Therapists Lose $71,500 Yearly to Paperwork, Klarify Aims to Cut
A Category Shift Goes Public
Klarify launched on June 30, 2026, with 8,300 therapists across five countries and a spot in Y Combinator's Spring 2026 batch, an AI-native operating system for therapists that refuses to sit in the therapy chair. By September the Toronto Guardian placed Canadian sign-ups above 11,000, with U.S. adoption accelerating after the YC demo day. The platform centers on Klara, an AI assistant that drafts clinical notes, generates treatment plans, prepares assessment reports and clinical letters, creates between-session resources, supports 104 languages, and produces visual session mindmaps that surface recurring themes across a caseload. Internal data shows 71 percent of Klara usage now occurs outside traditional note-taking: insurance support, treatment planning, operational tasks, letters, and the administrative work that surrounds the session.
The distinction defines the market's bifurcation: durable value lies in provider-facing operating systems that survive insurance audits and multi-jurisdiction privacy rules, not in consumer chatbots. Consumer chatbots market direct-to-patient CBT conversations. Klarify's position is explicit: AI should not replace therapy. The platform handles the operational, administrative, and financial infrastructure around modern mental healthcare, including clinical documentation, treatment plans, insurance workflows, assessment reports, between-session resources, and practice growth. "The therapy itself stays human. Always," said Moody Abdul, co-founder and CEO. "AI should handle the operational burden so therapists can spend more time helping people."
Abdul and CTO Alexander "Bergie" Bergholm met at the University of British Columbia. Abdul previously co-founded Circleback.ai, an AI meeting-notes platform, and led $20 million in enterprise contracts at LinkedIn. Bergholm built self-driving car research at UBC, conducted deep-learning signal processing in Finnish military special forces, and most recently led machine-learning infrastructure at Workday. The spark came from Abdul's own clinician, who saw Circleback and said, "I wish I could use this, but I can't." That constraint — privacy rules, consent requirements, audit exposure — shaped every architectural decision. Client consent comes first. The therapist records the session, Klara drafts the note, the therapist reviews and signs. The company is contractually barred from training AI on clinical data. Data stays in Canada. The platform meets HIPAA, PHIPA, Quebec Law 25, and UK GDPR.
NewMediaWire reported that Klarify estimates therapists sit at the center of a $22 billion operational economy across the U.S., Canada, U.K., Australia, and New Zealand, spanning reimbursement infrastructure, compliance systems, documentation software, scheduling platforms, outsourced billing, and administrative labor. According to NewMediaWire, the addressable market could exceed $50 billion at maturity as the platform expands into client-side experiences, institutional contracts, and adjacent infrastructure. A 103,000-subscriber audience built through The Future of Therapy podcast and newsletter provides a distribution channel most vertical SaaS companies would envy.
The launch arrives as insurers operationalize automated reimbursement infrastructure years ahead of practitioners. "Therapists are entering an increasingly automated reimbursement environment badly outgunned," Abdul said. "They deserve modern infrastructure on their side too." That asymmetry — payer-side AI denying claims faster than providers can document medical necessity — is the fault line the next section explores.
Why Therapists Are Drowning
The average therapist spends 11 hours a week on unpaid documentation. A 2025 industry burnout report translates that to roughly $71,500 in lost revenue per clinician each year, time that generates zero billable units while rent, malpractice premiums, and student loans compound. The same survey found 54 percent of mental health clinicians report frequent burnout, with administrative burden the primary driver. Only 8 percent reported no symptoms.
The mechanics are brutal. A single 50-minute session (CPT 90837) demands 15 to 20 minutes of documentation afterward. Multiply by a 30-client week and you exceed eight hours of paperwork, purely on progress notes. Research from Eleos Health puts documentation at 35 percent of working hours, roughly 16 minutes per client encounter. The American Psychological Association's 2019 survey had already clocked psychologists at 20 to 25 percent of work hours on session documentation alone, with another 10 to 20 percent consumed by authorizations, correspondence, and billing. Community mental health nurses and social workers fare worse: EHR tasks consume 37 percent of shift time.
