The NHS spent millions of pounds on a flawed AI chatbot whose creator used aggressive sales techniques and overpromised what it could do, former staff have claimed.
Babylon Health, a tech start-up championed by Matt Hancock and advised by Dominic Cummings, promised that its AI chatbot could keep patients who didn’t need to be seen by a health professional out of the overstretched NHS.
But the technology was not as sophisticated as the company claimed, with former staff now claiming that what began as a crude tool based on “decision trees written by doctors, put into an Excel spreadsheet” never realised its promised potential. Concerns — including the fact the app missed clear signs of a heart attack or dangerous blood clots — were raised with regulators, which failed to intervene or lacked powers to do so.
Babylon’s deals with the NHS, which saw it receive at least £22 million over the past three years alone and helped it to woo investors, were in part due its links with the Conservative Party and the backing of Hancock, the health secretary from 2018 to 2021.
The Tories received more than £250,000 in donations from individuals and companies with stakes in Babylon Healthcare, including Hancock, whose failed Tory leadership bid in 2019 received £10,000.
Advertisement
Between 2015 and 2022 the company’s executives had 17 separate meetings with a total of 19 ministers, including prime ministers Theresa May and Boris Johnson, and Philip Hammond and Liz Truss during their time at the Treasury.
The promise of Babylon’s founder, Ali Parsa, to “revolutionise healthcare” was an intoxicating one. Holding forth to a rapt audience at the Royal College of Physicians, a 1970s concrete block near Regent’s Park, in central London, he explained that his health-care start-up had developed an AI-based system, a “doctor in your pocket”, that was, he said, better than the real thing.
Babylon’s app scored better on a set of standard medical exam questions than real doctors, he claimed, and would provide top-tier care for a fraction of the cost. “We will do with healthcare what Google did with information,” he liked to say. “We will make healthcare accessible and affordable to every human being on earth.”
Babylon, he said, was building a super-powered machine to diagnose conditions, manage health and suggest treatments for billions of people. Investors and influential names in and around the Conservative Party lapped it up. Between its founding in 2013 and its 2021 stock market float in New York, the company raised $1.2 billion, making it one of the most lavishly funded medical start-ups in the world. By comparison, Theranos, the fraudulent Silicon Valley blood-test firm founded by Elizabeth Holmes, raised $700 million.
Matt Hancock publicly and repeatedly promoted the “revolutionary” company. Dominic Cummings, Boris Johnson’s former adviser, agreed a contract for ten days in July 2018 to advise on communications and recruitment. The company also had advice from those with inside knowledge of how the UK health system worked. Several of the company’s investors donated to the Conservative Party and to Hancock’s Tory leadership campaign in 2019. Demis Hassabis, the 47-year-old British billionaire founder of DeepMind, Google’s AI arm, invested early in Babylon. In 2018, Hassabis was appointed by Hancock to a government AI council convened to harness technology that promised to “revolutionise our economy and our society”.
Advertisement
For Parsa, an Iranian immigrant turned healthcare entrepreneur, these were salad days. With Hancock’s support, the company picked up numerous NHS contracts, a stamp of approval that helped Parsa expand into 16 countries, where he claimed Babylon was used by more than ten million people. His wealth exploded. The Sunday Times Rich List put his wealth at £825 million in 2021.
What has unfolded in the years since for Parsa, the company he started and his former government cheerleader, Hancock, is a reversal of cinematic proportions. Hancock, who resigned as health secretary over an affair with his aide, Gina Coladangelo, now spends his days eating worms and getting punched in the face by footballers as a reality-TV contestant. Babylon, once valued at $4.2 billion, collapsed into administration in August.
Its British assets were sold for just £500,000 to a little-known US rival, which picked up a ragbag of other assets sold for another £6.3 million. The US operation has been put into a court-supervised liquidation. Parsa, once the standard-bearer for start-up Britain, moved to the US and has disappeared from view; most of his £825 million fortune has evaporated.
