Healthcare AI

Healthcare AI Clinical Validation Is the New Bar: What Menlo's 2026 Report Means

Every buyer in healthcare wants proof, but not the same proof. Menlo Ventures shows where the AI money is going. We explain what each buyer needs to see.

Ruby Gadelrab  |  Founder and CEO, MDisrupt
November 2026  |  About 15 minute read

Executive summary

Providers and payers need different proof. Providers need clinical validation and a fit with their workflow. Payers need a budget impact analysis and HEOR studies. Menlo Ventures’ 2026: The State of AI in Healthcare (November 16, 2026) shows why this matters now: healthcare AI spend is up 2.5x to $3.7 billion, and buyers are running more head-to-head vendor comparisons.

  1. 1Buyers want healthcare-specific AI, and about half of spend now goes to it.
  2. 2Buying is slower and more rigorous: head-to-head vendor comparisons, longer pilots and fewer platform partners.
  3. 3As AI nears patient care, the open questions are clinical: validation, liability, regulation and reimbursement. Clinicians in the loop, with commercial and regulatory experts alongside, are how builders answer them.
  4. 4Evidation, Dandelion Health, LeanTaaS, Nudj Health, SPARK Neuro and Inceptive, six MDisrupt clients, illustrate how evidence, workflow fit and oversight look in practice, drawn from each company’s own materials.

A healthcare AI vendor can win the provider and still lose the payer, because the two are not asking for the same proof. Menlo Ventures’ new report shows why that matters now. Greg Yap and his co-authors have mapped where healthcare AI money is going, and they describe AI moving from the back office toward the bench and the bedside. We read the report as operators, so we ask what it changes for the people building, buying and validating these products. MDisrupt is the AI-powered health expert marketplace that connects healthtech and life sciences companies with vetted clinicians and health experts. That is the vantage point we write from.

01  The data

What did Menlo’s 2026 report find?

Healthcare AI spend grew 2.5x to $3.7 billion, and about half of it goes to healthcare-specific products. Menlo’s team worked with Morning Consult to survey 550 director-level and above executives in September 2026, across providers, health plans and benefits administrators, and pharma and biotech companies. Four numbers frame the rest of this piece (Menlo Ventures, November 16, 2026).

$3.7Bhealthcare AI spend, up 2.5x year over year
50%of that spend goes to healthcare-specific applications, about twice the share across all enterprises
42%of healthcare organizations have implemented healthcare-specific AI, up from 22% a year earlier
77%of the spend comes from providers, about $2.9B

Healthcare is buying AI built for healthcare.

Healthcare puts about twice the share of its AI spend into industry-specific products than enterprises overall do. In our view, a generic model with a healthcare logo on it will have a hard time in the room. Teams that can show how care is actually delivered have the edge, and teams that can’t will feel it in every evaluation.

The limits of the data

This is a survey of executives plus an estimate of spend. It tells us what buyers say and pay for, not what happens to patients. We use only the headline figures, and we label our own view as ours.

02  The buying cycle

Why is healthcare AI buying getting slower?

Health system buying now takes about 10 months, up from seven, and outpatient providers take about 11 months, up from five. Menlo says providers are returning to traditional procurement: more head-to-head vendor comparisons, longer pilots and closer scrutiny, with buyers consolidating around a smaller set of trusted platform partners.

We read slower buying as a request for proof, and evidence and clinical validation become the bar between vendors. A comparison only works if the buyer can see how each product performs in a setting like theirs, and who validated it. A platform partner is a bet that someone will stand behind the clinical and operational claims for years. Vendors who arrive with that evidence shorten the cycle. Vendors who assemble it during the pilot lengthen it.

Where we build on Menlo’s finding

Menlo concludes that, among providers, distribution advantages currently seem to weigh heavier than product advantages. We agree, and we would add what distribution is made of in healthcare: knowing who sits on the buying committee, which code pays for the product, and which clinical champion has to say yes. Those are expertise problems.

Payers feel the other side of this. Menlo reports that 43% of payers say provider documentation and coding AI raised their 2026 medical cost trend, up from 23% in 2025, and 57% of large payers say the same. When one side of the market adopts AI quickly, the other side asks harder questions about what it is paying for.