Insurers have grown more demanding. They want more detail, more clinical justification, more evidence of medical necessity. Clinicians respond by writing longer notes. Longer notes take more time. The cycle compounds. Most EHRs were built as transactional billing systems on legacy architectures, not as interactive platforms for behavioral health. They lack built-in SOAP, DAP, or BIRP templates. They force irrelevant fields. They require constant toggling between separate systems for notes, scheduling, and billing. Every context switch costs attention.
The cognitive load differs from other specialties. Writing a progress note means recalling the arc of a 50-minute session, selecting clinically relevant details, framing them within a diagnostic and treatment context, and producing a document that must withstand insurance review, licensing board scrutiny, or subpoena. That is not transcription. It is active clinical work performed after the session, often at home, on nights and weekends. The term "pajama time charting" entered the burnout literature to describe it. Seventy-seven percent of clinicians report finishing work later than desired or documenting at home.
That after-hours pattern is one of the strongest modifiable predictors of emotional exhaustion. A 2017 Annals of Internal Medicine study found that for every one-hour increase in EHR use outside office hours, burnout risk rose significantly, independent of total hours worked. Therapists who cite documentation as their primary stressor show notably higher exhaustion scores than those who cite caseload size or client severity. The "re-exposure loop" compounds it: a therapist who spends 25 minutes writing a detailed note for a client processing domestic violence or suicidal ideation re-immerses in that material without the session's clinical structure, alone, fatigued, at day's end.
Attrition signals flash. Fifteen percent of clinicians intend to leave their organization within 24 months. Twenty-six percent plan to reduce clinical hours in 2025. Thirty percent say they have considered leaving the field due to documentation load. Forty-eight percent say workforce shortages have already pushed them toward the exit. SAMHSA projects a shortfall of roughly 31,000 full-time-equivalent mental health providers by 2025; HRSA extends that gap past 8,000 FTEs by 2030. Clinician turnover costs half to two times an annual salary per departure, and behavioral health runs three times the healthy turnover rate.
Private practice therapists absorb the full weight. No medical scribes. No administrative support. No EHR helpdesk. The person in the therapy chair all day writes every note, updates every treatment plan, chases every prior authorization. A therapist billing $150 per session across 25 weekly clients generates $3,750 in weekly billings. Fifteen hours of documentation at their effective hourly rate represents $2,250 per week in unbillable time, over $108,000 per year. Documentation errors increase in late-shift and post-shift work, with omissions more common than commissions. In mental health, an omitted safety assessment or unreported substance use is a liability exposure, not a typo.
The workforce shrinks while demand grows. Burnout-driven attrition from documentation burden actively reduces an already undersupplied labor pool. Depleted clinicians have less capacity for the therapeutic alliance. Attrition means longer waitlists. The documentation burden doesn't just cost therapists their evenings; it costs clients access. That pressure is why nearly three in ten psychologists now use AI tools monthly for administrative tasks, and why 73 percent of clinicians agree systemic solutions are needed. The adoption curve isn't driven by curiosity. It's driven by survival.
The Audit Gauntlet
In March 2026, a federal Medicaid audit uncovered $285.2 million in improper payments for ABA therapy in Colorado alone: documentation failures, uncredentialed staff, missing diagnostic referrals. A parallel Maine state audit found $45.6 million in improper Medicaid payments for the same service line. Medicaid spending on ABA therapy hit $10 billion in 2025, a 421 percent jump from 2021, while children served grew just 67 percent. Spending grew six times faster than utilization. CMS released a 173-page toolkit in August 2026 to help state agencies combat fraud, waste, and abuse in ABA programs, signaling that scrutiny applied to autism services is the template for general behavioral health.
| State / Audit | Improper Payments | Key Findings |
|---|---|---|
| Colorado (federal Medicaid, March 2026) | $285.2 million | Documentation failures, uncredentialed staff, missing diagnostic referrals |
| Maine (state audit) | $45.6 million | Improper Medicaid payments for ABA therapy |
| Collective reviews (CMS toolkit) | At least $198.4 million | Fraud, waste, abuse patterns across multiple states |
The audit environment hardened in 2026. Mental Health Parity has been federal law since 2008, but enforcement was uneven until this year. Payers now enforce MHPAEA requirements aggressively. CPT code time-based documentation rules have tightened. The definition of "adequate medical necessity language" has narrowed. Notes that passed utilization review in 2023 trigger denials today, not because clinical work changed, but because the documentation bar moved.