Parsa did not return calls or emails about the business. He is not accused of criminal wrongdoing. However, as investors comb through the wreckage, the question they will be asking is: what was behind Parsa’s reverse alchemy? How did he turn $1.2 billion into less than $10 million? Did Parsa lie to investors, or was he simply an entrepreneur who tried a hard thing and failed?
Interviews with former employees paint a picture of a tireless salesman who made promises that stretched far beyond the capabilities of the company’s technology. One employee, who left in 2021 after five years of developing Babylon’s AI models, claimed: “There was just a lot of smoke and mirrors. We were promising something that would realistically require another 30 years of research and development before it would ever be ready to see the light of day.”
Advertisement
Parsa’s event at the Royal College of Physicians in 2018 was one such instance. He showed how Babylon’s AI used a phone’s camera to analyse the facial expression of a female patient to pick up subtle cues that a doctor might miss. “This is a real consultation,” Parsa said on stage. “This is what we have built. None of this is a show.”
It was a show. The facial-analysis tool, a prop for a demo, never made it to market. The “patient” in the video was an executive assistant at Babylon, according to former colleagues and her employment record on LinkedIn. This sleight of hand was a small example of a culture fixated on form over substance, a trait common in Silicon Valley but dangerous in healthcare.
And it was by employing Silicon Valley techniques that Babylon secured NHS work and won the support of senior politicians, using its deals with the health service to sell itself to private investors and expand globally, including in the US, Rwanda and beyond.
A former employee and doctor who left the company because of concerns about the safety of Babylon’s bot claimed: “We were rushing to push stuff out before it was safe.” Dr David Watkins, a consultant oncologist at the Royal Marsden who became Babylon’s most outspoken critic, tweeted last month that it “promoted dangerously flawed technology and put the public at risk of harm ... You have to ask: why the hell was Matt Hancock and NHS England ever involved with Babylon?”
Investors entranced
At Babylon’s Knightsbridge headquarters, a plush office near Harrods with fake grass, free food and blackboards covered with scribbled maths equations, the sales team would bang a gong every time they made a sale. Hugh Harvey hated the gong. A radiologist, he joined in 2016, aged 35, enthused by Parsa’s vision of using AI to deliver better results for patients. Yet he soon realised the company was not what it seemed. At the time, the “AI” Parsa was selling was a symptom-checker that would suggest possible treatments based on answers to prompts. Behind it, though, was not a complex algorithm but “decision trees written by doctors, put into an Excel spreadsheet”, Harvey said. It was, in short, a zippier way to look up answers in a medical textbook. Harvey was dismayed. But these were early days. The company was hiring fast and beginning work on a probabilistic decision system that would be “intelligent”.
Advertisement
That system, however, was only just taking shape — though the salespeople kept banging the gong. “When the gong went off, the whole company would get up and cheer,” Harvey recalled. “But every time it went off, you’d find out that we were going to build something different. The product basically became whatever the customer wanted.” Harvey left in 2017, convinced Babylon would fail. He now runs a health-tech consultancy.
Parsa migrated to Britain from Iran at 16 and earned a PhD in engineering physics from University College London. He set up Babylon in 2013 after stepping down from Circle Health, a company he co-founded that became the first private operator to manage an NHS hospital, Hinchingbrooke in Huntingdon, Cambridgeshire. The results were mixed: at first, standards at Hinchingbrooke improved, but inspectors later found neglected patients and poor staff hygiene.
By the time Circle cancelled its NHS contract, Parsa had quit and started Babylon. He quickly entranced investors with his promise of remaking healthcare through the magic of AI. More than one former employee said he was the consummate salesman. He knew which buttons to push. He leant on his immigrant story, mined his financial nous from his days as a Goldman Sachs banker and told a story of a magical technology that was the salve for a cash-strapped NHS.
