AI Pulse · Business Models & Winners

Kinney Drugs launched an AI phone assistant called Burt in May to handle inbound prescription refill calls. Within three months the chain pulled it back to outbound texts only after hundreds of complaints about incoherent answers, wrong dosage information, and missed refill notifications. Customers calling in for medication help now get the old touch-tone menu again.

The lesson isn’t that voice AI doesn’t work. It’s that Kinney put it on the highest-stakes surface in the business, inbound medication calls, with no opt-in and no fallback path from day one. When you’re dealing with dosages and refills, the cost of a single error is enormous. The chain treated the assistant like a finished product when it needed guardrails, human handoff triggers, and a slower rollout on lower-risk tasks first.

This is exactly the kind of scoping mistake we flag when building something like the Omni Command Centre. Voice AI can handle a lot, but you start with tasks where a mistake costs you time, not trust. Outbound appointment reminders, order confirmations, basic FAQs. You build confidence, collect real conversation data, tune the prompts and handoff rules, then move up the ladder to higher-stakes interactions. You don’t launch on the thing that can hurt someone if it hallucinates.

Kinney’s pivot to outbound texts only is smart. It keeps the efficiency gain, lowers the risk, and gives them time to fix what broke. If you’re scoping voice AI for your business, write down the three calls where a wrong answer would cost you the most. Those go last, not first. Build the system that earns the right to handle them.