Three People Schedule 1,000 Providers. The $4B Giant Failed.

A $4 Billion Unicorn Vanishes

On October 31, 2023, Olive AI told its remaining customers it would wind down in 45 days and sell its assets to the highest bidders. Lincoln International managed the out-of-court sale to avoid bankruptcy. When it closed in early 2024, investors who had committed $902 million recovered pennies on the dollar. No Olive-branded product remained.

The company started in 2012 as CrossChx, building digital patient-ID verification for hospital registration. It pivoted to Olive AI around 2018, rebranded, and pitched a universal AI operating system for healthcare. COVID accelerated the thesis. In 2020, Olive raised $225 million. The next year brought the peak: $400 million raised, a $4 billion valuation, headcount pushing 1,200, and the platform in more than 900 hospitals across 40-plus states, including over 20 of the top 100 U.S. health systems. Tiger Global, a16z, General Atlantic, and Drive Capital led the rounds.

Cracks appeared in 2022. Rates rose, tech valuations compressed, and growth investors retreated. Olive's burn rate, over $100 million a year, became unsustainable without a path to profitability. Revenue sat between $30 million and $60 million. Layoffs began: 450 employees in July 2022, roughly half the workforce, and CEO Sean Lane stepped down. Another 200 cuts followed in February 2023.

The asset sales split Olive's technology between two buyers. Waystar acquired the RCM automation assets: prior authorization, eligibility, and claim-status workflows that represented Olive's core competency and most of its real client value. Humata Health took the clinical AI capabilities aimed at documentation and decision support. For Olive's former hospital clients, the Waystar transition was generally smooth for prior auth and eligibility, but clients relying on Olive's broader roadmap promises for use cases that never shipped faced disruption.

CommonSpirit Health, one of the largest U.S. Catholic health systems and a flagship Olive reference, terminated contracts after deployments failed to deliver promised ROI. Customer churn surfaced visibly in 2022. The collapse didn't just kill a company; it reset buyer expectations. Health system CIOs and revenue-cycle leaders grew skeptical of "AI operating system" and "universal automation" pitches. Investor scrutiny tightened; the days of funding healthcare AI at 20–30x revenue on a vision pitch largely ended. The market now demands something narrower, proven, and integrated. Vertical scheduling tools answer that demand.

Why Horizontal Platforms Fail in Hospital Operations

Olive AI pitched a universal AI operating system that could automate any hospital workflow. The actual technology was a workflow automation platform using RPA-style automation with some machine-learning enhancements, applied to specific administrative tasks. That gap between vision and execution is why horizontal platforms collapse in healthcare.

Healthcare workflows are deeply specific. A claims adjudication system needs different AI models, data pipelines, validation logic, and compliance frameworks than a supply-chain optimization system. Olive's approach was to build a general-purpose automation layer across revenue-cycle management, supply chain, clinical operations, and patient access — all at once. The breadth looked impressive in demos. In deployment it proved fatal.

The expansion strategy was incoherent. In 2021, Olive announced simultaneous moves into clinical trial matching, supply-chain optimization, and clinical documentation. A company that had built credibility in revenue-cycle administrative automation suddenly claimed competence in entirely different functional domains. Each new vertical demanded its own integration architecture, regulatory pathway, and clinical evidence base. Olive had none at scale.

Integration complexity compounds the problem. Common healthcare infrastructure includes Epic, Oracle Health (formerly Cerner), Meditech, Waystar, Availity, Change Healthcare-style clearinghouse workflows, HL7 and FHIR interfaces, X12 transaction formats such as 837, 835, 270, and 271. A horizontal platform must navigate all of them for every workflow it touches. A vertical tool navigates one set for one workflow. The vertical tool ships faster.

Unit economics broke early. At peak, Olive spent roughly $100 million a year while generating $30–60 million in revenue. It burned through more than $800 million in funding by mid-2022 without achieving sustainable profitability. The ZIRP-era valuation assumed 20–30x revenue multiples. When the Fed raised rates in 2022 and tech multiples compressed 60–80%, that capital-efficiency story became impossible to tell.

Customers noticed. Clients reported long response times, lack of proactive assistance, and a poor understanding of the product's capabilities. Olive failed to give clients a clear picture of return on investment. Revenue-cycle leaders care about days in accounts receivable, denial rate, clean claim rate, prior authorization turnaround, cost to collect, staff productivity, and cash acceleration. Olive sold AI branding instead.