For investors — and for a health secretary desperate to right the NHS ship — the app proved irresistible. This was nearly a decade before ChatGPT, OpenAI’s powerful but error-prone chatbot, awakened the world late last year to the potential of the technology.
Advertisement
Babylon had two main products: a bot to assess symptoms and an NHS telehealth service called “GP at Hand” that would connect people, free, to real NHS doctors on a video call. In Parsa’s pitch, they were inextricably linked. By building a bot that would analyse ailments, Babylon could get rid of the need for a swathe of consultations with doctors — freeing up those doctors to better spend their time elsewhere. And when it could not, it would act as a co-pilot to the GP. It was the fuel for Parsa’s vision for a modern, streamlined healthcare delivery model — one that could save the NHS millions and make him a fortune.
Yet it turns out that building the machine peddled by Parsa was far harder than he bargained for. If a person is, from a medical perspective, a collection of risk factors, diseases and symptoms, Babylon had to build AI that would use probabilities to link all of those together. But every patient answers questions differently, and experiences ailments in varied manners. The result is that the basic health data that was meant to train Babylon’s AI was a tangle of vagaries and contradictions. This meant the model had to account for a universe of possibilities for any single symptom.
“That type of general health data is absolute trash,” said the employee who worked on the model. “The model over time just became overly complex, and the linkages just became almost impossible to give a sensible clinical probability of anything connecting.”
But Parsa was impatient. He had become obsessed with generating an “AlphaGo moment”, employees said, referring to when Google’s AI system beat the world’s best player of the board game Go. “Ali [Parsa] was very besotted with this idea and the way they were able to spin the PR around the world for it,” the former employee said.
Impossible tech
A GP who used to work at the company recalled a meeting at which Parsa demanded Babylon’s AI be capable of creating a “digital twin” of every patient, based on the data it collected from their medical files. “He sat us all down and said, ‘Right, what I want is [that] I open the app and see an MRI-quality image of my entire body in real time, and I can fast-forward, given my current risk factors, I can fast-forward 20 years in the future and see what the structure of my liver will look like. And I want it to be 100 per cent accurate.’ We all just laughed. That was, and is, completely impossible.”
Another worker claimed: “People who had dissenting voices were very quickly chewed up and spat out.” At its peak, the company employed nearly 2,600 workers around the world. Former employees claim that as Parsa brought on more investors, the pressure to deliver on his seemingly impossible demands grew. He would, it is claimed, dress down junior managers in front of their colleagues for missed targets. He handed out copies of Blitzscaling by Reid Hoffmann, the founder of LinkedIn, which is something of a bible for Silicon Valley techies hoping to build the next trillion-dollar company. “Safety and being realistic was not good. It was not rewarded. Agreeing, even if you didn’t think it was realistic, was kind of encouraged,” the worker said.
Just as doubts about Parsa’s promises stirred internally, he attracted a loud, and initially anonymous, public critic. Watkins had watched Babylon from afar and was, in his words, “quite interested” in the idea of a bot capable of making critical healthcare decisions. He began testing the bot for himself — and was appalled. He started publishing worrying interactions with the AI, using the Twitter alias DrMurphy11. He said the AI would suggest, for example, that what seemed to be a clear heart attack was likely to be a panic attack, or that an ingrown toenail was gout.
Watkins complained to the Care Quality Commission (CQC) and the Medicines and Healthcare products Regulatory Agency (MHRA) that Babylon had put into the world an “unvalidated and dangerous” product. He claims the regulators did nothing to check whether the app lived up to Parsa’s pronouncements. In 2017, the CQC raised concerns about potential misuse of the app and how it shared information with GPs. The MHRA said Babylon’s medical device did not require independent assessment, allowing the company to “self-declare” that it conformed with relevant standards. The Advertising Standards Authority also censured some claims made by Babylon.
Paul Bate, a former director at the CQC left the watchdog to join Babylon in 2016 as director of NHS services and then as its commercial vice-president. He is now the chief executive of the UK Space Agency.
