TL;DR
A GP practice at Armthorpe, near Doncaster, has withdrawn EMMA, the AI phone receptionist used across many UK surgeries, and recruited more reception staff instead. The trigger was Judith Butterfield, 71, whose speech was affected by a stroke in January and who gave up trying to book an appointment after five attempts.
Two failures compounding
Butterfield spent two months in Doncaster Royal Infirmary. The stroke weakened her right side and fragmented her speech. Both consequences mattered on the phone, and they interacted badly. The system asks callers to avoid speakerphone; she holds the handset in her working left hand, which leaves nothing free for the keypad her right hand can no longer manage. When she was asked to repeat her name and could not be understood, the call ended.
Her point about what a person would have done is the sharp one: a receptionist can ask a caller to spell it out. She has since joined another Doncaster practice and described her surprise at reaching a human being.
Practice manager Claire Oxley confirmed the system has been decommissioned at the surgery, said more reception staff have been hired, and invited her back. QuantumLoopAI, which supplies EMMA, apologised and said any practice can enable priority routing that puts callers through to the reception team directly on every call.
A pattern, not an incident
Resultsense covered the same system a week ago, when Healthwatch Rotherham reported patients unable to make it understand broad South Yorkshire accents, some giving up and walking to the surgery instead. The watchdog says Butterfield’s experience is not isolated.
The two cases are worth separating, because the remedy differs. Accent coverage is a training-data problem a vendor can work on. Speech disability is a reasonable adjustments question, and NHS bodies carry duties under the Equality Act that a supplier’s optional configuration setting does not discharge. Priority routing that exists but is switched off is not an adjustment.
Looking forward
Practices adopted these systems to clear the 8am queue, and on call-handling volume they largely do. What this case establishes is that the people most likely to need a GP are also the people most likely to fail an automated voice gate. Expect commissioners to start asking what the fallback path is before signing, rather than after a patient leaves.