What exit criteria should be in the contract from day one?

VetScribe AI listens to vet-client conversations and drafts clinical notes for veterinary practices. Ingrid Castellan is Director of Clinical Operations at Brightfield Veterinary Group, fourteen clinics. Farid Salame runs operations at a sibling veterinary group that tried to switch away from a similar vendor.

How to answer this, stage by stage

Nobody is scoring you on whether you can list contract clauses. They're scoring whether you can rank them by which one you truly cannot fix later.

Let's learn

Picture the day you sign an AI vendor contract, before you've thought once about the day you might need to leave it.

Say a veterinary group buys an AI assistant that listens to vet-client conversations and drafts the clinical note automatically. Before it, a vet at Brightfield Veterinary Group finished each visit, then typed up the clinical note by hand, roughly 90 minutes a day per vet across a busy clinic. VetScribe AI drafts that same note during the visit itself, cutting typing time to about 15 minutes a day for review and correction.

Brightfield signed a three-year contract. Nobody at the table asked what would happen if, someday, they wanted out. The contract said nothing about data export, because nobody had a reason to think they'd need it.

Every later step in leaving a vendor depends on this first box being settled. Skip it, and the rest of the chain never gets to run.

Here's the turn: the missing clause itself never cost Brightfield anything, for years. The real risk was what it would cost the day someone actually wanted to leave, because by then, three years of clinical notes existed only inside VetScribe's own system, in a format nothing else could read.

At its worst: a clinic that genuinely needed to leave, for cost or quality reasons, would find itself choosing between paying tens of thousands of dollars to get its own patient records back, or staying with a vendor it no longer trusted, simply because leaving had become more expensive than staying.

What I would leave alone: the day-to-day contract terms around price and support response time don't need this level of urgency. Those can be renegotiated at any renewal, and a slightly worse price this year doesn't compound the way a locked data format does.

The lesson: a contract's hardest clause to fix is never the one you argue about hardest at signing. It's the one nobody thinks to ask for, because it only matters on a day nobody's picturing yet.

Now here is the same thing as a story

The short version above is what you'd say defending this clause to Brightfield's board. Read this one for how the gap actually surfaced.

Every quarter, Ingrid Castellan pulls the VetScribe contract binder and reads the renewal clause first, out of habit, more than any real concern. For three years, that habit had never once turned up anything worth a second look.

Farid Salame runs operations at a veterinary group two states over, using a similar AI scribe vendor. The two of them talk shop once a year, at a veterinary practice management conference, over a hallway coffee that always runs long. This year, Farid mentioned, almost as an aside, that his group had spent the spring trying to switch off their AI vendor after a bad run of service outages.

Farid's group could still argue about price. They could not get their own three years of notes back in a shape anything else could read.

"We got quoted forty thousand dollars," Farid said, stirring his coffee like it was nothing. "Four months. Turns out our notes only ever export as flat scanned pages, one per visit. Our new system needed actual fields. We're paying someone to retype three years of our own patient records."

Ingrid laughed along, the way you do at a conference story that isn't yours. Then she flew home and pulled Brightfield's own VetScribe contract for the first time in three years, not to read the renewal clause, but to read everything else.

There was no data export clause. There was an auto-renewal clause, silently in place since year one, that nobody had ever needed to notice. Three years of clinical notes, tens of thousands of patient visits, sat inside VetScribe's system in a format the contract never once mentioned exporting.

Brightfield's contract had none of these four. It had a tool that worked, and nothing written down about the day it might stop.

Ingrid didn't leave VetScribe. The tool still worked well, and the relationship was still a good one. But she reopened the contract that quarter and added exactly the clause Farid's group had learned about the hard way: structured export, on demand, at a fixed cost cap, with clear ownership of the content itself.

The clause costs nothing to write on day one. It costs forty thousand dollars to write three years late.

Here's what I'd take back. Brightfield signed a three-year term where the trial period only ever tested whether VetScribe wrote a good note, never what a clean exit would look like. That was a reasonable place to focus attention when the whole team was deciding whether the tool worked at all. It stopped being reasonable the moment three years of irreplaceable clinical history accumulated behind a door nobody had checked for a handle.

I would go back and put the export clause in on day one, next to the clauses about accuracy and price, not three years and one hallway conversation later. And the part I'd tell myself: we didn't need to distrust VetScribe to ask for this. We needed to remember that trusting a vendor today says nothing about whether we'll still want them in three years.

ORDER, in one screenNot "list every clause a lawyer might want." ORDER is what tells you which one to fight for first, because it's the one you can't fix after the fact.

Only the first of these five gets harder to fix with every passing year. The rest can wait for a renewal conversation.

The recap, one line per letter: outcome is a clean future exit, reversibility is the export clause being the one thing you can't fix retroactively, dependency is every other exit step waiting on that one being answered first, and evidence is a free sample export during the trial, before three years of data pile up behind it.

Rank. Data export first, ownership second, performance-based exit triggers third, the transition window fourth, and price or exclusivity terms last, because those are the ones you can still fight for even after the ink is dry.

And if you want to be sure it really works, try it somewhere elseSame five letters, a municipal permitting office's AI review tool instead of a veterinary scribe. A different record that gets harder to reclaim every year.

Solange Mercier runs the technology contract for her city's permitting office, which uses an AI tool to pre-review building permit applications for completeness before a human inspector ever opens the file. Mapped onto ORDER: outcome is the city's ability to switch permitting vendors without losing years of applicant history and inspection notes. Reversibility is the same shape as Brightfield's: a missing export clause means permit records accumulate in a proprietary format that becomes harder to extract every year the office waits. Dependency is that no competing permitting vendor can even be evaluated fairly until the city knows what a real data handoff would cost. Evidence is requesting a sample export of ten years of historical permits during any renewal negotiation, before committing to another multi-year term.

The old decision here isn't an unread contract binder, it's a different reversal: the permitting office's original contract auto-renewed every year by default, and nobody on staff was ever assigned to review it, because renewal felt like the safe, do-nothing option. That made sense when the tool was new and nobody wanted to disrupt an unproven rollout. It stopped making sense once a decade of permit records had piled up with no export clause ever attached to any of those renewals.

Same two doors. A different decade of records stuck behind the second one.

Swap the trigger and it still runs.
Speed: an interviewer caps you at sixty seconds. Say "the data export clause first, because it's the one you can't fix after the fact," and stop.
Cost: legal review time is limited and can't cover every clause equally. Say so honestly, and spend that limited attention on the export and ownership clauses, not the ones you can still renegotiate at renewal.
The model gets better, for real: if the vendor's export tooling genuinely improves later, that's evidence the switching cost has dropped, not a reason to stop asking for the clause up front, since the next vendor might not be as good.

Where people run it wrong.
They spend the whole negotiation on price and none of it on data portability.
They treat every exit clause as equally important instead of ranking by what's actually hardest to undo.
They wait until they want to leave to find out what leaving actually costs.

How to use it live. The moment someone asks "what should be in this contract," ask back: which of these clauses gets more expensive to fix with every year we wait? Put that one first, and let everything else follow.

Flashcards (tap any card to flip it)

Check yourself Score: 0 / 0