Contract management case study: finding clauses and payment discrepancies in large PDFs

The challenge: helping reviewers navigate long, complex documents

The client's teams review contracts and payment documents, and before the build they searched each document manually. Finding a clause, payment term or particular wording could mean working through hundreds of pages, including tables, scanned pages, embedded content and complex formatting.

The build needed to meet three requirements:

• Search that stays accurate. Highlighting and search results had to be just as accurate whatever the structure of the PDF.
• Large files in every browser. Big PDFs had to work smoothly in all browsers, though their formatting could slow loading and navigation.
• No lag as volume grows. Reviewers had to search, read and move around a document in real time, as the number of documents grew.

What we built: document review with search, a PDF viewer and AI-assisted checks

The platform brings upload, viewing, search and analysis into one review workflow. It connects to the client's document management system and comes with add-on modules, among them a contract engine and a payment engine, which look for discrepancies, anomalies and potential problems in the documents.

Four pieces do the work:

• A viewer built for large PDFs. Documents render in a viewer built on @react-pdf-viewer and pdfjs-dist, adapted to the client's review workflows. Front-end optimisations improved loading and navigation, and perceived waits on large documents fell by about 30 to 40%.
• Search with highlighting. A reviewer searches for a term and sees the matches highlighted in the document, so a clause, a payment detail or a contract term turns up without a read-through. The time to find a specific term or clause fell by as much as 50 to 60% against searching by hand.
• One state for the whole review. Zustand keeps the document data, the search state and the reviewer's actions in sync from start to finish.
• AI-assisted checks in the contract and payment engines. AI workflows, built with @ai-sdk/react, help analyse the documents and flag anomalies. Reviewers use them to uncover inconsistencies in contracts and discrepancies in payments before the documents go on to validation.

The interface is Next.js, styled with Tailwind CSS, Bootstrap and Sass.

Where the AI stops, and who takes over

The AI's job here is narrow. In the contract and payment engines it helps with the analysis and raises flags, and people decide what the flags mean. The reviewer checks what's flagged, and the documents then go on to validation.

The two measured gains on this page come from the search, the viewer and the front-end work. The AI checks have no figure of their own. What they changed: teams uncover inconsistencies and payment discrepancies before validation, which makes that step simpler.

Results

Review now follows a consistent workflow: the documents, search and checks are in one place, helping reviewers reach well-supported decisions on contracts and payment records sooner.

Technologies and tools

Timeline

The build took three months.

If your team checks contracts against its own positions, our page on AI contract review sets out how we'd build that review on your playbook. When contracts arrive alongside invoices, claims and forms, our intelligent document processing solutions cover the reading and the checks.