The situation
Public tenders arrive as 150 to 300 pages of PDF with the decisive constraints scattered through the annexes. A senior architect spent two to three days per tender deciding whether to bid at all — which meant the firm passed on tenders it might have won, simply because the reading capacity was already booked.
What we built
We built an extraction pipeline that pulls deadlines, eligibility criteria, required certificates, penalty clauses and evaluation weightings into a structured checklist, each field linked to the page it came from. A second pass compares the terms against the firm's own red-line playbook and flags what a bid manager needs to read personally.
How it integrates
Tenders are dropped into the existing document management folder and the structured result appears in the internal bid database. Draft response sections are generated in the firm's own Word templates, so the bid team edits where they always have.
Governance & compliance
No decision is automated: the pipeline produces a recommendation and a reviewer decides. Extraction accuracy is measured per field against a labelled set of past tenders, and any field below threshold is shown as unverified rather than silently filled in.
Result
The bid/no-bid decision now happens in an afternoon instead of across three days, so the firm evaluates more tenders per quarter without adding people. The page-level citations turned out to matter more than the extraction itself — reviewers trust a field they can verify in two clicks.