Counterparty paper has always been the hardest review. Here's the math behind why, and what changes when AI is actually grounded in your playbook.
Every attorney who has done outside counsel work knows the feeling. A counterparty sends back the MSA with tracked changes. You open it. The redlines start on page 3. There are 64 of them. Some are administrative. Some are nuclear. And you have no way to know which is which without reading the entire document, side by side, against what you'd accept. That's how three hours disappear from a Wednesday.
Reviewing a contract you drafted is hard but bounded. You know your standards, you know what you're looking for, and the structure is yours. Reviewing counterparty paper is a different problem. You're working inside someone else's framework, against terms you didn't write, with their definitions and their numbered sections and their order of provisions. Every redline is a question about your tolerance and an answer about theirs.
And the redlines aren't equal. A change to the notice address is administrative. A change to the indemnity cap is structural. A removed carve-out on consequential damages is a deal-killer. You have to read every one to know which is which.
Pick any solo attorney with a healthy client base. Conservative estimate: 15 counterparty papers a month. Conservative time per: 3 hours. That's 45 billable hours dedicated entirely to redline review, before drafting, before negotiating, before any client meeting that didn't involve a Word document.
For a small firm running multiple attorneys: multiply by headcount. The senior attorneys at the firm get the dense MSAs because the associates haven't built the pattern recognition yet, which means your highest-leverage hours are spent on the lowest-leverage activity.
None of this is wasted, exactly. The review is real work. But it's also the bottleneck. While you're reading line 47 of clause 9.2 you're not on a client call. You're not drafting an engagement letter. You're not doing the thinking that actually distinguishes your firm. You're parsing somebody else's structure.
The obvious answer is "use AI." And every attorney who tried the first wave of AI contract tools knows the obvious answer didn't work. The reasons are specific.
Hallucination. Ask a generic model whether a liability cap is "unusual" and it will tell you with confidence, citing standards that don't exist. You can't bring that to a client.
No playbook anchor. "Unusual" means nothing without "compared to what." Your firm's playbook says one thing about indemnity carve-outs. The model doesn't know what your playbook says. So it gives you generic answers to a question that only has firm-specific answers.
No tracked-change parsing. Most AI contract tools treat a redlined .docx as a clean document. They don't pull out each insertion and deletion as its own object. You still have to find every change yourself.
No anchored citations. When the model says "this clause is risky," it's referring to which clause exactly? In a 40-page MSA with sub-sub-numbered sections, "the clause about indemnity" isn't precise enough to act on.
The version that works treats every tracked change as a first-class object, not a sentence inside a document. Each insertion and deletion gets pulled out and analyzed individually. Each one gets:
The last point matters more than it sounds. If the model says "this is high risk," the next question is always: based on what? Grounded review means every insight points at the source clause it came from. You can verify in three seconds. The model isn't asking you to trust it. It's showing you its work.
Drop the redlined .docx into the platform. The next screen lists every tracked change as its own row. High-risk changes at the top. Each row tells you what the other side changed, what they took out, what they put in, and how that maps against your firm's position on that kind of clause.
For high-risk items: read the analysis, read the suggested counter-language, decide if you want to use it as-is or modify. One click to accept the recommendation. One click to mark as "needs revision." One click to flag for the client.
For low-risk items: skim and accept in batch. Notice address change. Date format. Defined-term capitalization. These exist in every counterparty paper and they don't deserve senior attorney time.
For the medium tier — and this is the largest bucket — read the analysis, decide if you're comfortable with the trade-off, push back where the playbook says push back, accept where the deviation is reasonable.
The math on the firm side is one thing. The math the client sees is different. To a client, the counterparty paper review is the wait. Their deal is sitting in your inbox. The longer the review takes, the longer their deal slides past the original timeline.
A 3-hour review becomes a meeting. A 15-minute review becomes "I'll have it back to you by end of day." That difference is the difference between the client renewing the engagement and the client looking for a different firm next quarter.
This isn't "AI replaces the attorney." Every recommendation still goes through you. Every counter-language suggestion is a starting point, not the final draft. The judgment that distinguishes a senior attorney from an associate is exactly what gets compressed into the 15 minutes — not eliminated.
What it means is the dumb part of the work — parsing tracked changes, classifying severity, finding the playbook precedent for each clause — stops being the bottleneck. The thinking is still yours. The hours stop being wasted on the work that doesn't need them.
This is also why grounding matters more than speed. A fast-but-hallucinating model is worse than a slow-but-careful one. The 15 minutes only counts if the analysis is right. Grounded means defensible. Defensible means the client can act on it.
Drop your next counterparty redline into Midly and see how the math actually changes for your practice.
Attorney-built · 14-day free trial · No card required