Olive AI’s $400M at $4B Valuation Collapsed After Layoffs

The Paperwork Problem That Brought AI to Therapy

Solum Health's Annie cut prior-authorization denials 38 percent at Always Keep Progressing, a multi-site ABA operator, and compressed insurance verification from 12-minute phone calls to 12 seconds, Solum Health's published figures show. The AI front-office assistant, launched from Y Combinator's S22 batch, handles intake, verifies insurance, manages prior authorizations, and fills schedules — not another dashboard to monitor, but a system that executes the work.

JP Montoya, Solum Health's founder and CEO, watched his father spend a career in healthcare before building his own outpatient therapy practice from a single location to a multi-country operation serving more than 20,000 patients. "Most software out there stores your data," Montoya said. "It doesn't do your work." That observation became Solum Health.

Adoption in applied behavior analysis and outpatient therapy has accelerated because the operational pain is concentrated and measurable. ABA practices run on CPT codes 97151 through 97158 (assessment, adaptive behavior treatment, protocol modification, caregiver and group codes), each with distinct place-of-service rules for clinic, home, school, and telehealth. Medicaid managed-care organizations, commercial payers, and third-party administrators each maintain different authorization requirements that shift without notice. A therapy practice with multiple locations faces a different payer mix, workflow rules, and scheduling preferences at every site. Solum's approach was to embed directly into the practice management systems these clinics already use: CentralReach, Motivity, Rethink, AlohaABA, Hi Rasmus, Catalyst, and Lumary, plus productivity tools such as ClickUp, Monday.com, HubSpot, Zoho, and Zendesk. If an EHR isn't listed, the company handles the integration so staff don't have to.

The deployment model reflects clinic reality. Solum's implementation team handles 90 percent of configuration, including EHR integration, payer-mix mapping, authorization workflow design, and staff training, with a two-week go-live target. The platform carries HIPAA business-associate agreements, SOC 2 Type II certification, AES-256 encryption, penetration-test verification, role-based access controls, and a 99.9 percent uptime SLA. For multi-site operators and management-services organizations, each location maintains its own rules while operations leadership gets a unified view with centralized reporting.

Hi-5 ABA's COO, Benjamin MacGowen, said AI innovation is opening new opportunities for expanding access to excellent services. Hi-5 ABA's data shows average annual administrative savings of $247,000 and measurable ROI within 90 days for most practices.

Annie's workflow covers the full authorization lifecycle. Real-time eligibility checks return full benefit breakdowns by CPT code, including deductible status, copay, session limits, and place-of-service rules. Referrals arriving by fax, portal, or email are captured, patient data extracted, missing documents collected, and family follow-up automated, all synced to the practice management system. Initial authorizations and re-authorizations submit across Medicaid, commercial, and managed-care payers with every submission tracked, every deadline monitored, and every follow-up automated, including re-auth renewals before coverage gaps disrupt treatment. Medicaid eligibility rules, secondary-payer coverage, and state-specific MCO requirements are navigated continuously. When a TPA switches, an MCO updates policies, or coverage terms change, the system detects it in real time and re-verifies eligibility so teams aren't blindsided by surprise denials. Parent and caregiver communication runs on automated, personalized outreach for authorization status, upcoming appointments, and required documentation, reducing no-shows and keeping families engaged.

Where the Money and Time Go

The manual prior-authorization machine is expensive, slow, and error-prone. Across roughly 5,000 distinct PA codes, the labor adds up: provider organizations devote the equivalent of more than 100,000 full-time registered nurses annually just to chase authorizations. The AMA's 2024 physician survey found physicians handle a median of 39 PA requests per week, consuming roughly 13 hours of staff time. Practice spending on PA staffing jumped 43 percent between 2019 and 2024 even as reimbursement lagged, and 86 percent of providers describe the burden as high or extremely high, per the Medical Group Management Association.

AI platforms attack the problem at three choke points. First, they eliminate the manual hunt for requirements. Traditional workflows forced staff to identify that authorization was needed, locate the correct payer form, pull clinical notes from the chart, fill the form by hand, fax or upload it, then wait days and follow up repeatedly. AI systems using clinical evidence extraction and LLM-enriched form filling parse unstructured notes, visit summaries, lab results, and medication histories to pull the exact evidence a payer needs: diagnosis codes, prior therapy history, lab values, clinical criteria — without manual chart review. They pre-populate payer questionnaires with citations from the clinical record and flag documentation gaps before submission rather than after denial. Early deployments show 83 percent reductions in handling time and approval cycles cut from weeks to hours, per Develop Health's analysis.

