The evidence behind each alert

We model soil moisture deficit (SMD) per property, using the FAO-56 water-balance approach, and compare it against a threshold set for that property's specific location. A single UK-wide number doesn't work. We backtested 51 years of weather (1975–2025) across 10 locations, and a flat threshold either false-fired on ordinary wet years in wetter regions, or missed real droughts in drier ones — deficit magnitude just varies too much by climate. So each location gets its own cutoff, drawn from its own historical percentile.

We checked those thresholds against known insurer-reported UK subsidence years (1976, 1995, 2003, 2006, 2010, 2011, 2018, 2022, 2025) against known UK-wide wet years (2007, 2012). But we didn't just trust that list. A second, independent source caught two problems in it: 2021 doesn't actually hold up as a UK-wide wet year (it was only true for the South East), and 1995 — the UK's driest summer on record — had been left off entirely.

UK home insurance treats damp and mould as wear-and-tear, not an insurable peril like subsidence. That means there's no claims-by-year dataset to backtest against, and there never will be. So validation here is simpler: does the trigger fire a sane number of times per damp season? Not constantly. Not never.

The threshold is anchored to BS 5250, the UK's own code of practice for condensation control. It documents that sustained indoor humidity above 70% pushes external wall surfaces into the range that supports mould growth. We read outdoor humidity as a proxy for that — there's no indoor sensor, this is a data-only product with no hardware to install. Our first guess, 85%, fired 5–8 times a year in an early pilot. That's routine weather, not a warning. At 90%, tested properly across all 10 locations, it settled into a genuine occasional-watch range: 1.9 times a year in Plymouth, 5.5 in Birmingham.

This one's simpler. 0°C is a physical fact, not something we tuned — the trigger fires when a hard frost is forecast at a property you've flagged as vacant, with no heating keeping the pipes above ambient temperature. What we can't calibrate in advance is whether a three-day-out forecast actually holds up. Checking that properly needs archived past forecast issuances, and MET Norway's API doesn't expose that history. So unlike the other three triggers, this one gets assessed by watching real alerts against real outcomes once it's live — not backtested first.

This one's different from the other three: the underlying scale isn't ours. It fires on the Environment Agency's own severity levels (1–3 of their 4-level scale) for active flood alerts and warnings, within 3km of the property. We spot-checked that 3km radius against five UK towns with well-documented flood history — York, Tewkesbury, Shrewsbury, Hebden Bridge, Cockermouth — and it caught every one of their real flood areas.

One honest caveat: a deliberately inland control town matched too. England's watercourse network is dense enough that "there's an EA flood area nearby" isn't much of a filter by itself — severity is doing the real work, not proximity. And this only covers England & Wales, the Environment Agency's own remit. A property in Scotland or Northern Ireland will never see a flood alert from us.