Brightwater Property
Real estate
2024
10 weeks
SharePoint / Azure OCR / Claude / Yardi

Fields extracted and checked per lease
To abstract the whole portfolio
In missed rent reviews recovered
[CASE] Field notes
Written with the client team. Numbers are measured against the eval set agreed in the audit.
The problem
Brightwater manages offices and retail units across the north of England. Its leases were scanned PDFs, some typed, some handwritten in the margins. Nobody could say with confidence when each rent review was due.
What we built
An extraction agent reads each lease and fills 22 fields: parties, term, break dates, rent review dates and basis, service charge caps and more. A second pass checks every value against the page it came from. Low-confidence fields go to a reviewer with the page open beside them.
How it runs
The output lands in the property system with a link back to the source page for every field. Reviewers checked 14% of fields by hand; the rest passed both passes.
Results
The whole portfolio was abstracted in three weeks. The first report found 37 rent reviews that had never been triggered, worth $410,000 a year.

[Q] From the client
Marcus Oyelaran, Asset Manager, Brightwater Property
Next case