Kinney Drugs put AI on its pharmacy phones. It lasted three months
A 95-store pharmacy chain made an AI voice the only way in for refill calls. Patients got wrong refills and delays. The keypad came back in August.
| Company | Kinney Drugs, a pharmacy chain headquartered in Gouverneur, New York, with 95 store numbers on its own store locator across New York and Vermont (kinneydrugs.com, our count, 5 November 2026) |
|---|---|
| What ran | Burt, an AI voice and text assistant built by the vendor Synerio, handling inbound refill calls, refill reminders and prescription notifications (VTDigger, 29 July 2026; Kinney Drugs Burt terms) |
| How long | inbound AI calls ran from May 2026 to early August 2026, about three months (VTDigger, 29 July and 10 August 2026; WCAX, 7 August 2026) |
| What went wrong | customers reported refills requested for drugs they did not need, delays between a doctor's order and the pharmacy, accounts not found and missed pickup notices (VTDigger, 29 July 2026) |
| What changed | inbound patient calls went back to the old touch-tone menu; Burt stays for refill texts, which now need the patient to opt in (WCAX, 7 August 2026) |
| In writing | Kinney's Burt terms say Burt may not be accurate, send disputes to individual arbitration and cap the company's liability at 100 US dollars if other limits fail (kinneydrugs.com/kxterms, edited 12 May 2026) |
| Cost | not disclosed; no figure for the contract, the call volumes or customers lost has been published |
Kinney Drugs, a pharmacy chain with about 95 stores in New York and Vermont, made an AI voice assistant called Burt the front door for prescription refill calls in May 2026. By early August its president, John Marraffa, had said publicly that the company did not get the experience right, and inbound patient calls went back to the old touch-tone menu. Burt survives only for refill texts, and only for patients who opt in.
There is no reported data breach and no regulatory action on record. The project failed on the parts a project plan usually treats as details: how a caller gets past the machine, what counts as consent, and what the company measured.
What ran, and for how long
Burt was built for Kinney by Synerio, a Texas company that sells AI tools to pharmacies, according to VTDigger's reporting. It placed calls and sent texts about refills, pickups, vaccines and unpaid balances, and it answered patients who rang in. Kinney's written statement to VTDigger gave the reason: handle routine, non-clinical contact so pharmacists have more time for patients.
The paperwork is older than the phone line. Kinney's terms for Burt's calls and texts carry an effective date of 18 August 2025 and were last edited on 12 May 2026, the month customers and the company date the switch on inbound calls to. VTDigger published a special report on 29 July built on accounts from nearly a dozen customers. Marraffa's admission came in a WCAX interview on 7 August. That is about three months from launch to retreat.
What went wrong
The failures customers described were not exotic. One customer told VTDigger that Burt rang almost daily about refills without saying which drug it meant, that she approved them assuming they were right, and that she now had four bottles of a twice-weekly medicine at home. Others reported longer gaps between a doctor's order and the pharmacy acting on it, Burt failing to find long-standing accounts, and pickup notices that never came.
Two design choices turned those errors into a failure. First, the old system let callers press keys to reach a pharmacist; one customer described Burt as offering no keypad route, so the only way through was to talk to the machine. Second, the terms say that giving Kinney your phone number is consent to Burt's automated calls and texts, with an opt-out afterwards. Several customers said they did not know the system had changed until they heard the voice.
We did not, and we own that.
That is Marraffa on WCAX, after saying the company had got privacy and security right but not the experience.
What did it cost?
Kinney has not published a figure for the contract, the staff time or the customers it lost. What is on record is softer. Customers told VTDigger they had moved prescriptions elsewhere, and VTDigger reported that confused patients were leaning harder on pharmacists, the people Burt was meant to relieve. A Kinney director quoted in the same report said the tool was reducing calls to the pharmacy. Both can be true: fewer calls answered by a person is not the same as fewer problems.
The written figure is in the terms. If the other limits fail, Kinney's liability to a patient for anything arising from Burt is capped at 100 US dollars, and disputes go to individual arbitration. The terms also state plainly that Burt may not be accurate and may answer a different question from the one asked. That protects the company in a claim. It did nothing to protect the relationship with a patient holding the wrong bottle.
As read on 5 November 2026, the calls-and-texts terms still carry the 12 May edit date and the clause that a phone number counts as consent, three months after the company said texts would become opt-in. That may simply be a page nobody updated. It is also the kind of gap a regulator or a journalist finds first.
Does this apply to a 200-person business?
Yes, more directly than most AI stories. Kinney is not a tech company. It bought a vendor's product, put it on the line its customers use most, and judged it partly by how many calls it kept off staff. Any firm putting AI on its phones, its order line or its customer inbox is making the same three choices, usually without noticing them.
What it does not mean
It does not mean voice AI cannot work for routine contact. Kinney kept Burt for outbound refill texts, the narrowest and most checkable job it had. It does not mean the vendor's technology was the main fault; the record does not show that. And it does not mean a disclaimer in the terms is protection. Here it was the opposite: a sign that the company expected errors and had planned for the claim, not for the caller.
What to do before you switch one on
Keep a route to a person that takes one key press or one word, and test it yourself from a landline. Make the new channel opt-in for existing customers, or at minimum tell them before the first call. Have the system confirm the specific item, by name, before it acts on anything that ships, charges or refills. Measure repeat contacts and complaints per thousand interactions, not just calls deflected, for the first eight weeks. Start in one branch or one customer segment, with a named owner who can switch it off the same day. And when you change the rules, change the terms page the same week.