Healthcare is one of the only industries where the consumer (the patient), the user (the provider) and the payer (health plans and self-insured employers) are completely different stakeholders, with different incentives and evidence requirements. AI exacerbates this.

Every buyer in healthcare wants proof, but not the same proof. A vendor needs evidence for two different audiences.

  • Payers absorb the cost. Providers buy most of this AI, and payers feel it in claims and prior authorization. Where a product is billed directly, they also decide coverage. They need to see whether the product will save them money. That is where a budget impact analysis and health economics and outcomes research (HEOR) studies come in.
  • Providers decide whether it gets bought and used. They account for 77% of spend in Menlo’s data. They need clinical validation showing the product is safe and effective compared with the standard of care, evidence it fits their workflow, and a clear answer on return: will it save money or make money?

Evidence built for one does not fully carry over to the other, but clinical validation is the shared starting point. It answers two questions: is the product safe, and is it effective compared with the standard of care? Providers need those answers directly.

Payers need them too, and then they ask a third question: will this save us money or make us money? Answering it takes a budget impact analysis and HEOR studies, and both depend on validated clinical results. Different experts build each side. HEOR and market access experts build the payer’s economic case. Clinicians and clinical AI validation experts design the safety and effectiveness evidence providers need. Health system executives and clinical informatics leaders test whether a product fits real workflow. Builders can bring these experts in early, through Health Expert OnDemand, to generate the evidence for both.

In our view, that is why clinical validation becomes the bar. Without it, there is nothing for the economic case to stand on.

03  The shift

What changes when AI moves from paperwork to patients?

When AI moves toward patient care, the open questions become clinical. We see the early wins as scribing and billing, work that is bounded and easy to check. Newer categories such as clinical copilots, patient engagement agents and drug discovery are less bounded. Menlo flags what is still open:

  • Patient-facing agents must handle unpredictable patients and edge cases.
  • Early state pilots run under tightly defined protocols and physician oversight.
  • Reimbursement remains an open question.
  • Liability, regulation and clinical governance are unsettled.

We read every item on that list as a clinical or operating judgment. Someone has to define the edge cases, decide when the agent hands off to a human, and sign off on the evidence a regulator or payer will ask for.

The more clinical the AI, the more clinical the evidence has to be.

Two kinds of expert carry that work.

Role 1: clinical expertise

Clinician in the loop

Practicing specialty or subspecialty clinician who works inside the build, from design through monitoring, so the product stays clinically sound.

  • Defines the training data
  • Labels high-risk edge cases
  • Evaluates outputs for correctness and safety
  • Stays on after launch to validate and monitor
Role 2: commercialization expertise

Health expert on demand

Leader or operator who has taken health products to market, brought in to get the product paid for and approved.

  • Reimbursement and market access leader
  • Regulatory strategist for approval pathways
  • HEOR specialist for budget impact analysis
  • Revenue cycle specialist for billing and coding

Read more in our piece on why healthcare needs its own version of Mercor, built around clinicians in the loop, or, if you are a clinician, start with our guide to working in healthcare AI.

04  In practice

What does evidence-led healthcare AI look like in practice?

Menlo’s numbers tell us what buyers are doing. Six MDisrupt clients show what proof looks like in practice, from fit with clinical workflow to validation a buyer can inspect, human oversight, payer economics and lab evidence. We describe each one from its own materials, so the figures are the companies’ and we have not verified them. This is a small, hand-picked sample, so read it as illustration, not proof.

Menlo finding: distribution weighs heavier than product

Menlo’s distribution point, covered above, is easiest to see in how software gets adopted. LeanTaaS shows what that looks like for health systems. It offers predictive analytics, generative AI and machine learning to manage capacity across operating rooms, infusion centers and inpatient flow, and counts more than 1,200 hospitals and centers, and 14 of the top 25 health systems, as customers.

The detail worth noticing is on its inpatient flow page. iQueue runs on automated data flows from existing systems, so frontline staff do not enter data separately, and Epic stays the system of record. A dedicated engagement team handles implementation, change management and governance alongside the software.

Our take: that is distribution in practice. The software has to disappear into the workflow, and someone has to own the change. Menlo does not break out hospital capacity as its own category, so we use LeanTaaS to illustrate its buying-cycle findings, not to measure a segment.

