AI for Pet Health: What We Learned Building DogMD

DogMD is an AI health companion for dog owners — symptom triage, photo analysis, food safety, and knowing when something is an emergency. It holds a 5.0 star rating on the App Store.

Consumer health is a different discipline from professional tools. The user is not an expert, the stakes are emotional, and the worst case is severe.

Design for the worst case

With a professional tool, the average interaction is a lookup. With consumer health, the average interaction is a worried owner at 11pm, and a small fraction are genuine emergencies.

You cannot design for the average. The product has to be built so that the rare emergency is handled correctly even at the cost of making the common case slightly less smooth.

Concretely: the system escalates readily. If symptoms are consistent with something time-critical, the answer stops being informational and becomes an instruction to seek care now. That is a deliberate bias toward over-referral.

The hardest problem is not diagnosis

It is calibrating urgency.

Owners arrive with a description and an implicit question: is this fine, or do I need to act. Get the urgency wrong in the reassuring direction and the outcome can be terrible. Get it wrong in the alarming direction repeatedly and the tool becomes noise people ignore — which produces the same failure eventually.

We resolved this by making urgency a visible, colour-coded, structured part of every response rather than something buried in prose. The owner sees the severity assessment before they read the explanation.

What we will not do

Diagnose. The product describes possibilities and directs to care. It does not tell someone what their dog has.

Recommend dosages. Weight-based medication maths is exactly the kind of task that looks well-suited to AI and is completely inappropriate for it in a consumer product.

Replace the vet. The framing throughout is "help you understand and know when to go," never "handle it yourself."

These constraints cost engagement. They are correct anyway.

What worked

Photo input. Owners are much better at showing than describing. A photo of a rash beats three paragraphs of adjectives.

Food safety as an entry point. "Can my dog eat this" is the highest-frequency question and one of the few with a clear factual answer. It brings people in and builds trust for the harder conversations.

Plain language. Veterinary terminology helps no one at 11pm.

Explicit emergency detection. A distinct, unmissable response state for the cases that need one.

The transferable lesson

In consumer health, the product's job is to make people better at deciding, not to decide for them.

That reframing resolves most design arguments. Ask whether a feature helps someone make a better call about care. If it substitutes for that call, it is probably the wrong feature regardless of how well it demos.

More about DogMD.