The pattern repeats. Babylon Health, valued at $4.2 billion, collapsed to zero. Pear Therapeutics, once worth $1.6 billion, fell to $27 million before bankruptcy in April 2023. Forward Health, a broad-spectrum primary care play at $1 billion, shut down in November 2024. Each tried to be the platform for everything. Each solved nothing well enough for production use.

The common thread is a fundamental miscalculation: the belief that a generalized AI platform can address the multifaceted, highly specialized challenges of healthcare. Horizontal platforms dilute focus across too many conditions or administrative tasks, making it difficult to generate the domain-specific evidence required to prove efficacy and secure reimbursement. Scaling a generalized platform across diverse healthcare needs leads to high operational costs without commensurate revenue. Effective AI in healthcare is not just about algorithms; it is about embedding those algorithms within deep clinical and operational workflows.

Navigating the regulatory landscape (FDA 510(k) clearance or De Novo classification) and securing reimbursement through CPT codes or NTAP is arduous even for highly specialized solutions. For platforms attempting to cover a multitude of conditions or services, this becomes an exponential challenge, often resulting in evidence-light offerings that fail to meet the stringent requirements of payers and providers.

The companies succeeding in healthcare AI in 2026 share a different pattern: deep domain specificity. They pick a defined operational area — claims processing, prior authorization, provider data management — and build AI systems that understand every edge case, every regulatory nuance, every integration requirement within that domain. Depth beats breadth in regulated industries. Healthcare AI must be designed for rapid deployment with existing systems, not multi-year transformation programs that lose executive sponsorship before they deliver value. Time to first production workflow is the metric that matters.

The Vertical Alternative: Durate and LeanTaaS Win with Narrow AI

While Olive burned cash trying to automate everything, a different class of company took the opposite bet: solve one workflow, solve it completely, and integrate with the systems hospitals already run. Two operators illustrate the shift: Durate, a Y Combinator-backed newcomer, and LeanTaaS, the category's revenue leader.

Durate launched in 2024 from a Rice University dorm room. Michael Mounajjed and Ben Liu, both applied math and computer science students, dropped out after their first startup (an Ethereum mining operation Mounajjed built in high school) taught them that infrastructure scales when the rules are explicit. Mounajjed grew up in Madison, Wisconsin, Epic's headquarters, with two physician parents; his mother runs the internal medicine residency at the University of Wisconsin. Liu watched his sister, a chief resident at Cambridge Health Alliance, spend weekends wrestling spreadsheets. They started as Docflow, a residency shift scheduler, and became the first Rice student company accepted into Y Combinator's Fall 2024 batch. The team is three people. Durate now schedules thousands of providers across more than 100 sites.

The product treats the rule set, not the grid, as the source of truth. A group encodes constraints once — coverage per site and shift, credential rules, duty-hour limits, fairness targets, individual preferences — and the optimizer generates a full schedule in seconds. When a new hire arrives or a hospital adds two swing shifts, the administrator changes the rule in plain English ("no one works more than two weekends a month") and regenerates. The system flags coverage gaps weeks out, when the fix is a staff swap, not an agency call. Every change is logged for payroll and compliance. A mobile app lets physicians view schedules, submit preferences, and pick up open shifts in real time. Durate connects to Epic, Oracle Cerner, ADP, and more than 1,000 other integrations so the schedule flows into the EHR and payroll instead of living in a silo. Implementation runs two to four weeks with simulations before go-live. The platform carries SOC 2 Type II certification with SSO, role-based access, and MFA.

The proof is at Brown Emergency Medicine. The academic group, 400-plus physicians, advanced practice providers, and residents staffing seven hospitals from Rhode Island to Massachusetts, moved from a static legacy scheduler to Durate in 2024. About 95 percent of scheduling and operations now run automatically. Physician complaints about scheduling fell more than 75 percent. Time-off requests are honored 100 percent. Locum usage dropped 96 percent. The group saved roughly two months of scheduling administrator time. These are vendor-reported figures from a flagship customer; apply the usual discount. But the pattern — rules-first, regeneration in seconds, integration native — is the vertical playbook.