Menlo finding: well-designed evals can differentiate vendors

Menlo says well-designed evals could differentiate new entrants. Two MDisrupt clients we highlight in this area are Evidation and Dandelion Health. Each makes its validation work visible.

Evidation provides longitudinal, multimodal health data for AI model development and evaluation, and develops models for selected applications. Through direct, ongoing relationships with individuals, the company links wearable signals, survey responses, clinical records and biospecimen-derived data from the same person over time. Researchers can investigate how these signals relate and collect additional information to develop and validate models. Evidation’s work spans dozens of conditions, including rheumatoid arthritis, hypertension, COPD and influenza.

Research on influenza-like illness, published in 2023 with collaborators including the University of Washington, illustrates this approach in practice. The team paired more than 14 million hours of wearable data with symptom reports and ground-truth laboratory PCR influenza test results (PCR is the lab test that confirms influenza) from self-collected nasal swabs. Using an evaluation framework that mimicked real-world model deployment, the researchers compared nine modeling approaches, testing performance on previously unseen individuals and later time periods. The framework also quantified uncertainty when selecting models for a setting where confirmed infections were rare. For AI developers, this work illustrates how linked data can support both model development and rigorous evaluation of how reliably models detect infection across individuals and over time.

Dandelion Health provides clinical intelligence for life sciences, combining multimodal data, clinical AI and scientific expertise. Validation came first. In a 2023 program, it offered developers a free service to evaluate algorithms for performance and bias across racial, ethnic and geographic subgroups, starting with ECG algorithms, on de-identified data covering roughly 10 million U.S. patients. Results went only to the algorithm’s developer.

That work now underpins its Clinical AI Marketplace, launched in September 2024, which checks third-party algorithms for performance and equity before running them on Dandelion’s real-world data. In its launch study, it ran a Pheiron algorithm that predicts cardiovascular risk from ECG waveforms. It estimated that GLP-1 drugs could lower predicted major cardiovascular event risk by 15% to 20% over three years across 44 million previously unstudied U.S. patients, in six weeks and, the company says, at under 1% of the time and cost of a clinical trial. Like Evidation, it states the limit: these are projections of predicted risk, not observed outcomes.

What both show: each company hands the buyer a way to check the model. A vendor that tests a model the way it will be deployed, or labels an estimate as a projection, is giving you the evidence to evaluate it. That is what treating evidence and clinical validation as the bar looks like.

Menlo finding: as AI nears patients, oversight becomes the question

Menlo flags oversight as an open question for patient-facing AI. Nudj Health is an intelligence layer with three AI offerings, one each for clinicians, operations and patients, built on a longitudinal dataset from more than 20,000 patients. It says its care is built so that health systems, employers and payers can measure outcomes. Its homepage states that a human is in the loop at all times, and an outcomes page shows what the care model delivers.

Why it matters: that is the line buyers will ask every patient-facing vendor to write. Saying it is easy. Naming who the human is, what they review and when they can override the model is the work of a clinician in the loop.

Menlo finding: payers react to provider-side AI

As covered above, payers are pushing back on what provider-side AI adds to their costs. SPARK Neuro builds for the payer side. It identifies high-risk plan members early for more proactive, tailored care, and offers Spark Predict, a product built on its claims intelligence engine.

The split: a plan asks whether a product lowers its costs. A health system asks whether it works in clinical workflow. Evidence that persuades one does not automatically persuade the other, so builders need to plan for both.

Menlo finding: newer categories like drug discovery are less bounded

Menlo lists drug discovery among the newer, less bounded categories, and 300 of the 550 executives it surveyed work in pharma and biotech. Inceptive builds foundation models that learn to design sequence-based medicines such as mRNA, siRNA, ASOs and peptides, and partners with drugmakers to customize them for discovery programs. It states that in silico validation is insufficient, and its in-house wet lab ranks models and validates designs in vitro and in vivo.

The difference: in discovery, the proof is physical. A design has to hold up in the lab, and whoever judges that needs bench experience.

The common thread in these examples is evidence a buyer can inspect. We think the strongest versions are built with clinicians and health experts early, before a buyer asks for proof. That is what we see at MDisrupt every day.

05  The playbook

What should builders and buyers do now?

