Social housing group
Twelve words for one wall, and four months of warning.
A social landlord managing eighteen thousand homes, where one structural defect arrived under a dozen different descriptions and reset the ticket counter every single time.
18k
Homes covered
4 months
Of warning already in the corpus
12
Vocabularies for one defect
3+
Contacts becomes an escalation
Deployment model
Branded portal on the shared multi-tenant backend
Live since
2026
Languages
English · Español · Français · العربية · Português · Polski · Türkçe · বাংলা · Русский
Damp, mould, condensation, black spots, wet wall, smell, patches. Twelve vocabularies for one structural defect, and a keyword-driven repairs system that connected almost none of them.
Every fresh contact opened a fresh ticket. A household reporting the same problem for the fourth time looked identical, in the data, to four separate households reporting once.
Meaning clustering went in first, so every phrasing of a defect collapses into one tracked issue with all the original wording attached.
Repeat-contact journeys were switched on, so three or more messages from one anonymous device surface as an escalating case rather than three isolated tickets.
Safety categories were floored to critical on arrival and routed past the local estate office entirely.
The first meaning cluster the console produced held four months of reports about one block, in language no search had ever connected. Remediation was scheduled the week it surfaced.
Repeat-contact escalation moved the group from counting tickets to counting households, which changed how the whole repairs backlog was prioritised.
“The pattern had been in the corpus for four months. We had no system that could see across twelve different words for the same wall.”
Director of housing operations, Social housing group
Illustrative composite, attributed to a role. Full disclosure on the stories index.
Run the same engine
on your own corpus.
Seven days, your own tenant, your own taxonomy, card-free.