Second, they cut denials by submitting cleaner packets the first time. PA denials follow predictable patterns — missing documentation, outdated payer criteria, incorrect diagnosis codes, insufficient evidence of medical necessity. Machine-learning models trained on historical determinations now flag requests with elevated denial risk before they leave the EHR, letting clinical teams strengthen the evidence package proactively. Calibrate's pharmacy team, for example, was submitting five to seven PAs per member because coverage was unclear. With benefits-driven routing informing the workflow upfront, that dropped to one targeted PA per member — an 80–85 percent reduction in PA volume for the same patient population, per Develop Health's case study.

Third, they compress the follow-up loop. When a denial does occur, AI systems analyze the denial reason, generate a clinically grounded appeal letter drawing on brand-specific guidelines and patient-record data, and route it for provider review and resubmission. Omni-channel submission pathways, including direct PBM integrations, HL7 FHIR-based Prior Authorization APIs, AI-generated fax, and phone-based AI outreach with human fallback, mean the platform can pursue a decision across every channel payers still use. Prosper's voice-first agent, Kate, automates calls to payers to initiate and check status, delivering results with 99 percent accuracy in under two hours, per the company's published data. One large national insurer reported its AI tool made the PA process up to 1,400 times faster, per GetProsper.ai.

The Patient Toll and Regulatory Response

The patient-access impact is measurable. That survey found 93 percent of physicians say PA delays care, and 29 percent have witnessed a serious adverse event (including hospitalization or permanent harm) because a treatment stalled waiting on approval. Ninety-four percent report PA delays access to necessary care; over 90 percent say it has a negative impact on patient outcomes. The Health Affairs Scholar survey found 31 percent of patient respondents reported negative impact on ability to seek treatment, and 92 percent of provider respondents reported care delays due to PAs. With an expected RN shortage of 200,000–450,000 in 2025, automating half of the PA workload would effectively return more than 100,000 RN-equivalents to clinical care, per the same study. PA is also a high contributor to burnout — 42 percent of provider respondents cited it, with follow-ups with the other party as the top factor.

CMS's Interoperability and Prior Authorization Final Rule (CMS-0057-F) now requires payers to respond to standard requests within seven calendar days and urgent requests within 72 hours, with full API-based electronic prior authorization required by January 2027. AI-powered platforms that combine direct PBM integrations, AI calling, and automated follow-up are already consistently achieving approvals in 20 hours or less for many therapeutic areas, well inside the new regulatory window. A 2025 Cohere Health national provider survey found 99 percent of clinicians and 96 percent of office administrators reported confidence in AI-driven PA, though implementation quality matters enormously.

The Automation Arms Race: UiPath, Olive, and the Clinic-Native Wave

UiPath enters the prior-authorization conversation not as a healthcare specialist but as the category-defining RPA platform now pivoting hard toward agentic automation. Founded in Bucharest in 2005 by Daniel Dines and Marius Tîrcă, the company went public in April 2021 with a $1.3 billion IPO (one of the largest software listings in U.S. history) after a $35 billion private valuation. In July 2022 UiPath acquired Re:infer, a London NLP startup that turns unstructured documents and communications into structured data, a capability directly applicable to the fax-heavy, PDF-laden prior-authorization workflow. By July 2024 the company released Autopilot, embedding generative AI and NLP into its Business Automation Platform. Three months later it previewed Agent Builder, letting developers combine traditional RPA robots with GenAI-powered agents. March 2025 brought the Test Cloud for AI-agent testing and the acquisition of Peak, a Manchester firm building agentic inventory and pricing optimization. April 2025 saw the launch of the UiPath Platform for Agentic Automation with Maestro, an orchestration layer that coordinates robots, agents, and people. Forrester named UiPath a Leader in Document Mining and Analytics Platforms; the platform earned AIUC-1 certification for agent security and reliability. Revenue reached $1.43 billion in 2025 against a $163 million operating loss and a $74 million net loss, with 3,868 employees. Healthcare customers include Omega Healthcare, citing a 100 percent productivity increase and 50 percent faster invoice turnaround, and NYPCC, using UiPath to serve mental-health communities. The platform is horizontal; clinics adopting it for prior auth typically configure it themselves or work with implementation partners.

Olive AI followed a different arc and serves as the sector's cautionary tale. The Columbus, Ohio startup raised $400 million at a $4 billion valuation led by Vista Equity Partners, positioning itself as the AI workforce for healthcare revenue cycle. By late 2023 the company had shut down, its assets sold off in pieces, a collapse Healthcare Dive attributed to overexpansion and a product architecture that couldn't deliver the autonomous denial prevention it promised. Olive's collapse demonstrated that healthcare-specific automation requires more than capital — it demands deep integration with payer rulesets, provider workflows, and the regulatory cadence that changes every January and July.