Insurers increasingly use AI and data mining to allege improper payments based on minor documentation deficiencies or obscure requirements. For payors seeking financial performance, the same tools used to identify fraud claw back reimbursements on far less sensational grounds. In a National Association of Insurance Commissioners survey published May 2025, 84 percent of 93 responding health insurers reported using AI or machine learning. Denials climbed alongside adoption: 74 percent of 1,000 practicing physicians surveyed by the AMA in 2025 said denials increased over five years, and six in 10 fear AI will push denial rates higher. Only 24 percent report denial reviews consistently conducted by qualified clinicians. Just 7 of 38 Medicaid MCO states require plans to disclose AI use in prior authorization.
An audit reviewer at a regional BCBS plan wrote in a 2024 industry compliance newsletter: when the Assessment section of ten consecutive notes contains word-for-word identical language, she treats that as evidence the clinician is not documenting the actual session.
AI-generated notes create three exposure points during an audit. First, tools that generate notes from session summaries describe what happened without checking whether that matches the documented treatment approach: the medical necessity gap. Second, tools using the same template with similar inputs produce notes that look nearly identical across sessions, signaling the clinician is not reviewing or individualizing. Third, tools that generate notes without prompting for session duration create CPT code vulnerability: billing 90837 when the note documents a 45-minute session, or when multiple sessions lack documented start and end times, creates an audit flag easy to find and hard to defend.
Automated transcription and summarization tools lack clinical judgment. They misidentify speakers, confuse similar-sounding names, diseases, or medications, omit acronyms or technical terms, misinterpret overlapping exchanges, or inaccurately capture medication names, diagnoses, dosages, or follow-up instructions. If AI-generated text enters the record without adequate review and later supports an inaccurate claim, the result may be overpayment exposure and, in some cases, False Claims Act scrutiny. The Department of Justice has already prosecuted schemes involving AI-generated fraudulent consent recordings. AI use in health care introduces new FCA liability theories across care delivery, documentation, and claims processing. Government agencies use AI to detect fraud: DOJ and CMS leverage tools like the WISeR Model and the Health Care Fraud Data Fusion Center to enhance detection and likely increase FCA investigations.
Six states enacted laws in early 2026 restricting how insurers may use AI in coverage decisions. Indiana bars downcoding a claim on AI output alone without a professional's review, and bars providers from submitting AI-prepared claims without human review. Utah requires insurers to disclose AI use in prior authorization to the state insurance department, providers, and enrollees, and requires reviewers to exercise independent medical judgment. Washington prohibits sole reliance on AI to deny, delay, or limit care, reserves adverse determinations for licensed professionals, and requires carriers to report AI-assisted denials. Alabama requires AI-assisted decisions to rest on the individual's medical history and clinical circumstances, with annual certification that tools do not rely on group datasets. Maryland requires quarterly reports on adverse determinations, services involved, and whether AI was used, and lets the commissioner investigate denial spikes. Georgia permits AI to automate tasks and support decisions but bars adverse determination without a licensed provider's review. A KFF brief found these statutes converge on human review of denials, individualized clinical grounds, disclosure of AI use, periodic accuracy review, privacy limits, auditability, and anti-discrimination language. Holland & Knight reduced the body of law to one principle: "AI may assist and streamline insurance operations, but a health insurer cannot rely upon it as the sole basis for denying care."
