PetsDetect & Microsoft Copilot

Pets at Home is a complex retail and pet care business: stores, ecommerce, veterinary practices, grooming and a large loyalty scheme all sit in the same customer world. The scale matters: around 450 stores, about 450 veterinary practices and grooming services handling up to 17,000 pets a week.

In that kind of operation, profit protection work is under pressure and heavy with data. The job is not simply to move fast. It is to move fast enough to protect the business while keeping fair human judgement in the loop.

That was the useful problem: reduce the time spent compiling case material, so specialists could spend more time on skilled review. Less hunting around. More thinking.

My role was hands-on delivery: I designed and built the Copilot Studio agent that became PetsDetect, then shaped the documentation and handover needed for a specialist team to use it with confidence.

The job was not to replace judgement. It was to remove the slow compilation work around review: bring case material together, make it easier to assess, and keep the final decision with the people who understand the context. That matters in fraud review, where speed is valuable but overreach can damage genuine customers.

Copilot Studio was the right fit because the work needed more than a demo. It needed a maintainable agent pattern: clear behaviour, readable outputs, documentation for the team, and enough structure for a live retail environment.

The value came from a very practical shift: less time spent assembling a case, more time available for skilled assessment. That is where the return showed up. Pets at Home linked the agent to potential annual savings in the seven figures. The external numbers followed the same pattern: decisions in seconds rather than 30 minutes, 10 times faster identification, and 20 times more cases processed.

Those numbers matter because they point to shipped value, not innovation theatre. The win was a faster, clearer way for a specialist team to handle demanding review work without moving accountability away from humans.

This is where agent work gets serious. Autonomy is useful only when the guardrails, handover and review model are clear. Otherwise, you have a fast system and a slow trust problem.

Done well, that kind of agent helps specialists move faster without making the process feel less accountable. The point is not to replace judgement. It is to give people better prepared material so they can focus on the decision.