But Parsa was irrepressible. The company labelled Watkins a “troll” and said its app was “100 per cent safe”. The latter claim raised eyebrows among doctors. No medical device could accurately be described as entirely risk-free except, apparently, Babylon’s.
And parts of the service were proving useful, especially the telehealth service. Launched in 2016, GP at Hand quickly drew in more than 100,000 patients. Many were young people attracted by the digital-first experience that let them jump the typical long queue for a GP and get a telehealth appointment instead.
Cracks appear
On top of that, Parsa had Hancock on his side. When Hancock was appointed health secretary in July 2018, the NHS was reeling. It had been stripped bare by years of austerity and faced an acute staffing crisis. England alone faced a shortage of more than 30,000 nurses. Given his prior role as secretary of state for digital, culture, media and sport, Hancock was, perhaps, uniquely susceptible to the idea that a technological miracle might deliver him a win in a difficult new job.
He sang Babylon’s praises in a supplement the company paid for in the Evening Standard newspaper in 2018, which led to claims he broke the ministerial code. The paper was edited at the time by the former Tory chancellor George Osborne. Labour would later demand a sleaze inquiry after it emerged that Babylon investors had donated to the Conservative Party and to Hancock’s leadership campaign.
Its links with the party allowed executives to meet with Theresa May and Boris Johnson when both were prime minister, as well as Jeremy Hunt, during his time as health secretary before Hancock. Oliver Dowden, now the deputy prime minister, hosted the company in 2021, as did Philip Hammond and Liz Truss when they were in the Treasury.
Between 2015 and 2022, the company had 22 meetings with government ministers.
The Conservative Party received more than £250,000 in donations from individuals and companies with stakes in Babylon Healthcare. Hancock’s failed bid for the Tory leadership in 2019 received £10,000 from Ian Osborne, a shareholder in Babylon via a subsidiary.
But in 2019, Parsa had powerful friends and a popular app. In August, he raised the largest single round for a European health start-up: $550 million from Saudi Arabia’s Public Investment Fund, Vostok New Ventures and Kinnevik from Sweden, and an unnamed US insurance company.
Yet cracks were already appearing, especially in the GP at Hand service. The NHS pays GPs roughly £160 per patient a year, regardless of how much care they require. This makes healthy people prized commodities. They require less intervention, which frees surgeries to bring in more patients.
Babylon turned that dynamic on its head. Its ease of use meant its patients asked for far more appointments than is typical. As a former employee put it: “What a surprise, all the 20-year-olds with anxiety from TikTok are just booking appointments all the time and asking for referrals to ADHD clinics.” The upshot was that Babylon was saddled with a patient population that was high-maintenance and expensive. Surgeries that lost patients to Babylon would often be left with those who were older and more frequently ill.
Babylon’s influx of patients increased the pressure on its AI efforts. Parsa desperately needed his machine to automate away consultations. Yet users appeared underwhelmed by the app. The Royal Berkshire NHS Foundation Trust and the University Hospitals Birmingham NHS Foundation Trust had signed deals in 2020 to use Babylon’s app to triage patients, an idea that would eventually be overtaken by the national NHS 111 system. Berkshire did not renew its contract and Birmingham’s came to an end in October last year after Babylon announced cost savings of £82 million.
The Royal Wolverhampton NHS Trust was perhaps Babylon’s biggest win. The trust signed a ten-year deal for a digital-first GP service that would allow patients to monitor their health with the aid of Babylon’s digital tools. As Babylon’s financial woes began to mount last summer, it cancelled the contract eight years early. Tim Rideout, Babylon’s general manager, told Wired magazine: “We just couldn’t afford to continue to invest.” The trust said the decision to end the deal was “mutual”.
As well as its dubious AI chatbot, Babylon also provided GP services to 100,000 people via an app that allowed patients to speak to doctors remotely. The standard GP contract was commissioned in 2017 by the NHS in Fulham, with the backing of NHS England. The GP At Hand service expanded to include Birmingham and other parts of the UK by January 2020.