On the provider side, Illinois barred AI from providing therapy or making treatment decisions in licensed behavioral health care (August 2025). Nevada bars providers from using AI in direct patient care while allowing administrative support with human review (effective July 1, 2025). Maine allows licensed professionals to use AI for scheduling, billing, and, with documented client consent, record-keeping and analysis of session notes, but not for treatment decisions, therapeutic communications, or independent client interaction. Arizona's Board of Behavioral Health Examiners adopted rules requiring licensees to obtain and document informed consent before delivering any service in which AI records or documents clinical services, effective January 1, 2027.
What auditors demand is a defensible documentation trail: medical necessity supported by observable evidence, internal consistency across the record, clear accountability. The most common reason notes get flagged: insufficient documentation of medical necessity: notes describe what happened but not why it was clinically indicated; treatment goals are vague or unchanged across sessions; progress notes read as social updates rather than clinical documentation; the diagnosis does not clearly connect to the intervention. Notes too similar across sessions draw scrutiny. Notes too polished — reading like textbook entries — draw scrutiny. Notes lacking clinical voice, sounding detached from the clinician's reasoning, read as boilerplate.
A template-first approach enforces completeness by design. If a progress note template requires "Session Duration," it cannot be omitted. If it requires "Medical Necessity," the therapist cannot save without addressing it. If it requires "Functional Impairment," the habit of documenting observable functional impact gets built into every session. The audit-readiness advantage is not intelligence. It is consistency. Consistency across 200-plus session notes per year makes audits predictable rather than panic-inducing.
The most audit-safe workflow: AI generates a draft from session data, templates, or transcription; clinician reviews within 24 hours; clinician edits for accuracy, adding observations, adjusting language, noting specific patient statements; clinician verifies medical necessity language is present and specific; clinician signs. Steps two through five are non-negotiable. An unsigned AI draft is not a clinical note. A signed AI draft the clinician did not read is a liability.
In-house counsel should treat AI governance as a legal and compliance priority, not a technology decision. Business Associate Agreements must address ownership and control of AI-generated output, align retention and deletion with organizational policies, and define breach notification procedures. Some states impose AI-specific consent requirements in provider-patient interactions. "Shadow AI" — employees turning to unapproved consumer tools — can move sensitive information outside monitored systems and turn a compliance lapse into a regulatory investigation. Transcription tools should function as documentation support, not the final clinical record. Human oversight is essential; responsibility for the medical record cannot be delegated to AI. A multidisciplinary oversight structure involving clinical leadership, compliance, information security, health information management, and legal counsel helps ensure governance is shared and concerns escalate before they become enforcement problems.
The compliance puzzle is not a tax on innovation. It is the moat.
Privacy Without Borders
The regulatory map for AI therapy tools has stopped being a single country's problem. A vendor signing up a therapist in Toronto, another in Austin, a third in Montreal now faces three overlapping regimes, each with its own enforcement teeth, its own definition of consent, its own timeline for when rules bite.
In the United States, the baseline remains HIPAA. But the application to AI scribes is stricter than many practices realize. Any vendor that receives, stores, or transmits protected health information on behalf of a practice is a business associate, and a written Business Associate Agreement must exist before the first session is processed. That BAA is not a formality. It obligates the vendor to maintain appropriate security safeguards for electronic PHI, to publish clear policies for how patient audio is stored, retained, and used, and to make those safeguards verifiable. HIPAA compliance is not a government certification; it is a contractual and operational posture every therapy practice must verify before hitting record. Upheal's implementation guide emphasizes that the practice owns the due diligence.
Quebec's Law 25 changed the calculus for any platform with Canadian users. Since September 22, 2023, the Commission d'accès à l'information has held penal authority: five years to initiate proceedings, fines up to C$25 million or 4 percent of worldwide turnover, whichever is greater. The law applies extraterritorially. A San Francisco startup processing data from a single Quebec clinician falls inside the perimeter. Law 25 also mandates privacy impact assessments for high-risk processing, explicit consent for any secondary use, and a right to de-indexation with no direct HIPAA analog. For a platform like Klarify, serving thousands of therapists across five countries, the Quebec threshold is not theoretical; it is a product requirement. The penalty regime in Law 25 — up to 4 percent of global turnover — is designed to make non-compliance existential for venture-backed companies.
