Every quarter a new vendor puts a chat box where your contracts already live. The pitch is always the same. Summarize this, flag that, draft me an NDA, compare these two deals side by side. The box is convenient and the box is dangerous in the same way, because every keystroke in that box is teaching the vendor something about how your business actually works. Before you sign anything, you should know what the box is going to do with the next two years of your contracts.
I am not writing this to talk you out of buying the tooling. Most teams genuinely need help reading page forty of an order form, and the cost of missing an auto-renew clause (a contract clause that silently extends your subscription unless you cancel by a specific date) is real. The point of this article is to slow down the thirty-second “yes” and replace it with a one-page checklist that protects you when the renewal shows up.
What the legal-AI pitch is actually selling you
If you strip away the legal-specific framing, the pitch collapses to three claims. You will get a better answer to a question you already ask. You will not need to teach your staff a new tool. Your data will not leave the room it already sits in. Each of those claims is partly true, and each conceals a downstream lock-in (the cost of switching to a different vendor later, which tends to grow the longer you stay) that nobody puts on the demo screen.
The first claim is easy to verify. Take the twenty contracts you already know well, run them through the assistant, and grade the output against what your senior reviewer would have written. If the assistant is faster and right ninety percent of the time, the productivity claim holds. If the assistant is faster and right sixty percent of the time, you are creating review work that you used to skip.
No new login means no new identity layer to manage, no new vendor to onboard, no procurement ticket. That part is real. The hidden cost is that the assistant is now sitting next to your inbox, your shared drive, and your existing permissions model, which means it inherits every access control problem your Drive folder already has. A misclassified contract that a junior lawyer would have caught by hand is now an output from a confident-sounding model that nobody double-checks.
Data residency is the claim that ages worst. Your data does not leave the room today. Vendor policies change, vendor leadership changes, vendor acquisitions happen, and the lawyers who negotiated your current contract will not be the ones negotiating the renewal in 2028. Read the data-residency (where your data physically lives, and which country’s laws apply to it) and exit clauses (the terms that govern what happens to your data if you leave) before you sign the first contract, not after the third.
Why every vendor is fighting over the same wedge
Walk through the last six months of enterprise AI announcements and the pattern is obvious. Google shipped a legal-tuned assistant. Microsoft shipped one. OpenAI shipped one. Anthropic shipped one. Every vendor picked legal as the first vertical (an industry or function that the product is tuned for) for the same three reasons, and once you see the reasons, you stop being surprised when the next announcement lands in finance, sales, or HR.
Legal teams give vendors three things almost no other department will give. They will run a sixty-day pilot (a small-scale trial of the tool on real work, usually free or discounted) without complaining. They will fill out the survey. They will write the procurement memo. They will tolerate a worse UI for a year if the underlying answer is better.
Beyond patience, legal teams produce documents every other team reads. The contracts legal writes shape sales, procurement, HR, and finance for the rest of the fiscal year. Win the legal team and you have a wedge into the workflows that depend on the contracts, plus a sponsor who will defend your tool in front of the C-suite.
Finally, legal teams are trusted to set policy on AI use inside the company. When the vendor gets pushed back on by an employee, the vendor can point to the legal team’s documented approval and the conversation ends. Few other teams carry that kind of authority.
- Patience, document volume, and trust are the three things every vendor is buying
- Winning legal means expansion into sales, HR, finance, and procurement is mostly a permissions change
- A vendor that owns your contracts folder owns the definition of risk for your whole business
- The wedge is the same wedge whether the vendor pitches legal, finance, or HR
The pre-sign checklist that survives the renewal
Most enterprise AI pilots fail at the same step. The team gets excited about the demo, runs a thirty-day pilot on ten contracts, sees a productivity win, and signs a three-year deal. Eighteen months later, the vendor has learned the shape of your business and the cost of switching is whatever you would have to pay to retrain every lawyer who got used to the prompts. The checklist below is what I would run before that signature, in this order, with the answers written down in the same document.
