VLM-driven rateable value discovery
£2–4m / year in missed business rates, surfaced by a vision-language model over Google Street View.
UK business rates are set from a property list the Valuation Office Agency maintains. Properties change use, get subdivided, or quietly appear, and the list drifts. I prototyped a way to close that gap using Google Street View and a vision-language model, then turned it into a product the council now plans to commercialise to other UK authorities.
Solution Architect / IoT Tech Lead — owned end-to-end from self-initiated prototype to shipped product.
Councils lose material annual revenue when the rating list is stale. Manual street surveys do not scale and the VOA has limited bandwidth. Nobody inside the organisation was treating this as an AI problem.
- 01 Self-initiated a prototype: VLM over Google Street View panoramas, cross-referenced with the live rating list, to flag properties whose visible use did not match their valuation.
- 02 Pitched to leadership, built the business case, ran stakeholder discovery with the Valuation Office Agency.
- 03 Designed the solution flow in Miro, planned the backlog in Jira, and developed product features using Claude as a coding partner.
- 04 Shaped a prototype-driven development process: iterate with service teams on early designs before locking architecture.
- ▹ Surfaced £2–4m of additional annual rateable value.
- ▹ Now being commercialised to other UK councils.
- ▹ Became the internal template for scoping AI-driven revenue initiatives.