Below the federal and provincial layers, a state patchwork hardens. Illinois, Nevada, and Utah have each passed laws restricting or prohibiting AI in mental health care, citing safety, effectiveness, inadequate emotional responsiveness, and threats to privacy. The statutes differ in scope — some target consumer chatbots, others sweep in provider-facing tools — but the trend is unmistakable: states are writing their own guardrails rather than waiting for Congress. A vendor operating nationally now needs a compliance matrix tracking which features are permitted in which jurisdiction, and that matrix changes every legislative session.
Professional bodies have weighed in with operational specificity. The American Psychiatric Association's 2026 position paper states that generative AI and large language models must not use protected health information for any purpose, including model training, without a clear legal basis and explicit, specified opt-in consent. AI tools must publish plain-language data-use notices, retention limits, and deletion pathways. The American Psychological Association's health advisory goes further, calling for "Safe-by-Default" settings where the most protective configuration is the factory default, not an option buried in a menu. It also demands a prohibition on the sale or unapproved commercial use of health data collected through AI interactions, and a right to "mental privacy" guarding emerging data types that can infer emotional state without conscious disclosure.
For provider-facing operating systems, these requirements cascade into product architecture. Data residency becomes a feature flag. Model training pipelines must be segmented by jurisdiction and consent status. Audit logs must satisfy both HIPAA's access-control requirements and Law 25's accountability provisions. Vendors that treat compliance as a checklist will ship faster, until the first enforcement action. Those that build regulatory logic into the platform's core will be the only ones still selling into Quebec, Illinois, and Texas simultaneously when the next wave of rules lands.
Three Lanes, One Winner
The AI therapy market has split into three lanes, and no startup leads all of them. The first lane is AI-guided entry into healthcare, including intake, triage, and assessment, where Limbic leads with more than 650,000 clinical assessments completed and adoption across roughly 45 percent of NHS Talking Therapies services. The second lane supports people already in care: between-session tools such as Limbic Care and Jimini Health's Sage that help patients practice skills and stay engaged between appointments. The third lane sells a direct relationship with an AI, such as Ash, The Path, Youper, Earkick, and Sonia, competing for users who want immediate, private conversations without arranging professional care. That third lane boasts the biggest user counts and the highest regulatory risk.
Consumer chatbots have accumulated staggering reach. Wysa reports more than six million users, one billion AI conversations, and 11 million covered lives across 65-plus countries, backed by 45-plus peer-reviewed publications and an FDA Breakthrough Device designation for a specific chronic-pain CBT product. Replika claims over 10 million users; Character.ai has been valued above $1 billion. The Path reports 50,000 active members and 2.5 million sessions. Ash, backed by $93 million from Slingshot AI, says it has reached around 200,000 people. Youper, Earkick, and Sonia have raised $5.2 million, $1 million, and $500,000 respectively. But the research record for autonomous generative agents remains thin: only 16 percent of LLM studies have undergone clinical efficacy testing, and no generative AI agent is ready to operate fully autonomously in mental health given the range of high-risk scenarios. At three-month follow-up, effect sizes in chatbot studies diminish and often turn nonsignificant. Regulators are noticing. Illinois restricts AI from independently providing therapy. Nevada limits systems presenting themselves as professional mental health providers. Utah requires clear AI disclosure and restricts personal-data use. Slingshot pulled Ash from the United Kingdom after concluding the country lacked a clear route for a wellbeing product behaving increasingly like therapy. Woebot shut down its consumer app in June 2025 after reaching 1.5 million people and raising nearly $100 million. Bloom was discontinued after acquisition. Replika faced FTC complaints. The consumer lane is being forced to choose: stay firmly in wellness, or invest heavily enough to enter regulated care.
