The Hazards a SIC Code Can't Tell an Underwriter

Companies House gives an underwriter a SIC code and little else, and a SIC code describes what a business is registered as, not the hazards it runs. Veridion scored 200 UK SMEs on seven layers of observable evidence, every hazard traced to a verbatim quote on a public page, and surfaced the operations the registry can’t see.

1. 1
   Went past the SIC code

   Companies House tells you the registered activity, not the hazards a business actually runs

2. 2
   Stacked seven evidence layers

   hazard flags, cert checks, exposure, building intelligence, satellite + Street View imagery, live registries

3. 3
   Sourced every hazard flag

   all 1,449 anchored to a verbatim quote on a public URL, substring-validated

4. 4
   Verified certs against third parties

   claimed vs registry-confirmed, the company's own site rejected as proof

5. 5
   Scored every business 0–100

   a deterministic 10-component score with a plain-English underwriter rationale per row

6. 6
   Surfaced the hidden hazards

   nearly 1 in 7 ran hazards their SIC code hides, now sorted to the top of the book

A SIC code names the label, not the business

The insurer underwrites UK SME commercial risk, and underwriting turns on knowing what a business actually does: the hazards it runs, the premises it occupies, the certifications it holds. The registry carries almost none of it.

Companies House gives an underwriter a SIC code and little beyond it, and a SIC code describes what a business is classified as, not the hazards it runs day to day. A company filed as a road haulier may also operate a workshop, a training centre and warehousing, operations its classification never mentions and that materially change its risk.

Across a 200-business pool, nearly one in seven carried hazardous operations their SIC code did not predict. An underwriter working from registry data reads the label, not the business. The brief was a per-business hazard footprint, the operational risk picture the registry cannot provide, with every claim traceable to public evidence.

Seven evidence layers, every flag traced to source

Veridion built the hazard footprint from public evidence under one rule: no flag without a verbatim quote, and no quote without a source URL. Each business is described by seven independently auditable evidence layers.

Operational hazard flags come first, 1,449 across the pool, each an excerpt on a public URL against a closed 58-ID taxonomy; then sub-segment qualifiers; a two-layer certification check that separates a claimed certification from one confirmed in a third-party registry, with the company’s own site rejected as proof; quantitative exposure hints such as fleet sizes, yard acreage and sites operated; building intelligence from public records, 519 building events drawn from 1,564 third-party sources; a structured vision read of every headquarters from satellite and Street View imagery; and live signals from five registries, including Companies House, Environment Agency flood data, HSE notices, FSA hygiene ratings and VAT.

Those layers roll into a deterministic 0–100 score, ranging 2 to 99 with a median of 41, and every row carries a one-to-two-sentence senior-underwriter rationale in plain English. Provenance is the discipline that holds it together: every excerpt is substring-validated against the live page, so a hallucinated quote is dropped before it reaches the workbook, and a reviewer reaches any claim in five clicks, from the score to the rationale to the flag to the quote to the source.

Underwrite what a business actually does

The insurer can underwrite on what an SME actually does rather than what it is registered as, and sort the book by risk so the highest-hazard accounts rise to the top of the file where the underwriter belongs.

Every hazard is evidence an underwriter can open and read, so the score is auditable rather than asserted, and the off-SIC operations the registry hides, the workshop behind the haulier, surface exactly where they change the risk.

Because the footprint is read from public evidence and the live company graph, it refreshes as the evidence does and runs across the whole SME book. The next step is calibrating it against the insurer’s own claims data, the route from triage to a loss-cost score.