Petco: a good sales associate should not play veterinarian

A customer walks in for pet food but brings something less visible: worry. Their dog has stopped eating its usual meal, their cat is behaving differently, and the internet has already supplied three diagnoses. An associate wants to help. That entirely understandable instinct needs a clear professional boundary.

Practise boundaries, not theatrical confidence

Petco offers dedicated veterinary services. That distinction matters: helping someone choose a product is not the same as providing medical care. What follows is my independent proposal for AI sales training, not a claim about a system Petco has deployed.

I would build a role-play where a virtual customer begins with an ordinary product question, then introduces a circumstance requiring specialist advice. The associate does not win by naming an illness. They succeed by recognising the limit of their role and explaining the next step calmly.

A second set of exercises covers everyday choices: pack sizes, the usual feeding routine, budget and information actually printed on a label. AI varies the context while the employee practises asking before recommending. Otherwise we get the familiar retail performance: an answer looking desperately for its question.

The third mechanism is feedback on an exact sentence. Rather than announcing that someone lacks empathy, the trainer identifies an interruption, an unsupported product promise or a missed clarification. A manager reviews disputed feedback and chooses one skill for the next short practice session.

A referral should end with a practical next step, not a vague instruction before returning to the shelf. The exercise can require an explanation of the available route to professional help, without guaranteeing an appointment or its outcome.

The model does not write the rules

My three primary measures would be correct specialist referrals in assessment scenarios, unsupported-claim frequency and customer ratings of how clear the advice was. Sales conversion can be observed separately, but a transaction that should never have happened must not earn the model a gold star.

The knowledge base needs current product information, approved consultation boundaries and verified referral routes. Qualified specialists must review the medical scenarios and escalation criteria. There is no need to upload identifiable customer stories. Synthetic situations can teach the skill without collecting a visitor's private concerns.

A pilot could cover one category in a small group of stores, alongside a comparable group using existing training. Establish a baseline first, allow several weeks of practice and have human assessors score the final conversations without knowing which training each employee received. Enthusiasm is not an evaluation method.

This is the kind of AI employee training I find useful: it helps someone say that a question lies outside their expertise while remaining helpful. Sometimes good service ends with a purchase. Sometimes it ends with an appropriate referral. A lasting customer relationship should have room for both.