LeanTaaS runs the same play at a different layer. As of December 2023, the company serves 1,000-plus hospitals and centers across 49 states, covering 185 health systems and 5,300 operating rooms. The iQueue for Operating Rooms product added 2,000 ORs in 2023 alone, growing that installed base more than 50 percent year over year. Revenue grew 170 percent over the trailing three years. LeanTaaS combines lean principles with predictive and prescriptive analytics to unlock capacity in the hospital's highest-margin assets — ORs and beds, rather than trying to replace the EHR.

Both companies avoid the horizontal trap by design. Legacy schedulers (QGenda, Lightning Bolt, ShiftWizard, Amion) digitized the artifact: a nicer grid that a human still repairs by hand every cycle. Durate's comparison pages target QGenda on exactly this point. LeanTaaS layers onto existing Epic and Cerner installations instead of asking CIOs to rip them out. The market is responding. The AI-in-medical-scheduling segment was valued at $200.4 million in 2025 and is projected to reach $1.45 billion by 2033, a 28.4 percent compound annual growth rate. North America accounted for 47.6 percent of revenue in 2025.

The window for vertical vendors is narrowing. Epic and Cerner are embedding scheduling intelligence into their core platforms, and health systems are consolidating tech stacks to cut vendor count. But the incumbents' roadmaps move at EHR speed — years, not quarters. Durate and LeanTaaS are compounding operational advantage now, one schedule at a time.

The Burnout Bill: A $4.6 Billion Annual Problem

That advantage is measured in burnout dollars. Physician burnout costs the U.S. healthcare system an estimated $4.6 billion annually, according to a study co-written by Christine Sinsky, MD, the AMA's vice president of professional satisfaction, published in Annals of Internal Medicine. The researchers' models put the range between $2.6 billion and $6.3 billion, but the $4.6 billion baseline, driven by turnover and reduced clinical hours, has become the industry's reference point. Broken down, that is roughly $7,600 per employed physician each year, with individual organizations seeing costs from $3,700 to $11,000 depending on specialty mix and how heavily they weight lost productivity versus replacement expense.

Turnover is the sharpest line item. Replacing a single physician costs two to three times their annual salary when recruitment, sign-on bonuses, lost billings, and onboarding are tallied. At Stanford Medicine, leaders calculated that if burnout went unaddressed, almost 60 physicians would leave within two years, creating a significant economic loss. The national figure does not even capture harder-to-quantify downstream effects: medical errors, malpractice risk, patient satisfaction declines, and disrupted continuity of care. Medicare beneficiaries alone spend an additional amount in the first year after losing their primary care physician, driven by increased specialist and urgent-care utilization.

The burnout crisis sits on top of a structural physician shortage decades in the making. The Balanced Budget Act of 1997 capped Medicare-funded residency positions at 1996 levels, a move endorsed at the time by the AMA and the Association of American Medical Colleges, which warned of a physician "oversupply." From 1997 to 2002, the U.S. residency pool grew by just 0.1 percent. HRSA now projects a shortfall of 124,180 physicians in 2027, widening to 187,130 by 2037. The AAMC's parallel forecast ranges from 46,900 to 121,900 by 2032. Because physicians tend to practice near where they train, and most residency programs cluster in urban centers and the Northeast, the shortage is geographic as much as numeric.

Understaffing and burnout feed each other. In a 2025 AMA survey, nearly half of physicians reported working with incompletely staffed teams more than one-quarter of the time; those physicians were more than twice as likely to report burnout. The share working short-staffed exceeded 60 percent in radiology, anesthesiology, pathology, and radiation oncology. Among respondents, 47.9 percent met burnout criteria, 26.4 percent intended to reduce clinical hours, and 15.4 percent reported intent to leave their organization. Inadequate team staffing was independently linked to both intent to cut hours and intent to depart.

Administrative burden, particularly EHR inbox management, is the other engine. "Physicians don't leave their careers. They leave their inbox," Sinsky has said. One study estimated that 78 percent of low-value work identified by physicians and patients was fully within health-system control, not mandated by regulation. At Cleveland Clinic, a dual-factor authentication requirement for ambulatory orders (justified as a regulatory rule) turned out to have sunset in 2021. Retyping passwords on 11 million orders annually consumed 12,376 physician hours, an indirect cost of $2.25 million. The AMA's "Reducing Regulatory Burden Playbook" now urges organizations to ask, "What can we stop doing?" — a question medicine historically struggles to answer.