First, where does the data physically live, and who at the vendor can read it. Cloud region is not enough. Subprocessor list (the other companies the vendor uses to actually run the service) is not enough. The question is which humans, under what circumstances, with what logging, can touch your contracts. If the answer is “none, under any circumstances, with full audit logging,” that is one answer. If the answer is “our support team can read it to debug your tickets,” that is a different answer, and your renewal terms need to say so.
Second, what happens to your documents if you leave. Most enterprise AI contracts bury the data-return clause in a section labeled “Termination.” You want a clause that says you get the original documents back in original format, you get a copy of the assistant’s learned preferences back in machine-readable format, and the vendor deletes its copy within thirty days, verified by an audit log entry you can request. If the vendor will not put that in writing, the trust claim is a sales pitch, not a contract term.
Third, who in your org owns the vendor relationship when the contract renews. Not who owns it on day one, who owns it on renewal day. If the answer is “the same person who negotiated it,” write that person’s name in the contract next to the renewal date. If the answer is “we will figure it out then,” assume the renewal will be auto-extended and your exit window will close before anyone reads the email.
Fourth, what does the assistant cost at three times your current contract size. Enterprise AI pricing is usually quoted at your current seat count. The per-document or per-query price at three times the volume is a separate conversation, and that is the conversation that determines whether the tool is still a win two years from now. Get the scaled price in writing before you sign the original deal.
How to run a pilot that teaches you something
The default pilot is a procurement checkbox. The team demos the tool, runs it for thirty days on real work, writes a paragraph saying it saved time, and signs the contract. That pilot teaches you almost nothing. The pilot that actually changes your decision-making looks different, and it is not harder to run, just less convenient.
Pick one legal workflow that already takes too much time. Contract summarization, NDA review, vendor clause comparison, renewal flagging. Pick a workflow your team does at least twenty times a month, because below that threshold you cannot measure improvement reliably. Pick a workflow whose output quality is easy to grade, because “feels faster” is not a metric.
Run the workflow on the same twenty documents through two vendors for sixty days. Yes, two vendors. Yes, sixty days. The cost of running two pilots is roughly the cost of one procurement cycle, and the comparison output is what protects you when vendor A tries to lock you in at renewal by telling you the migration cost is too high. If vendor B produced equivalent output at sixty percent of the cost, you have a real negotiation lever. If vendor A produced meaningfully better output, you have a real reason to pay the premium.
Track total cost per document, not per seat. Per-seat pricing hides the math. A tool that costs fifty dollars per seat and processes ten documents per month is five dollars per document. A tool that costs two hundred dollars per seat and processes two hundred documents per month is one dollar per document. Most teams optimize per-seat because that is the number the vendor publishes, and most teams discover the real cost eighteen months in.
Grade the output the way your senior reviewer would grade a junior associate. Not “did it feel right.” Did it miss the auto-renew clause. Did it flag the indemnity language. Did it correctly identify the cap on liability. If you do not have a graded answer key for the workflow you picked, your pilot will produce enthusiasm, not evidence.
Trade-offs
The productivity win is real, and pretending otherwise would make this article useless. A well-tuned assistant on a well-defined workflow will save your team real hours, and the cost of missing an auto-renew clause is not theoretical. For most legal teams the productivity claim holds up under measurement, which is why the contracts keep getting signed.
The lock-in is also real, and the lock-in is the cost nobody budgets for. The data-residency and exit terms you negotiate on day one are the terms you will live with for the next three to five years. The vendor changes its policy, the vendor gets acquired, the vendor raises prices, and you discover that walking out the door is a project, not a decision. Trained users are a bigger switching cost than trained models. Your lawyers who got good at prompt engineering (writing the specific instructions that get the best output from the assistant) on vendor A’s interface are starting over on vendor B’s interface, and that retraining window is when the worst work product tends to ship.
Most teams will end up on the middle path, and it is worth saying out loud. You will probably sign the contract. You will probably run the pilot the way the procurement team suggested, not the way this article suggested. You will probably pay per-seat and not per-document. None of those choices are wrong on their own. They become wrong when you do not write down, before signing, what would make you regret them two years from now.
If you take one thing from this article, let it be the three questions. Where does the data live. What happens if you leave. Who owns the renewal. Write the answers down, attach them to the contract, and revisit them before the auto-renewal window opens.