All the NHS contracts, which were due to run for several years more, have now been torn up and the future is uncertain for GP at Hand, now owned by a US telehealth operator. Parsa told investors last year that the company would be cautious in any new GP services in the UK as it was effectively losing money on every patient it saw.
Wes Streeting, shadow health secretary, said: “Having promoted this company to the NHS, Conservative Ministers are now sitting back and leaving the health service to pick up the pieces as it collapses.
“The Conservatives should give the hundreds of thousands they received from individuals linked to Babylon to the NHS, and help fix this mess they’ve made.”
A spokesman for the Royal Berkshire Trust said: “Babylon Health were contracted to deliver our emergency department’s triage service as part of a pilot scheme to reduce attendances. At the end of the pilot, we found that there had been no significant reduction in attendances, and the launch of the 111 service fulfilled many of the same functions as the triage service — as a result, the contract was not renewed.”
The Royal Wolverhampton Trust said it had worked with Babylon for 2½ years on a digital-first service for local people, which had 7,000 registered patients. It said the decision to end the contract early was “mutual”.
Parsa’s dream, at least in Britain, appeared to be dying. Babylon was comprised of a loss-making telehealth service and an underpowered AI system incapable of coming to the rescue. By that time, though, Parsa had left. He moved to the US in 2021, having decided that the NHS, with its stingy spending habits, was not going to take Babylon to the promised land. So he bet everything on the biggest healthcare market on the planet.
American gamble
Southeastern Missouri in 2020 was in the grip of a public health crisis. The opioid epidemic was ripping through a population that was already dying, on average, four years earlier than the national average. Obesity rates were higher, education and income rates hovered far below national averages, and Covid was killing thousands.
Parsa showed up with the same pitch he had been giving for years in Britain. And it worked. Babylon signed a deal with Home State Health, a healthcare provider, to become the primary care service for patients in ten southeastern counties. “There has never been a greater need for a digital healthcare solution that can go directly to the source,” Parsa said when the deal was unveiled. “Meeting patients exactly where they are and when they need care most.”
Parsa’s new focus on the US hastened Babylon’s demise, because what Parsa signed up for in Missouri, and several other states, were so-called “value-based care” deals. Instead of the traditional “fee for service” model, where providers get paid when they treat people, Babylon would instead receive a lump sum for each patient in its coverage area.
To begin with, such deals looked great on a balance sheet. But as healthcare bills piled up, profits would diminish until they turned into losses. Unless, of course, a machine could take on the work for a tiny fraction of the cost of doctors.
For a former employee on the AI team, the Missouri deal sealed his decision to leave. “My first thought was, ‘Oh, my God, you’ve just taken on financial responsibility for a bunch of fentanyl addicts’,” he said. “We had just taken on all the financial risk of an incredibly vulnerable population, and this was all supposed to work because we could generate better outcomes through mass ingestion of data and use of AI and personalised care? It was never going to work. I couldn’t believe we had actually signed up for it.”
Money runs out
The folly of that strategy was not immediately apparent. Indeed, Babylon’s sales exploded. Its losses did too, but over time, as the AI improved, the whole model would turn right-side up. This was Parsa’s gamble — and it worked one last time. In July 2021, he announced a deal to float the company on the New York stock exchange. Babylon would merge with a special-purpose acquisition vehicle, or Spac, called Alkuri. Just two years ago, Spacs were all the rage, a phenomenon of the dying days of the zero-interest era. Spacs raised a pot of money with the mandate to find suitable firms to merge with. What they became in practice was a back door to the public markets for high-risk companies that otherwise would not meet the requirements of Wall Street.
Once a deal was announced, investors had the power to keep their money in the vehicle or pull it out before the deal was done. The plan was for Alkuri to inject $575 million into Babylon, valuing it at $4.2 billion. Before the merger closed in October 2021, however, most investors pulled out, leaving Babylon, which by this point was haemorrhaging cash, with $300 million less than it had expected.