Scheduling sits at the intersection of these forces. It determines whether a physician works nights three weekends in a row, whether coverage gaps get filled by expensive locums, and whether the schedule respects stated preferences, the control factor that research consistently ties to lower burnout. When Brown Emergency Medicine deployed Durate's AI scheduling across its seven hospitals, the results tracked the burnout literature: complaints plummeted, locum spend collapsed, and the group recovered months of administrator time. The ROI case is why scheduling AI is the first operational workflow hospitals buy: a medium-sized community hospital implementing AI scheduling assistance at $175,000 reached breakeven in 4.5 months. The bill for doing nothing has become impossible to ignore.

Why Epic and Cerner Haven't Solved Scheduling

Epic and Oracle Health control the operating system of the American hospital. Between them they run the clinical record for the vast majority of U.S. inpatient beds. If scheduling were a feature problem, they would have solved it a decade ago. They haven't — and the reasons reveal why vertical AI vendors still have a runway, even as that runway shortens.

Epic's dominance in large health systems is undisputed. Its single-database architecture gives it a stability advantage that Oracle Health's new cloud-native platform is still proving. But Epic's AI push (branded "Art" for charting, "Penny" for prior authorization, "Emmie" for patient-facing scheduling) targets documentation and patient access first. Emmie has rescheduled more than 14,900 appointments at Ochsner Health, saving roughly 750 staff hours. That is real throughput. It is not physician scheduling. The rules that govern an ER group's night coverage, a cardiology call rotation, or a multi-hospital trauma roster live outside the patient portal. They involve union contracts, credentialing matrices, fatigue rules, and equity formulas that no patient-facing agent touches. Epic's Agent Factory, announced in March 2026, promises a platform for building custom agents that "reason, decide, and act across workflows." Until those agents ship and mature, the gap remains.

Oracle Health is in the middle of a harder transition. The 2025 release, what executives call a "clean break" from Cerner Millennium, replaces a click-heavy, on-premise legacy with an OCI-native stack built around a Clinical AI Agent that listens to encounters, drafts notes, and surfaces scheduling actions through "Ask Oracle." The roadmap is dense: AI-driven specialty scheduling for infusion and surgical protocols, unit scheduling that coordinates patients, staff, and resources, multi-patient appointment coordination, and provider matching by payer. But migration is a multi-year undertaking. Health systems that spent years customizing Millennium with alerts, order sets, and scheduling rules cannot port them; they must rebuild. KLAS Research has documented a "partnership gap": customers report declining proactive communication since the Oracle acquisition. Anecdotal reports put implementation and data-migration fees above $10,000 even for modest ambulatory clinics, before ongoing OCI hosting costs. CIOs budgeting for hidden costs (data mapping, interface rebuilding, staff downtime) have reason to pause.

That pause is the vertical vendor's window. LeanTaaS grew to 1,000-plus hospitals and 185 health systems by sitting on top of the EHR. Durate, founded in 2024, automates scheduling for 400-plus providers across seven hospitals at Brown Emergency Medicine. Both integrate rather than replace. They ingest the EHR's provider directory, room inventory, and encounter data, then apply the combinatorial optimization the EHR's native tools don't handle.

The window is narrowing. Epic's Agent Factory and Oracle Health's expanding agent library (scheduling agents for surgical protocols, multi-patient coordination, unit-level resource alignment) signal intent to own the workflow. Both vendors are opening APIs and FHIR endpoints to third parties, but the strategic vector points toward native depth. Health systems fatigued by vendor sprawl are already consolidating RCM and point solutions; scheduling is next on the consolidation list. Vertical vendors that prove ROI now, measurable reductions in locum spend, overtime, and turnover, will be acquired or become the de facto scheduling layer the EHRs embed. Those that don't will be displaced by the very platforms they currently complement.

On October 31, 2023, Olive AI gave its customers 45 days. The vertical vendors selling scheduling AI today don't have that luxury — they have the length of a single contract renewal to show the numbers. The $4 billion unicorn is gone. The $4.6 billion burnout bill is still coming due.

Working in AI? Zero G Talent tracks the openings: see every open Durate role, browse AI jobs, the companies hiring, and the people building the field.