The challenge
Every day, the firm’s team reviewed more than 50 sales and purchase contracts by hand. Each one took two to four hours, as staff read through the document and keyed its data into the firm’s systems. That added up to between 100 and 200 hours of skilled work every day.
The data was also hard to trust. About a quarter of everything entered could not be traced back to the contract itself. Some of that gap (10%) came from human error. The larger share (15%) came from emails, phone calls, and other documents that never made it into the deal record.
What we built
Progentic AI designed an extraction service that reads each contract in stages and checks its own work across several passes before handing data on. It reached 95% accuracy across all documents.
Using it is simple. A user answers four questions and drops in the documents. In under 3 minutes, the data is ready.
We also widened the scope to bring in the emails, call notes, and other supporting documents that had been sitting outside the deal record. The AI now does the first round of extraction, and the firm’s staff review and verify its output.
The firm’s own staff can add new document templates as new contract types and needs come up, with no developer involved. Because it runs as a shared service in the background, it connects to the firm’s existing systems, and new systems can plug into it as they come online.
The results
Work that took two to four hours per contract now takes under 15 minutes, including human review. Data that could not be traced back to the contract dropped from 25% to less than 5%, giving the firm a much clearer audit trail.
The service launched with two use cases in a single region. It now supports a dozen use cases across every region the firm operates in, and the relationship has grown to three separate engagements with Progentic AI.