The competitive pressure is visible in acquisition behavior. UiPath's purchase of Re:infer, Cloud Elements, StepShot, ProcessGold, and Peak in successive years reads as a deliberate stack completion: process mining, API connectivity, desktop capture, and now domain-specific agentic AI. UiPath also acquired WorkFusion to strengthen financial-services agentic solutions. For therapy clinics, this means the prior-authorization automation they buy today may arrive as a module inside their EHR, a bot deployed from their RPA center of excellence, or a purpose-built agent from a startup that survives the funding winter.

Funding Follows the Integration Layer

Venture capital has flooded into healthcare administrative automation over the past three years, though the trajectory resembles a boom-and-correction cycle more than a steady climb. The clearest signal came in 2021 when Olive AI secured that round. The capital didn't vanish with Olive. Tebra, a practice-management platform formed from the Kareo-PatientPop merger, secured $250 million in 2024 to accelerate AI innovation across its revenue-cycle suite. Medusind, a revenue-cycle management services firm, announced a strategic affiliation with Tebra to expand RCM solutions for independent practices. Kareo (backed by Travis Kalanick) merged with a healthcare IT firm in a deal Bloomberg covered, consolidating another piece of the outpatient therapy stack.

The funded competitors cluster around three models: EHR-embedded modules (Epic, Oracle Health, athenahealth), standalone prior-authorization platforms (Cohere Health, Waystar, Inovalon), and horizontal RPA vendors moving vertical (UiPath). The horizontal players bring deeper automation stacks but shallower clinical logic; the vertical players bring payer-specific rules engines but narrower integration surfaces. Public-market proxies are mixed: UiPath (PATH) trades well below its 2021 peak. No single analyst firm has published a consensus TAM for "AI-driven prior authorization in outpatient therapy"; the category is too new and too nested inside broader RCM forecasts. The therapy-specific slice (ABA, PT, OT, speech) is smaller, faster-growing, and more denial-dense, which is why Solum and peers target it first.

The funding pattern reveals a strategic shift: early checks went to end-to-end automation dreams (Olive). Recent rounds back platforms that own the practice-management layer (Tebra) or the payer-connection layer (Cohere, Waystar) and layer AI on top. That architecture wins because prior-authorization logic lives in the payer rules, not the provider workflow — and the rules change weekly. Companies that control the integration surface can update the logic; pure-play AI vendors cannot.

For therapy clinics, the investment signal is practical: the tools that survive will be the ones embedded in the EHR or practice-management system they already use. Standalone prior-authorization bots face a distribution moat they rarely cross. The next wave of capital will likely fund the data layer (structured denial-codes, payer-policy graphs, real-time eligibility feeds) because that's what makes the AI accurate enough to replace a human biller.

The Workforce Splits in Two

The automation wave hitting outpatient therapy revenue cycles is not just changing software stacks — it is rewriting job descriptions. Clinics that adopt AI prior-authorization tools like Solum Health's Annie, which handles front-office workflows 24/7, report that staff previously spending hours on phone trees and fax confirmations are being redirected to exception handling and patient communication.

What happens to the medical biller who used to chase denials? The research is thin on therapy-specific headcount data, but adjacent signals are clear. Olive AI, which once commanded a $4 billion valuation and $400 million in capital, collapsed after layoffs — a trajectory of rapid funding, aggressive hiring, then shutdown that illustrates how fast the talent demand curve can invert when automation matures.

First-party hiring data from Zero G Talent's board, while not healthcare-specific, shows the salary bands tech companies now pay for adjacent skill sets. Both companies added dozens of openings in a single week. When EHR giants embed native prior-authorization AI and platforms bet on autonomous agents, they compete for the same talent pool. A medical biller who learns to configure denial-prediction models or audit agent decision logs moves into that band; one who only works queues does not.

The net effect is polarization. Clinics that automate early shed low-leverage follow-up labor and hire up-market for AI oversight. Those that lag retain manual teams but face rising denial rates and staff burnout — the very conditions that drove Solum's founders to build Annie from their own clinic operations. ABA and multi-site therapy groups with volume to justify AI oversight hires will pull ahead; solo practices may rely on vendor-managed services, effectively outsourcing the skill shift to platforms like Tebra. The next hiring wave will not be for people who call payers. It will be for people who teach machines which payer rules matter.

Working in frontier tech? Zero G Talent tracks the openings: see every open ASML role, browse frontier tech jobs, openings at Stripe, and the people building the field.