The graveyard of healthtech is not full of bad products. It is full of products that worked clinically and still could not survive as a business. That distinction matters, because it means the lesson is almost never "build better technology." It is almost always about distribution, regulation, or capital structure, and the pattern repeats with uncomfortable consistency across very different companies.

Three recent post-mortems, three different failure modes

Forward Health raised more than $650 million and reached a $1 billion valuation building a membership-based, tech-forward primary care clinic model, then shut down and closed its locations in late 2024. The ambition was genuine: replace the ordinary doctor's office with a sleek, tech-driven experience. The failure was structural: physical clinics are capital-intensive, slow to scale, and the membership model never reached the density needed to make the unit economics of a real-estate-heavy business work at venture speed.

Cydoc, a bootstrapped health AI company, survived seven years with paying customers, patents, and demonstrated clinical impact before shutting down in 2025. Its founder's own framing is the sharpest line in any of these post-mortems: deploying health AI is only 20% of the challenge, the other 80% is workflow integration, sales cycles, and the operational grind of getting a clinical tool actually used inside real institutions.

Kintsugi, a depression-detecting AI startup, shut down in 2026 after struggling to get its tool through the FDA, ultimately open-sourcing the technology rather than continuing to fund the regulatory pathway for depression-detecting AI, a pathway that has proven far harder than many mental health AI teams initially modeled.

The pattern underneath the differences

Different products, different clinical categories, different capital amounts. The recurring thread across most digital health failures is overestimating adoption speed, underestimating regulatory friction, and burning cash on growth before product-market fit was actually proven, what one advisory firm bluntly calls confusing fundraising momentum with product-market fit.

Break that down into the specific traps:

  • Confusing a signed pilot with a repeatable sales motion. One enthusiastic health system does not prove a go-to-market model. The founders in most of these post-mortems had real, credible early customers and still could not translate that into the second and tenth customer at the pace their burn required.
  • Underpricing the regulatory pathway, in time and money. Mental health AI, diagnostic AI, and anything touching a clinical claim faces a slower and more expensive FDA and clinical evidence pathway than most technology founders, coming from non-regulated software backgrounds, initially budget for.
  • Building a capital-intensive model that needs venture-speed growth to work. Physical care delivery, like Forward Health's clinics, has real-estate and staffing economics that do not compress the way software economics do, no matter how much capital is available.
  • Treating clinical validation as a launch requirement instead of a day-one build. Retrofit evidence generation after the product exists is slower and more expensive than building the clinical evidence pipeline alongside the product from day one.
  • Mistaking a great product for a solved distribution problem. Health systems and payers move on institutional timelines. A product that is clinically excellent still needs a sales and integration motion built for a 12 to 18 month enterprise cycle, not a consumer growth loop.

What survivors did differently

The companies that avoided this pattern generally did three things early: they picked a distribution model that matched their capital intensity (software-only companies stayed asset-light rather than opening clinics), they built regulatory and evidence strategy into their product roadmap rather than bolting it on before a funding round, and they treated the first handful of enterprise customers as a repeatability test, actively diagnosing why each sales cycle took as long as it did rather than attributing delays to bad luck.

A pre-mortem checklist worth running now

  • Does our growth model require venture-speed scaling of something capital-intensive (real estate, hardware, staffing)? If yes, is our funding runway actually long enough for that asset class's real growth rate?
  • Have we mapped the actual regulatory pathway, timeline and cost for our specific clinical claim, with input from someone who has done it before, not just our own optimistic estimate?
  • Can we point to two repeatable, independent sales cycles that closed on comparable terms, not just one relationship-driven pilot?
  • Is clinical evidence generation built into our product roadmap as an ongoing workstream, or is it a one-time item we plan to "do before the next round"?

The takeaway

Nobody in these post-mortems built a bad product. They built a product whose distribution model, regulatory pathway, or capital structure could not keep pace with the growth their investors expected. Read the failure pattern as an operating checklist, not a cautionary tale about someone else's company, because the pattern is generic enough to apply to almost anything currently being pitched.