Provider-facing operating systems play a different game. Their customer is the clinician, not the patient. Upheal says it saves therapists 6 to 40 hours per week. Limbic Access holds Class IIa UKCA medical-device status, forcing documented risk management, testing, usability, security, and change control that wellness apps skip. Wysa's FDA Breakthrough designation covers a clinician-linked product for chronic musculoskeletal pain with related anxiety or depression. The Path has built a separate crisis protocol that hands off to human help when high risk is detected. Jimini Health's clinician-supervised design fits the direction regulators prefer. Products sold as supervised tools for clinicians have a cleaner path than apps claiming to replace a therapist. Regulation increasingly favors companies that support clinicians over products that quietly present themselves as therapist replacements.
That regulatory tailwind compounds a commercial one. Solo practitioners unknowingly spend nearly $26,000 annually across fragmented operational infrastructure. They lose $1,000 to $2,500 per clinician each month through missed or denied reimbursement as insurers deploy automated infrastructure years ahead of clinicians. The average private-pay session reimburses at roughly $159 versus $111 through insurance. A note takes most therapists 15 minutes; with Klarify it takes about two minutes to review and sign. The platform is designed to help practitioners respond to increasingly automated reimbursement systems with AI infrastructure of their own, including claims preparation, CPT coding optimization, eligibility verification, and denial appeal drafting.
The durable moat is not a friendly chatbot, memory, voice, or a collection of therapy prompts. Clinical evidence, regulated processes, proprietary care data, provider integrations, and trusted distribution are much slower to reproduce. Limbic's cognitive-layer system scored above standalone language models and participating human clinicians across measures of CBT competence; users with the highest exposure had a 52 percent recovery rate versus 33 percent for lower exposure. Enhanced models scored 43 percent higher on the Cognitive Therapy Rating Scale, and clinicians preferred them on core clinical criteria 83 percent of the time. Assessments took 12.7 minutes less, patients were seen five days sooner, and dropout fell 18 percent. Wysa's referral technology reports a 91 percent completion rate and roughly 30 minutes saved per assessment; across 100,000 referrals, half an hour per assessment releases approximately 50,000 staff hours. Researchers estimated the 21-point increase in reliable recovery from Limbic's between-session tool could generate about £228 of value per patient using standard NHS cost assumptions.
General AI will wipe out weak therapy wrappers, but it will struggle to replace companies with clinical evidence, regulated workflows, and healthcare distribution. Dedicated products still win on safety. Consumer AI therapy will remain the easiest way to reach people who have nowhere else to go; the likely winners will give that conversation a safe next step rather than treating it as the entire product. Hybrid models — clinician-supervised chat, biomarker-informed interfaces — may be the most defensible path forward. Investors and users lean into platforms demonstrating transparency, accuracy, and safety alongside innovation. Healthcare systems, universities, and employers increasingly deploy AI mental health tools as part of comprehensive wellness programs. Klarify believes therapy represents one of the clearest early examples of a true vertical AI category: the profession combines high documentation burden, reimbursement complexity, fragmented tooling, regulatory sensitivity, and emotionally intensive human work, creating conditions where AI can dramatically expand practitioner capacity without replacing the practitioner. The next generation of category-defining software companies will own a single professional workflow end to end. AI has changed what a small team can ship, and Klarify's advantage is that it only builds for therapists.
The M&A Surge
Behavioral-health M&A didn't just tick up in 2025; it jumped. Deal volume rose more than 42 percent to 104 publicly announced transactions, up from 73 in 2024 and 83 in 2023, according to Becker's Behavioral Health. The first quarter of 2026 alone recorded 149 health-tech deals, putting the sector on pace to exceed 2025's full-year total of 555, Corum Group data shows. Disclosed deal value in that quarter hit $22.7 billion, already approaching 2025's $30.3 billion total.
| Year | Behavioral Health Deals (Publicly Announced) |
|---|---|
| 2023 | 83 |
| 2024 | 73 |
| 2025 | 104 |
| 2026 YTD (July) | 44 |
Source: Becker's Behavioral Health; Capstone Partners
The buyer mix has shifted. Strategic acquirers accounted for 68.2 percent of sector volume in 2026, per Capstone Partners. Private strategics — not private equity — now drive thesis-driven roll-ups across the fragmented outpatient landscape. PE add-ons, which historically dominated, have been dampened. Twenty-eight active U.S. behavioral-health PE platforms operated across eight sub-segments in 2024–2026, but the exit math has tightened. Optum's 2022 acquisition of Refresh Mental Health from Kelso & Company created the single largest U.S. outpatient platform — 300-plus sites across 37 states, 1,500-plus employees — and effectively removed the top-tier buyer from the sponsor-to-sponsor pool.
