Client
An engineering and contracting company that bids regularly on public and private tenders. The client’s name is withheld for confidentiality.
Business challenge
Each tender pack runs to dozens of pages of requirements, deadlines and guarantees. A single missing or expired certificate is enough to lose an opportunity, and checking was done by senior staff reading documents line by line.
Existing process
Tender documents downloaded manually; requirements copied into spreadsheets; company documents kept in folders; expiry dates tracked from memory.
Implemented solution
A system that discovers relevant tenders, reads PDF, Word and scanned documents, extracts each requirement with a reference to the page and clause, and compares it with the company's stored certificates and records. Missing items and documents that expire before the deadline are flagged early, and bid-envelope checklists are produced. Nothing is submitted automatically; every approval is logged.
Technology
OCR for scanned documents, LLM-based extraction with retrieval and citations, a company document store, web application with role-based access.
Implementation approach
Demo on sample tenders → pilot on the client's real documents → deployment on a dedicated server with backups.
Impact on the work
Reviewers now check the items the system flags, each linked to its page and clause, instead of reading every page. Missing and soon-to-expire documents surface well before the deadline rather than at submission, and the bid team works from one checklist instead of scattered spreadsheets. Client-specific figures are not published.
Lessons learned
Citations to page and clause were what made reviewers trust the output; extraction accuracy alone was not enough. Expiry tracking delivered value before any AI feature did.