Skip to content
English - United Kingdom
Contact us
  • There are no suggestions because the search field is empty.

EMMA is Cutting the Call When It Cannot Understand the Patient

Purpose of Document

This is one of the most emotive complaints a practice can raise. When a patient reports that EMMA cut their call, it creates immediate concern about patient safety, access, and the practice's reputation. Staff must understand exactly how EMMA behaves when it cannot understand a caller, what safeguards are in place, and how to triage the complaint accurately before responding to the practice.

Background: How EMMA Handles Calls It Cannot Understand

EMMA does not simply drop calls when it cannot understand a patient. There is a structured sequence of behaviours designed to give every patient multiple opportunities to be heard before the call ends.

The key components that govern this behaviour are:

Clear Voice is EMMA's speech recognition layer, trained specifically on NHS primary care audio. This covers regional UK accents, clinical vocabulary, and real-world phone call conditions such as background noise, poor signal, and calls made from busy environments. As of V12 (April 2026), Clear Voice has been significantly improved for accuracy across all patient demographics.

Sentinel monitors every call in the background. When EMMA encounters something unfamiliar, including a heavy accent, background noise, or a poor line, Sentinel ensures EMMA falls back to a clinically safe pathway rather than attempting to guess. This means EMMA will redirect to a safe option rather than continue processing an unclear request.

Flow is EMMA's conversational timing layer. It reads each patient's speaking rhythm, adjusts to their pace, detects when they have finished speaking, and handles interruptions without losing the thread. This reduces false cut-offs caused by EMMA responding before a patient has finished their sentence.

Lifeline is EMMA's dropped-call fallback. At the start of every call, EMMA tells the patient that if the call drops or the signal is poor, they will receive a secure link to continue their request through SOFIA. The patient completes their request in writing and the practice receives the same structured outcome. Nothing is lost.

When a patient goes completely silent, EMMA now uses a two-step check (April 2026): first asking "Hello, can you hear me?" and then "Hello, I can't hear you, are you on mute at all?" before disconnecting. Mobile callers who disconnect at this point receive an SMS continuation link via SOFIA. Landline callers receive next-step signposting. All disconnect messages include 999 and 111 signposting for callers with urgent needs.

Common Causes and What to Check

1. The patient's audio quality was too poor for EMMA to process reliably

Calls made from a busy environment, on a poor mobile signal, on a bus, or with significant background noise can make it difficult for any speech recognition system to interpret what is being said. EMMA's primary response to poor audio is to use SOFIA, the text message fallback, which sends the patient a link to continue their request in writing.

What to do:

  • Review the call recording for the affected call if available
  • Check whether the audio quality was clearly poor throughout the call
  • If so, confirm that SOPHIA sent the patient a continuation link
  • Reassure the practice that the patient's request was not necessarily lost. If SOPHIA was sent, the patient had a route to complete their request
  • If SOPHIA was not triggered and the call simply ended, escalate to engineering with the call ID

2. The patient was silent or unresponsive and EMMA reached its timeout

Some patients pause for a long time, go on hold from their end, or experience a dropped line mid-call. Before April 2026, EMMA repeated the same prompt when a caller went silent, which felt robotic. As of April 2026, EMMA now uses the two-step natural check described above before disconnecting.

What to do:

  • Check the call recording to confirm the patient went silent rather than speaking unclearly
  • Confirm that the two-step check was completed before the call ended
  • If the call ended before the two-step check, this may indicate a pre-April 2026 call or a regression. Escalate to engineering with the call ID
  • Confirm that mobile callers received an SMS continuation link after disconnection

3. EMMA entered a call loop due to repeated speech recognition failures

Before the March 2026 speech recognition stability fix, EMMA could enter call loops in some cases when it could not interpret spoken input, particularly around medication names and clinical intent. This presented as the call going in circles with no resolution before eventually ending.

What to do:

  • Check the call recording for signs of repeated misunderstanding and looping behaviour
  • Check whether the incident predates the March 2026 stability fix
  • If it is occurring after that date, escalate to engineering with the call ID and a description of the loop behaviour observed

4. Audio distortion or call failure during peak morning hours

Before the March 2026 peak hour infrastructure fix, audio distortion and dropped calls were occurring during Monday morning peak hours between approximately 8:00 and 8:10 AM. The infrastructure was underprovisioned for the volume of concurrent calls during that window.

