The AI healthcare cohort that didn't make it

On January 21, 2022, IBM sold what was left of Watson Health to the private equity firm Francisco Partners for a reported price above $1 billion. Watson had once been pitched, in a 60 Minutes segment, as the machine that would cure cancer. By the time it was carved out and rebranded as Merative, it had spent the previous decade trying, and largely failing, to convert oncology pattern-matching into clinical decisions that doctors trusted and insurers paid for. The sale was the first of six obituaries that would run between 2022 and late 2024, covering more than $3.2 billion in disclosed venture funding and one corporate division of indeterminate but enormous internal cost.

The five startups that followed Watson out of the category were not fringe operators. Mindstrong Health had raised $160 million from General Catalyst, ARCH, and Foresite. Pear Therapeutics had raised $396 million and gone public through a SPAC that valued it at $1.6 billion. Babylon Health had raised $1.15 billion and gone public at $4.2 billion. Olive AI had raised $902 million at a peak valuation of $4 billion. Forward Health had raised $657 million from Founders Fund, Khosla, SoftBank, and a roster of celebrity angels. They were not the same company. They had, on inspection, the same problem.

Mindstrong was the cleanest version of the early thesis. Founded in 2014 by Paul Dagum, Rick Klausner, and the former NIMH director Tom Insel, the company proposed to read mental illness off a smartphone, treating keystroke cadence and scroll patterns as digital biomarkers for depression and relapse. The science papers were real. The product, when it shipped, was a telehealth therapy app with a passive sensing layer that few payers reimbursed and few clinicians used. By the time Mindstrong wound down on March 10, 2023, it was selling sessions, not signals. SonderMind bought the remaining technology and the brand quietly disappeared.

Pear Therapeutics was supposed to be the company that proved the regulatory path. Its lead product, reSET, received FDA de novo clearance in September 2017 as the first prescription digital therapeutic for substance use disorder. Somryst, for chronic insomnia, followed. By 2021 Pear had three cleared products and a SPAC deal with Thimble Point Acquisition. By April 7, 2023, it had filed for Chapter 11. The clearance was real. The reimbursement was not. Commercial insurers, citing weak real-world evidence and the absence of an established code path, declined to pay at scale. Pear's assets sold at auction that summer for $6.05 million, less than a typical Series A.

Babylon Health is the case study for what happens when a chatbot meets a payer contract. Founded in London in 2013 by Ali Parsa, Babylon ran an AI symptom checker, an NHS-contracted virtual GP service called GP at Hand, and a US business built on capitated Medicaid and Medicare Advantage deals. The SPAC merger with Alkuri Global closed in October 2021 at a $4.2 billion pro forma equity value. The medical loss ratios on the US capitated business never came down. A take-private deal with the Swiss neurotech firm MindMaze fell apart in August 2023. The US entity filed Chapter 7. The UK arm went into administration. UK operating assets sold to eMed for £500,000.

Olive AI told a different story and ended in the same place. Founded in Columbus in 2012 by Sean Lane, Olive sold an AI worker for hospital back-office tasks, prior authorization, claims status checks, eligibility, revenue cycle. The pitch was that robotic process automation, wrapped in a friendly persona named Olive, would let health systems shed administrative cost. A 2022 Axios investigation, drawing on customer interviews, reported that much of what Olive shipped was conventional RPA glued together with services labor, and that promised savings often did not materialize. Olive sold off two product lines in 2023 and shut the rest of the business on October 31, 2023. The peak valuation, set in 2021, had been $4 billion.

Forward Health was the last of the cohort and in some ways the most theatrical. Adrian Aoun, a former Google special-projects lead, founded Forward in 2016 as a membership primary care clinic, $149 a month, body scanners in the lobby, a software-first medical record. In November 2023 Forward announced the CarePod, a self-contained AI-powered medical booth meant to deliver primary care without a clinician in the room, and a $100 million round to build them. The pods, deployed in malls and offices, reportedly trapped patients inside, misread blood draws, and required staff dispatch to recover. On November 12, 2024, Forward closed every location with minimal notice. Landlords were left with empty kiosks.

