How to Automate In-House Contract Risk Scoring with Local AI

Reviewing contracts for risk is slow and expensive. Sending contracts to a public AI means your documents leave your premises. Running a risk scorer on a local model, on hardware you own, keeps everything in-house with no per-document fees.

What you need

• A machine running a local large language model (Hard Inf. installs and runs these for you).
• A way to extract text from contracts — PDFs, Word files, or scanned documents via OCR.
• A risk rubric: the clauses and terms that matter to your business.

How it works

1. Extract the text. Pull the contract text out of the document so the model can read it.
2. Split into clauses. Break the contract into sections (or keep it whole if it fits in the model’s context window).
3. Score each clause. Feed each section to the local model with a structured prompt asking it to flag risk items against your rubric — category, severity, the quoted text, and a short reason.
4. Roll up a score. Combine clause-level flags into an overall risk score (for example, 0–100), weighted by severity. High-scoring contracts go to legal review; low-scoring ones can be cleared automatically.

Example risk rubric

• Liability and indemnity — uncapped liability, broad indemnities.
• Termination and notice — short notice periods, no termination for convenience.
• Payment terms — late fees, upfront deposits, price escalation.
• Confidentiality and data — data use, subcontractors, cross-border transfers.
• Intellectual property — who owns deliverables and improvements.
• Auto-renewal — evergreen clauses with long notice windows.

Example scoring prompt

You are reviewing a contract clause for risk. Return JSON only, with these fields: category, severity (low, medium, high), quote (the exact text), and reason (one sentence). Score against this rubric: liability and indemnity, termination, payment, confidentiality, IP, auto-renewal. If the clause carries no risk, return an empty array.

Keep it private

Because the model runs on hardware you control, contract text never leaves your premises. There are no API calls to a third party, no per-document fees, and you control the rubric — so it stays aligned with your business, not a vendor’s terms.

Limitations

A local scorer is a triage tool, not legal advice. Models can miss nuance, so always have a human review contracts flagged as high risk before you sign.

Next steps

Hard Inf. can install and run a local contract risk scorer on your hardware, integrated with your document store. Get in touch to discuss your setup.