For builders

  1. Name your validators. Put practicing specialists from the right specialty, subspecialty and care setting on the build at design. Write down what they did and when.
  2. Settle who pays, and prove it, before the pilot. Menlo calls reimbursement an open question. Bring in a reimbursement or market access leader early. Plan a budget impact analysis and HEOR work for payers alongside the clinical validation providers expect, so the pilot tests the business model along with the product.
  3. Map the buying committee. Plan for a 10-month health system cycle, a clinical champion, and the platform partners you will have to fit alongside.

Takeaway for builders: put clinicians in the loop on your side, because clinicians will be on the buyer’s side asking the hard questions.

For buyers

  1. Who validated this, and when? Ask for named roles and specialties, and whether they took part at design or only reviewed the finished tool.
  2. Were they practicing in a setting like ours? An emergency department, a community clinic and a payer call center are judged by different people.
  3. Who monitors it after launch? Standards of care and patient populations change. Ask who reviews performance over time and what happens when the model is unsure.

Takeaway for buyers: put clinicians on your evaluation panel and ask every vendor the hard questions.

06  Next steps

How can healthcare AI teams get started?

Start before the model, the contract or the pilot is locked, while expert input can still change the training data, the evaluation and the payment path. With Health Expert In The Loop, that can be one clinician on a labeling sprint or a standing panel for ongoing quality assurance, with reimbursement and regulatory experts alongside when you need them. See our Clinicians in the Loop solutions for the full lifecycle.

Build with experts

Healthcare AI that’s built with clinicians in the loop

For companies

Find clinicians and health experts for your healthcare AI products

Tell us what you are building. We will match you with the right clinicians and health experts by specialty and availability, from our network of 4,000+ vetted healthcare experts.

Start a project → Contact us
For clinicians and health experts

Join our network and shape how healthcare AI products are built

Put your clinical experience to work on healthcare AI. If you want to help shape how it is built, we would like to hear from you.

Become an expert → See why experts join
07  Questions

Frequently asked questions

What is the main finding of Menlo's 2026 State of AI in Healthcare?

Healthcare AI spend grew 2.5x to $3.7 billion, and about half of it goes to healthcare-specific products, roughly double the share across all enterprises. Menlo also reports slower, more rigorous buying and AI moving closer to patient care (Menlo Ventures, November 16, 2026).

Why is healthcare AI procurement getting slower?

Menlo reports that providers are running more head-to-head vendor comparisons and longer pilots, and consolidating on fewer platform partners. Health system procurement rose from about seven to about 10 months. Our read is that buyers are asking for proof, and each needs different proof. Providers want clinical validation that shows a product is safe and effective compared with the standard of care. Payers want a budget impact analysis and HEOR studies that show what it does to cost.

What is a clinician in the loop?

A practicing specialty or subspecialty clinician who works inside the build of a healthcare AI system, from defining training data and labeling to evaluation, validation and ongoing monitoring. Learn more about a clinician in the loop.

What is a health expert in the loop?

A health expert who works inside the build of a healthcare AI system, with checkpoints and the authority to override the model. The clinician in the loop is the most common one. Commercial and regulatory specialists, such as reimbursement and market access leaders, HEOR specialists and regulatory strategists, are health experts on demand. They join through Health Expert OnDemand to get the product paid for, approved and adopted.

What counts as healthcare AI clinical validation?

Healthcare AI clinical validation is the evidence that a healthcare AI product is safe and effective for the patients and care settings it serves, compared with the standard of care. A clinician who only signs off on the finished output is not enough. A clinician in the loop works inside the build, from defining training data and labeling edge cases to evaluating outputs and monitoring after launch. Health experts on demand, such as reimbursement, HEOR and regulatory specialists, build the payment and approval case. Together they give a buyer evidence to inspect.

How do healthcare AI teams get clinicians and health experts involved?

Start a project with MDisrupt through Health Expert In The Loop. Engagements range from one clinician on a labeling sprint to a panel for ongoing quality assurance, with reimbursement and regulatory experts alongside when needed.

08  Sources

Sources

About the author

Ruby Gadelrab is the Founder and CEO of MDisrupt, the AI-powered health expert marketplace that connects healthtech, life sciences, medical device, diagnostics and investor organizations with a curated network of vetted healthcare experts. She has been commercializing health and life science products for over two decades, with leadership roles at Affymetrix, Invitae and 23andMe.

The AI-powered health expert marketplace behind healthcare’s boldest innovationsmdisrupt.com