There is a pattern to how new technology enters the NHS. It arrives with bold promises. It encounters real problems in the real world. Critics call for it to be abandoned. Slowly, sometimes painfully, it improves. Eventually it becomes so embedded in the health service that nobody can imagine functioning without it. And then the next technology arrives and the cycle begins again.
Understanding this pattern matters right now, because AI voice systems in GP surgeries are somewhere in the middle of it. Not at the beginning, and not yet at the end. In the middle, where things are genuinely better than they were and genuinely not yet as good as they will be.
In April 2013, NHS 111 launched nationally. The British Medical Association had already written to the Secretary of State for Health warning that the service was a disaster in the making and recommending the launch be delayed on safety grounds. Their concerns proved well founded. Within days of going live, the service was unable to cope with demand. Response times stretched to five hours in some areas. Several regions suspended the service entirely within weeks of launch. NHS England's own deputy chief executive publicly admitted that patients had been let down by what she described as unacceptable failures.
The headlines at the time were scathing. Investigations uncovered poor staffing, inadequate training and non-clinical staff making decisions they were not qualified to make. Parliamentary committees concluded that the national rollout had been undertaken prematurely and without a sufficiently sound evidence base.
Today, NHS 111 handles nearly 20 million calls a year. In 2024 and 2025, patient satisfaction sat at 80 per cent, with 87 per cent of calls answered within 60 seconds. The service books over a million patients directly into GP practices every year and has become a cornerstone of how the NHS manages urgent care demand. None of that would have happened if the service had been abandoned in 2013 because its launch was rocky.
The contrast with NHS 111's eventual success makes another NHS technology story worth telling, because it illustrates what failure actually looks like and why AI voice systems in GP surgeries are not it.
The NHS National Programme for IT, launched in 2002, was described at the time as the world's biggest civil information technology programme. Its ambition was sweeping: a single, integrated electronic patient record system across the entire NHS, procured centrally, rolled out uniformly, whether individual trusts wanted it or not. By the time the programme was abandoned in 2011, it had consumed somewhere between ten and twenty billion pounds of public money and delivered a fraction of what was promised. A parliamentary committee described it as one of the most expensive IT failures in UK history.
The reasons for its failure are instructive. It was too large, too rigid and too top-down. Clinicians were not meaningfully consulted. The system was designed without sufficient understanding of the people who would use it or the environments they worked in. When problems emerged, there was no feedback loop capable of addressing them at the pace required.
The lesson is not that technology in the NHS is doomed to fail. It is that technology deployed at impossible scale, without genuine engagement with end users, without flexibility to adapt and without honest feedback mechanisms, tends to fail very expensively. The lesson points in exactly the opposite direction from centralised, monolithic IT programmes: towards smaller deployments, real-world testing, genuine user feedback and iterative improvement.
QuantumLoopAI's EMMA was not deployed through a national top-down programme. It went into individual GP surgeries, in real NHS environments, handling real patient calls. Problems that were not visible in any controlled testing environment became visible in deployment. The phonetic alphabet requirement that early patients found frustrating was identified, worked on and removed as the underlying name recognition capability improved. Accent recognition has improved through exposure to the genuine diversity of patient voices across different parts of England. Call handling times have fallen as the system has learned from thousands of real interactions.
This is not a story of a technology that failed. It is a story of a technology doing exactly what the NHS's most expensive IT lesson says it should: deploying at a scale that allows real problems to be found, maintaining the feedback loops required to address them, and improving continuously rather than claiming perfection upfront.
When NHS 111 launched badly in 2013, the response from NHS leadership was not to abandon the concept of a non-emergency telephone service. It was to fix the problems, improve the staffing, refine the clinical decision support tools and keep going. That decision, to persist through a difficult early phase rather than retreat to what existed before, is why 20 million people a year now have access to a service that genuinely helps them.
The same logic applies to AI in primary care. The question is never whether a new system has problems in its early deployment. Every significant healthcare technology does. The question is whether the problems are being identified, taken seriously and fixed. Whether the feedback loops are working. Whether the people running the system are honest about what is not yet working and committed to addressing it.
The technologies that survive their early phase and go on to transform care are not the ones that were perfect from the start. They are the ones that learned fastest from being imperfect.
Sources: