AI for reviewing vendor agreements

If you run procurement, vendor paper shows up on its own schedule: a SaaS subscription from marketing, a facilities contract, a supplier's own MSA they insist on using instead of yours, all needing a read before anybody signs. AI can do the first read of that pile and score it against your own standards. It cannot decide what you accept.

Where the time goes on inbound vendor paper

The setup takes longer than the reading: working out which entity is you, finding the governing law, deciding what counts as a problem, and then holding that standard steady across the next thirty contracts. When two people review the same supplier agreement and come back with different lists, the standard was never written down.

The WorldCC Foundation's contract design pattern library describes a "business terms first" layout, where the core commercial terms go up front and what it calls the passive terms come later. A lot of vendor agreements read that way: pricing and scope on page one, and the terms that decide who pays when something goes wrong in the back half, or in a schedule the vendor sent as a separate PDF.

What happens when you click review

You upload the vendor agreement and click Review with AI. It does not ask you anything. It reads the document and returns a setup card already filled in: a one-line scope of what it plans to check, both parties pulled from the document with your side preselected, the governing law read off the page, and a four-level review depth that defaults to Issues, described in the app as "Flag potentially critical issues that could be worth redlining only." Every setting is editable. A drop zone takes reference documents, so the price schedule, the SLA exhibit and the supplier's security addendum can go in as .pdf or .docx alongside the main agreement.

Pick a playbook, or leave it at "No playbook", and run it. A review takes a few minutes, and the app says so. Kick it off and come back.

What you get back

Findings sit in a rail on the left, with the contract in a Word-style editor on the right. Each finding is named for the problem it found: "Data privacy indemnity is uncapped", "Amendment process is one-sided", "Execution block drafting defect in SOW". They are grouped High, Med and Low, and each carries an ID like SD-4 so two people can talk about the same item.

Open a finding and you get the original clause with its section number and a jump to the text, the suggested redline in full, a plain English note on why the wording is a problem and which side it hurts, and the actions: Flag, Dismiss, Apply redline, Add to playbook, Mark as unacceptable. Nothing is applied to the document until you apply it. The suggested redline is drafting language. Whether you accept anything close to it is your call.

The review tracks findings decided as a count and a percentage, and the chat stays gated until every one has a decision: "9 review findings still need a decision. Accept or dismiss to keep chatting." You close every finding; goHeather only keeps count.

Your standard, applied to every supplier

Without a playbook the review is general and unopinionated about your own positions, and the app labels it "No playbook - general law". That is fine for a first look. It will not make review consistent across a category.

A goHeather playbook is a numbered list of rules, and each rule has more than a description. Under the advanced options you can add fallback positions ("we would accept 90 days extra notice of termination beyond our primary stance"), unacceptable deviations, and guidance notes such as a deal-size threshold where you are willing to negotiate. Because the walk-away is written down, a finding can be marked unacceptable on the spot.

The playbook changes what comes back. One MSA in a goHeather account was reviewed against five different playbooks and returned 17, 16, 8, 8 and 8 findings. Same document, and the rule set decided how long the list was. There is a longer post on how your playbook decides what an AI review flags.

You can generate a playbook from one of your own signed supplier agreements in about a minute; the app calls the result AI-drafted, you approve, and it needs a read and an edit before you trust it. Or go the other way: when a finding on a live contract reflects a call you will make every time, hit Add to playbook and it becomes a rule on the next vendor agreement.

Where it stops

It reads what it is given. If the vendor's MSA points to a rate card or a service guide you did not upload, that document is not in the review, though you can add it as a reference document and run again. It flags, compares and drafts language. Deciding what you accept and negotiating with the supplier stay with you and, where it matters, with your lawyer.

CIPS defines contract management as "the process of creating, implementing, and reviewing contracts." The reviewing part is where inbound supplier paper piles up, and it is the part a first-pass AI contract review can take on while the decisions stay with you.

This is legal information, not legal advice; consult a lawyer for legal advice.