Risk selection is the part of underwriting that decides which submissions a carrier accepts, on what terms, and which ones it declines. Done well, it also decides the order in which underwriters work their queue, so time goes to the risks that fit the carrier’s appetite and are likely to bind.
The three jobs inside risk selection
| Job | Question it answers | How UrbanStat does it |
|---|---|---|
| Augment | What do we know about this property? | Risk Analysis APIs add over 30 data sets for every address in the US, Canada, Europe and the UK. |
| Prioritize | Which submission should be worked first? | AI-based risk scores and “more likely to bind” scores replace first in, first out. |
| Select | Does this risk fit our appetite? | A scoring algorithm identifies submissions beyond the carrier’s appetite automatically. |
More than half of submissions are beyond a carrier’s appetite.
What data risk selection needs
Property risk is local. An underwriter needs to know the hazards at the address (for example tornado, hail, flood and wildfire), what surrounds the property, how much exposure the carrier already has nearby, and whether any underwriting guideline or moratorium applies. Typing in an address, a policy number or coordinates returns more than 300 data points in UrbanStat.
In the US, parcel and permit data adds property-level facts such as construction date, building value, other structures value and building permits, across all 50 states.
Risk selection on renewals
Selection does not end when a policy binds. Reviewing renewals with year-over-year portfolio context shows whether a policyholder is still a good fit.
Where machine learning fits
Machine learning is useful for scoring, but it is not sufficient on its own. Claims are rare (most insurers see a claim frequency of around 1 to 6%), outcomes take years to appear, and regulation limits which variables may be used. UrbanStat combines machine learning with traditional probabilistic modeling and human intuition, which it calls the Three Pillars of Risk Analysis, and never uses personally identifiable information or credit scores. The reasoning is set out in Is machine learning the silver bullet in underwriting?
Common questions
What is the difference between risk selection and pricing?
Risk selection decides whether to accept a risk and on what terms. Pricing decides how much premium to charge for it. A carrier needs both: a well-priced book can still lose money if the wrong risks are selected.
What does “appetite” mean in underwriting?
Appetite is the set of risks a carrier wants to write, defined by its underwriting guidelines: the property types, locations, hazards and values it will and will not accept.
What is a “more likely to bind” score?
A score that estimates how likely a submission is to become a bound policy. Used with a risk score, it helps underwriters prioritize work that is both acceptable and likely to convert.