. under_250k cyber ≤ SGD 80k) and scaled by headcount for employer and business interruption. We never invent Fortune-500 loss figures to look impressive.
Model pipeline
The model returns structured JSON; OPAL normalizes scores and exposures; a Zurich public-line matcher attaches directional product families. Confidence starts as intake_only. It becomes enriched only when extra signals are actually on the run — rescoring the same 12 answers does not fake enrichment.