What to do:

  • Check whether the reported call drop occurred on a Monday morning between 8:00 and 8:10 AM
  • If before March 2026, this was a known infrastructure issue that has been resolved
  • If occurring after March 2026 during peak hours, escalate to engineering as a potential infrastructure regression with the call timestamp and surgery ODS code

5. EMMA transferred the call rather than cutting it, but the practice perceived it as a cut

In some cases, EMMA transfers a caller to reception rather than continuing the call. If the transfer is not answered, the patient may experience what feels like being cut off. This is not a call cut by EMMA but an unanswered transfer.

What to do:

  • Check the call recording to determine whether EMMA transferred the call or ended it
  • If the call was transferred and not answered at the surgery's end, advise the practice to review their transfer handling
  • Confirm that the transfer number configured for that surgery is correct and active

6. The patient had a strong accent and EMMA could not confidently process their response

EMMA's Clear Voice speech recognition is trained on NHS-specific audio including a wide range of UK regional accents. However, very strong accents combined with poor line quality can still present a challenge. When Sentinel detects that EMMA cannot confidently process a response, it routes to a safe fallback rather than guessing.

What to do:

  • Check the call recording to assess whether accent recognition appears to be the issue
  • Confirm that SOPHIA was triggered as a fallback and an SMS continuation link was sent if the caller was on mobile
  • If accents from a specific demographic are consistently causing issues at a particular surgery, raise this with the engineering team as a potential tuning request for that surgery's patient population

What EMMA Does NOT Do

It is important to be clear with practices on the following points:

  • EMMA does not simply hang up on patients who are difficult to understand. There is always a fallback pathway.
  • EMMA does not abandon patients mid-call without attempting to re-engage.
  • Sentinel ensures EMMA always routes to a safe option rather than guessing when uncertain.
  • Lifeline ensures patients on poor connections always have a route to complete their request via SOFIA.

If a practice believes a patient was cut off with no fallback offered, this should be investigated and escalated. It is not expected behaviour.

Step-by-Step Triage Process

  1. Ask the practice for specifics. Which patient? What time? What was the patient trying to do?
  2. Pull the call recording and listen to the full call.
  3. Identify the cause: poor audio, patient silence, call loop, peak hour distortion, or transfer not answered.
  4. Check whether the relevant platform fixes were in place at the time of the incident (March 2026 speech recognition, March 2026 peak hours, April 2026 silent caller handling).
  5. Confirm whether SOPHIA was triggered and an SMS continuation link was sent to the patient.
  6. If the call ended without any fallback being offered and the cause is unclear, escalate to engineering.

What to Tell the Practice

Keep communication calm and factual. Example wording:

"EMMA is designed to never simply cut a call without offering a fallback route. We are reviewing the specific call you have flagged to understand exactly what happened. In most cases where a call ends unexpectedly, the patient will have received an SMS link to continue their request through our text-based channel. We will confirm this and come back to you shortly."

Do not confirm that EMMA cut the call without fault until the recording has been reviewed.

Common Mistakes

  • Telling the practice EMMA cut the call before reviewing the recording
  • Not checking whether SOPHIA was triggered and an SMS link was sent
  • Assuming accent recognition issues are unfixable without checking whether a surgery-level tuning request is appropriate
  • Not checking whether the incident predates the relevant platform fixes
  • Treating a missed transfer as a call cut by EMMA

Escalation Guidance

Escalate to engineering if:

  • The call recording confirms EMMA ended the call without attempting the two-step silent caller check
  • SOPHIA was not triggered when it should have been
  • Call loops are occurring after the March 2026 speech recognition stability fix
  • Audio distortion or call drops during peak hours are occurring after the March 2026 infrastructure fix
  • A specific patient demographic or accent pattern is consistently causing call endings at one or more surgeries
  • A clinical safety concern exists because a patient in distress was unable to complete their call

When escalating, always include:

  • Surgery name and ODS code
  • Call ID or timestamp
  • Description of what the call recording shows
  • Whether SOPHIA was triggered
  • Whether the issue is isolated or affecting multiple patients

Last Reviewed: May 2026 Owner: Support and Customer Success