Quantifying flood exposure across an insurance portfolio

The question behind the work

An insurer arrives with a property portfolio — anything from a few thousand single-family policies to several hundred thousand commercial and residential lines — and asks the obvious question: what is our flood exposure?

Underneath that question are three sub-questions the work needs to answer, in this order:

1. Where, geographically, is the exposure concentrated?
2. How much annualised loss should we be carrying for this book?
3. Where should we adjust pricing, capacity, or reinsurance treaty terms?

A defensible portfolio exposure quantification answers all three with traceable inputs. A black-box score answers none.

What we actually do

The mechanics are unglamorous and that is the point. There is no clever AI step that magicks the answer.

Geocode. Property addresses are matched to coordinates and, where data permits, to building footprints rather than just centroids. We flag rows that fail to geocode and rows that geocode with low confidence. Both are reported back so your team can decide whether to manually verify or accept the broader uncertainty.

Hazard attachment. Each geocoded property is intersected with the available flood hazard layers. The depth at each return period is read at the property location. Where coverage is in beta or in our roadmap rather than production, the result is annotated with that fact — we do not silently fill gaps with zeros or fabricated regional defaults.

Loss modelling. Depth at return period drives damage using depth-damage curves. The curves are agreed up front: industry-standard, the insurer’s in-house curve, or a transparent blend. Expected annual loss (AAL) is computed by integrating loss across the return-period set.

Aggregation. Results roll up to whatever grouping is useful — policy, postcode, municipality, reinsurance treaty, custom. The aggregation is just a sum; the value is in choosing the right groupings.

Sensitivity. We run the same portfolio through alternative hydrology assumptions or alternative depth-damage curves and report how the headline numbers change. Sensitivity that does not appear in the deliverable is sensitivity that does not protect the reinsurance review.

What you get

A report and accompanying data files. Not a dashboard, not a portal, not an API. The report walks through methodology, assumptions, and headline numbers with sub-portfolio breakdowns. The data files (CSV, Parquet, and GIS extracts where relevant) let your analytics team plug the per-property results into whatever they already use.

A consolidated subscription platform is on our roadmap — the Insurance Risk Platform page covers what is and isn’t planned. Today everything ships as a per-portfolio engagement.

The honest part

Two things will surprise an insurer running this exercise for the first time:

Coverage gaps are real. Even in a well-modelled metro, the rural fringes of a portfolio often sit outside the hazard layer. A defensible engagement reports those properties as “no layer” — not as zero — and flags the slice in the executive summary. Hiding the gap behind a small default is a near-irreversible mistake the first time the reinsurer asks how it was handled.

The biggest risk concentration is rarely where the gut says. Property-level peril and accumulation often diverge from the underwriter’s mental map of the portfolio. We have seen books where the headline accumulation comes from suburbs the team did not consider high-risk, simply because of how many policies sit there. The quantification reveals these patterns; the dashboard prints them.

Next step

If you have a portfolio you want quantified, book a discovery call. We will agree the geographies, return periods, and depth-damage assumptions up front, scope the engagement, and tell you exactly which parts of your book we can cover today.

For an overview of how this fits with our other insurance offerings, see the Insurance sector hub.