AI for first-pass review of master service agreements

A first-pass AI review of a supplier MSA gets you one thing: every term that needs a decision, surfaced and ranked, before you have read page one. It does not tell you what to accept. Your playbook does that, and if you have not built one yet, the review you get is a general one.

Why the MSA is the hard one

LexisNexis defines a master service agreement as "a framework contract setting out the core legal and commercial terms". The statements of work and order forms that supplier sends you afterward are called off under it, so a bad term in the MSA follows you into every one of them.

And the terms that get argued over are not always the ones that hurt later. WorldCC, back when it was IACCM, wrote in 2018 that there is often a disconnect between the terms that are most negotiated and the ones negotiators think are actually the most important. Its list of the ten most negotiated terms starts with limitation of liability and indemnification. On a first pass you want the whole document surfaced, not just the two clauses at the top of that list.

Upload it and click Review with AI

Drop the MSA into goHeather and it lands in the contract list with the start date, end date and notice to terminate date pulled out on its own. Open it and click Review with AI. You are asked nothing.

What comes back is a setup card goHeather has already filled in from the document. It names both parties and preselects your side. It reads the governing law off the contract. It writes a one-line scope for the review and sets a review depth, a four-level setting running Existential, Issues, Warnings, Everything, with Issues as the default. Every field is editable, so if it picked the wrong side, change it.

There is also a drop zone for reference documents. If the MSA points to a rate schedule or a service guide, feed those in here. goHeather reads what it is given and nothing else, so a document the MSA names but you did not upload is not in the review.

Then click Run full review and go do something else. It takes a few minutes, the app says so, and the work runs server-side, so you can close the tab.

What a first pass on an MSA looks like

The review opens as two panels: findings on the left, the full contract in a Word-style editor on the right. Findings are grouped High, Med and Low, each with an ID like SD-1 that you can quote back to the supplier.

Each finding is named as the problem. "Data privacy indemnity is uncapped" tells you more than a section number would. Open one and you get the original clause with its section reference and a jump to the text, a suggested redline written out in full, a plain English note on why the wording is a problem and which side it hurts, and the buttons: Flag, Add to playbook, Dismiss, Apply redline, Mark as unacceptable.

This is where the software stops and you start. Nothing is applied to the contract until you click Apply redline. The suggested redline is drafting language, not a position on what your company should accept. And goHeather keeps count. "9 findings left undecided" sits at the top until you have accepted or dismissed every one, and the chat will not move on until you do.

Without a playbook it is a general review

Run an MSA with the default No playbook setting and you get a general law review. It will not know that your company walks away from any supplier who refuses a 60-day termination for convenience, because nobody told it.

A playbook is where you tell it. Each rule has a name, a primary position in plain language, and under advanced options a fallback position, an unacceptable deviation and guidance notes, which is where deal-size conditions go. The quickest way to build one for supplier MSAs is to generate it from your own standard MSA. goHeather drafts the rules and you read and edit them before saving. Read that first draft properly before you trust it; the app itself calls it AI-drafted, you approve.

The playbook changes what counts as a finding. One MSA in our account reviewed against five different playbooks returned 17, 16, 8, 8 and 8 findings. The standard decided the result, not the model.

Where the first pass ends

The review also gives you a Legal Summary and a Deal Abstract, both downloadable as DOCX. The Deal Abstract is the one you will probably forward to the category manager: parties, term and renewal, fees, payment terms, termination and notice requirements, as bullets. If the contract is edited after the review, both carry a banner saying the output is stale and to re-run.

What goHeather will not do is decide. It reads, flags, drafts language and keeps score of what you decided. The minutes the review takes are the small part of turnaround, and this post on how long AI contract review takes explains where the rest of the time goes. The AI contract review workflow page shows the full product.

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