Chapter 31 — Key Takeaways

Data-Driven Underwriting: How Information Technology Is Transforming Risk Assessment. A one-page field card. Everything here is illustrative; the discipline is not.

The core claim

Data and IT changed the inputs and the speed of underwriting — not the job. The submission now arrives pre-filled and scored before a human reads it; for simple, high-volume, well-described risks the machine binds it straight through. But the selection-and-pricing judgment is untouched: when information was scarce, judgment filled the gaps; now that it is abundant, judgment decides which of it to believe.

The five moves

1. Read the data picture as a first draft written by a machine — fast, useful, and to be checked. Corroborate the fields that drive the price.
2. Treat a pre-filled field as a hypothesis, not a fact. It arrives wearing the costume of verified data; its three failure modes are wrong match, stale data, false precision.
3. Read the score as one voice in the room. It arrived first — don't let arriving first make it the answer. Ask what it could not see (relationship, intent, a fix in flight, a novel exposure).
4. Let automation bind what needs no judgment; refer what does. A good referral rule is an honest statement of where the data and the model stop being trustworthy.
5. Distrust the confident number. Says who, and how do they know? Make missingness visible; never silently default a price-driving field.

Rules of thumb

• Corroboration beats any single source. Independent data agreeing with the file (as Harbor Steel's satellite roof read agrees with the loss runs, inspection, and broker note) is the gold standard; data disagreeing is a flag to chase, not noise to average.
• The most expensive data error is the silent default — a missing price-driving field filled with a guess, so the file looks complete and nothing flags.
• Automation's verdict is a combined-ratio verdict. A lower expense ratio in year one is not a win; the loss ratio lags two-to-three years. "We automated and cut costs" is only good news if the combined ratio fell.
• Garbage in, garbage out is worse in the data age — automation removes the human friction that caught errors, then acts on them faster and at scale.
• You outsource the data but keep the risk. A vendor's match rate and currency are now underwriting decisions; audit the feed, not just the decisions.

Alternative data sources at a glance

Key terms

• Data enrichment / pre-fill — auto-populating a submission's fields from third-party data; faster and often more honest, but it imports its sources' errors.
• Real-time risk scoring — an automatic risk score generated the moment a submission arrives; an input, not the decision.
• The underwriting workstation — the integrated environment that pulls, enriches, scores, surfaces guidelines, and records the decision.
• Alternative data sources — third-party data (imagery, IoT/telematics, public records, aggregators), often gathered for another purpose, drawn in to enrich a risk assessment.

What you could defend to your manager

"The pre-fill and the satellite roof read corroborate our manual assessment of Harbor Steel — the imagery, the loss runs, the inspection, and the broker's note all agree, which raises my confidence in the risk picture rather than changing the disposition. The real-time model recommends decline; I've logged it as an input, not acted on it, because it has never seen the signed roof contract, the hot-work controls, or the new plant manager — and a single two-loss count is not the story of those losses. I'd never let a large-limit, catastrophe-exposed, loss-flagged account like this bind straight through; it trips half our referral triggers by design. Chapter 32 opens the model and earns the override; this chapter confirms the read."