Not long after the deal was announced, Parsa gave a presentation meant to convince investors to keep their money in rather than to cut and run. But it was peppered with promises about AI capabilities that were “under active development”. They had not made it to market and had no guarantee “if/when the product will be delivered”.
It was a “jam tomorrow” pitch, a plea to come along for an incredible ride that was just getting started. What investors saw was losses spiralling as quickly as sales. Parsa later called the decision to go public his “biggest mistake”.
The company raised $275 million — versus the proposed $575 million — just as the costs of all of those value-based contracts were exploding. Its share price tanked. By March this year, Babylon’s cash pile had dwindled to $25 million after losing $63 million on $311 million in sales in the first quarter. Its accumulated deficit tipped past $900 million.
AlbaCore Capital, a London debt investor that had previously given Babylon $300 million in a convertible loan, attempted to engineer a merger with a mental health app in which it was also an investor. That too fell apart. The house of cards was collapsing.
Babylon’s US operation, which accounted for 94 per cent of its sales, filed for bankruptcy in August. The British operation called in the administrators weeks later. The assets, including GP at Hand, were bought by eMed Healthcare, an American telehealth operator, which excised the name Babylon from its app.
Per Brilioth, whose Swedish investment firm Vostok New Ventures saw more than $100 million vaporised, said Babylon’s collapse had left “a scar that will hurt for a long time”. He also made clear, in a 2022 interview, the extent to which Parsa’s deals with the NHS helped woo investors.
“The NHS as the first customer, and a very politically sensitive customer, was also very keen on trying to demonstrate to its population, voters ultimately, that this was a product that worked,” he said.
David Watkins, whose attempts to raise the alarm about Babylon went unheeded, said: “The NHS endorsement and contracts were central to Babylon’s success. Big investors trusted Babylon and their AI because of those NHS contracts. They assumed the NHS had undertaken due diligence and as a consequence, handed over millions.”
A spokesman for the MHRA said it was working to strengthen its regulation of medical devices, including AI chatbots, but under current rules, the Babylon technology was able to be self-declared.
It added: “Therefore, the chatbot has not been required to have been independently assessed or certified at the pre-market stage.”
The CQC said the chatbot did not come under its remit but that it would assess the quality of care delivered to patients regardless of how that care was provided.
Hancock, whose support lent an air of legitimacy that helped Parsa take Babylon around the world, declined to comment.
Failed revolution
Parsa’s vision was, at least in the broadest of strokes, not wrong. He pushed telehealth years before Covid forced us into it and turned it into a pillar of the healthcare model that will not be reversed. AI, meanwhile, has arrived. When San Francisco’s OpenAI released its ChatGPT bot last November, it stunned the world. Here was a wildly powerful machine capable of writing screenplays and software code and acing standardised tests.
The tech that underlies it, a type of AI called neural networks, is aeons beyond the spreadsheets with which Babylon started. And a crop of start-ups seeking to capitalise on this new AI revolution are making the Babylon pitch all over again.
Emad Mostaque, founder of the London-based Stability.AI, which develops AI models that others can build apps on top of, said this year: “Imagine having a professional doctor in your pocket. We’re at the edge of that right now.” Limbic, a British start-up, recently won certification for its AI therapy chatbot that helps people self-refer for mental health disorders.
Meanwhile, eMed, which bought Babylon’s assets, said it “will continue to provide Babylon’s digital-first 24/7 primary care” and recently informed users via text that it would soon add a “medical weight management” service, a likely reference to the weight-loss drug Wegovy.
Parsa sold a story of towering ambition: an AI revolution that would allow anyone to access world-class medical care through their smartphone. Instead, nearly a decade of toil and $1.2 billion from some of the world’s top investors yielded a renamed telehealth app looking at hawking weight-loss pills.
This was not the revolution Parsa promised.