Read together, the six failures rhyme on three points. The first is the gap between FDA clearance and insurer reimbursement, which turned out to be wider and meaner than the pitch decks assumed. A de novo clearance, as Pear discovered, is a permission slip to sell, not a contract to be paid. Commercial payers want randomized evidence, a CPT or HCPCS code, and a comparator. The prescription digital therapeutics category never got a stable code. Without one, every sale was a custom negotiation, and custom negotiations do not scale to the unit economics of a venture-backed company.

The second common thread is the cost of capital. Pear, Babylon, and the late-stage rounds at Olive and Forward all priced off a 2020 to 2021 environment in which SPAC sponsors, crossover funds, and growth-stage investors were willing to underwrite long paths to profitability at double-digit revenue multiples. When the Fed began raising rates in March 2022, that market closed. Companies that had been built to raise a $300 million round in 18 months suddenly had to extend their runway by cutting the operating model that the round had been built to fund. Pear and Babylon ran out of cash with cleared products on the shelf. Olive ran out of cash with paying customers still on the books.

The third thread, and the one most often missed in the post-mortems, is hospital procurement. Health systems are not a market in the way that small businesses or consumers are a market. They are a set of committees, value analysis groups, IT security reviews, compliance officers, and CFOs who will, when sufficiently annoyed, kill any pilot before it touches a production system. Olive's customers, according to the Axios reporting, included systems that had bought AI and received scripted automation. They told each other. Watson's customers, including MD Anderson, which terminated its Oncology Expert Advisor project in 2017 after spending roughly $62 million without a clinically usable system, told each other too. By 2022, the buyer was suspicious before the first demo.

The clinical evidence problem sat underneath all of it. A 2018 STAT News investigation, drawing on internal IBM documents, reported that Watson for Oncology had been trained on a small number of synthetic cancer cases rather than real patient data, and that company specialists and customers had identified multiple examples of unsafe and incorrect treatment recommendations, including a recommendation that a 65-year-old lung cancer patient with severe bleeding receive a drug carrying a black-box warning against use during serious bleeding. A physician at Jupiter Hospital told IBM, according to STAT, that the hospital had bought it for marketing and could not use it for most cases. That was the public artifact. The private version of that conversation happened at every health system in the country for the next six years.

Babylon's symptom checker had its own version of the problem. In 2018 a British GP, David Watkins, began publishing examples in which the chatbot missed presentations of heart attack and other emergencies in women. The MHRA reviewed the product. Babylon disputed the findings. The dispute mattered less than the fact that it was now in the air, and that NHS commissioners and US Medicaid plans had to weigh reputational risk alongside cost. By the time the SPAC closed in 2021, the company's largest US contracts were already underwater on medical costs. The chatbot was a marketing surface on top of an insurance company that could not price risk.

Forward's collapse looked, on the surface, like a hardware story. The CarePods, photographed in glossy press tours, had the aesthetic of an Apple Store crossed with an MRI suite. The actual failure was more boring. Primary care, as practiced, depends on a clinician taking a history, examining a patient, ordering tests, and following up. Forward built a kiosk that automated the middle steps and left the others unstaffed. The kiosk could not, by itself, write a controlled substance prescription, manage a complex chronic patient, or handle the moments when something went wrong inside the pod. Membership growth stalled. The unit economics never closed. The November 2024 shutdown surprised members and landlords but not, by then, the people who had been watching the deployments.

The combined disclosed funding across Mindstrong, Pear, Babylon, Olive, and Forward comes to roughly $3.265 billion. Watson Health's IBM-internal spend is not public but is widely estimated in the billions across the 2015 to 2022 period. None of the six companies survived as an independent operator. Watson's assets live on inside Merative. Mindstrong's tech sits inside SonderMind. Pear's products were repurchased in fragments at the bankruptcy auction. Babylon's UK shell was sold for the price of a London flat. Olive's remaining product lines were absorbed by Waystar and Humata Health. Forward's kiosks were repossessed.

What the category demonstrated, in expensive detail, is that healthcare punishes the parts of a software business that are usually its strengths. Speed of iteration runs into regulatory review. Network effects run into hospital procurement. Marketing claims run into clinical evidence requirements. A growth-stage round of $200 million, deployed into a product that cannot get a billing code, does not buy time. It buys a longer obituary. The builders, investors, and operators who lived through this cohort are the ones now making the next round of decisions in the category, with, one hopes, a more accurate map of where the bodies are.