Shipping an AI copilot vets say saves them about 15 minutes per case

The biggest stall in a veterinary appointment happens after the client leaves.

I helped lead market research: interviews and case walkthroughs with five practicing DVMs and techs.

We mapped where a visit stalls and found four points: searching protocols, calculating doses, explaining the plan to the client, and re-entering the history once the room is empty.

That last one falls outside the appointment entirely, and it was the largest of the four.

We designed the whole records system. We could build one piece of it.

The first plan was a full EHR replacement: patient workspace, medical history, medications, and an assistant, all mocked up on the design system I had built for the product.

The team was part-time, ten hours a week at most, and the dev team could not take on a build that size.

I recommended to the founder that we cut to the one piece that did not need a records migration to be useful: the assistant. It ran on the same design system, and we kept the patterns simple enough for the team to maintain at that capacity.

The pilot set the roadmap: dose calculator, discharge generator, lab interpreter.

We ran ChatVET in one working clinic and expanded to three. We logged what vets asked for.

Three tools came back in demand: a medication dose calculator, a discharge generator, and a lab interpreter. Dosing was the most-asked question by a wide margin.

All three open from a new chat or from the middle of one, so the case already on screen carries into them.

Answers are restricted to the Merck Veterinary Manual and leading journals.

A template library replaced prompt-writing with a few fields.

In the pilot, vets wrote prompts that returned inconsistent results, and the same clinical question came back differently depending on how it was asked.

I designed a template library organized by clinical job: SOAP notes, discharge instructions, differentials, and client emails. Vets replace the highlighted details and press enter.

Consistency mattered as much as the time saved. The same question returns the same shape of answer regardless of who asks it.

The discharge generator writes the client handout from the consultation that already happened.

I designed the print output and reworked the discharge generator around the chat context.

It reads the consultation back, asks which of the medications discussed should go home with the client, and fills the handout from there. The owner leaves with instructions they can follow, and nobody stays late writing them.

500 vets logged in last month, with no sales effort behind it.

The web app is live. 500 registered users signed in over the past month, worldwide, and clinics are piloting the business tier ahead of a paid rollout. Growth has been bottom-up.

Vets in the pilot reported saving about 15 minutes per case. That was never instrumented, so it is what they told us and nothing more.

The screens in this study are the version I designed and shipped. My role shifted in June from design to investor and marketing material, and the product has kept changing since.

If I ran this again, I would instrument the discharge flow before shipping it. We never set up a way to find out whether those handouts changed anything after the client left.