The question nobody can answer quickly
Ask a mid-sized business which contracts auto-renew in the next ninety days and the answer usually involves someone opening files. The information exists; it is in PDFs in a shared drive, in language that varies by counterparty.
The cost of not knowing is concrete: renewals that pass unnoticed, notice periods missed, obligations nobody tracked, and no view of aggregate exposure across agreements.
Extraction and interpretation are different jobs
| Task | Feasibility | Human involvement |
|---|---|---|
| Find the renewal date | High | Spot-check |
| Identify the notice period | High | Spot-check |
| Find the liability cap value | Good | Review flagged cases |
| Classify the governing law | High | Spot-check |
| Decide whether a clause is acceptable | Poor without legal input | Lawyer decides |
| Assess overall risk of an agreement | Not a machine task | Lawyer decides |
Keeping the project on the top half of that table is what makes it deliverable. The value is in finding and structuring, not in judging.
The document problems come first
Contracts arrive as scanned PDFs, signed copies with handwriting, documents with amendments attached separately, and occasionally photographs of pages.
- Scanned documents need OCR, and quality varies enormously
- Amendments and side letters change terms in the main agreement
- The operative version is not always the newest file
- Schedules and annexes carry terms the main body references
- Multi-language agreements in international groups
The amendments issue is the one that causes real errors. A system confidently reporting a renewal date superseded by a variation letter is worse than no system, so amendment handling belongs in the design rather than as a later refinement.
Confidence and review
Every extracted field should carry a confidence level and a link to where in the document it came from. That link is what makes review fast - a reviewer can confirm in seconds rather than reading the contract.
Route low-confidence extractions to review rather than accepting them. For fields driving a deadline, we would suggest reviewing everything initially and relaxing only once measured accuracy justifies it.
Where the return actually comes from
The obvious benefit is time saved. The larger one is usually the questions that become answerable: total exposure across agreements, how many contracts contain a particular obligation, which counterparties have unusual terms.
Those were previously unanswerable at any reasonable cost, so nobody asked them. Making them cheap changes how the business manages its commitments, which is worth more than the hours saved on any single renewal.
The contract you forgot renews on time, whatever your intentions were.