"The top-tier outpatient mental-health exit market is structurally smaller than the platform count suggests," the CTA Acquisitions roll-up tracker notes. "If you remove Refresh from the addressable buyer pool, the universe of buyers capable of absorbing a $25M-plus EBITDA platform contracts to LifeStance, Optum itself, regional sponsor-backed platforms not yet at national scale, and a small set of growth-equity buyers focused on virtual-first models."
Three landmark deals in the first half of 2026 illustrate where capital concentrates. United Health Services acquired Talkspace for $835 million in March, at 3.0x EV/Revenue and 26.1x EV/EBITDA, advancing its continuum-of-care goals and expanding payer coverage diversity. Spring Health bought Alma in May, combining an AI-native mental health platform with Alma's provider-network software to attack the continuity-of-care problem. Hims & Hers spent $1.2 billion on Australian digital health company Eucalyptus. Meanwhile, Frazier Healthcare Partners paid $490 million for MatrixCare, New Mountain Capital rolled up Smarter Technologies into a $1.49 billion RCM platform, and Thoreau Group signed a $12 billion agreement for Ensemble Health.
Drivers converge. Companies acquire to expand into new markets, add complementary products, and build broader platform offerings, Fierce Healthcare reported in September. The rise of AI accelerates product development and intensifies competition, prompting build-or-buy decisions. A tougher funding environment pushes venture-backed firms toward consolidation as a path to scale and profitable growth. Rock Health counted 115 digital health acquisitions in H1 2026, above 2025's pace of 199 and well above 2024's 121. Galen Growth found M&A captured 97.6 percent of digital health exits in the same period, 84 exits total.
Regulatory overhang shapes every valuation. The Mental Health Parity Final Rule's enforcement pause in May 2025 stalled the anticipated reimbursement rerate. Mertz Taggart's Q4 2025 report described multiples as "stable" rather than expanding, with a widening gap between premium and distressed assets: a flight to quality. Parity-driven valuation lift remains an option, not a fact, through at least 2026. Two enforcement actions — Cerebral's $3.65 million non-prosecution agreement over Adderall prescribing and Done Global's criminal fraud indictment — permanently raised the underwriting bar for telehealth platforms prescribing controlled substances. SAMHSA's 42 CFR Part 2 final rule took active enforcement effect in February 2026; buyers now run pre-LOI Part 2 audits as standard practice because a misconfigured EHR can expose the post-close operator to liability.
The compliance burden itself consolidates. The additional cost of Part 2 compliance is small at platform scale but meaningful at single-site or sub-scale operator level. The rule consolidates the platform advantage. Early-stage companies consolidate horizontally, driven by severe capital concentration and longer timelines between Series A and B raises. Well-capitalized later-stage platforms buy earlier-stage health-tech companies to expand product offerings. Medical front-office AI agents emerge as the next ripe consolidation target.
Investors and executives expect the surge to continue through 2027. Publicly traded and large private healthcare companies will drive more M&A, using public stock as acquisition currency, a backdoor exit for venture-backed firms. Private equity interest remains strong so long as rates stay reasonable. Market valuations have mostly reset to a disciplined baseline, except for a handful of high-profile AI companies still commanding outsized multiples.
The next phase isn't about rolling up clinics. It's about owning the operating system that makes the roll-up auditable, billable, and defensible across jurisdictions. The buyers who understand that will set the price. The ones who don't will pay it.
Working in AI? Zero G Talent tracks the openings: see every open Databricks role, browse AI jobs, openings at Anthropic, and the people building the field.