The Math Doesn't Close the Deal: McKinsey Says AI Can Do a Fifth of Outpatient Care. Your Buyer Isn't Buying That Fifth Yet.

Why this issue exists

One buyer signal. One claim. Tested against what’s actually happening in the market. No panels. No roundups. A claim on trial.

Issue #7 gave us the Shadow RFP: the shortlist forms through trust before procurement ever opens a door. Issue #8 showed that trust being filtered by four committees, not one. Issue #9 showed the trust layer turning into paid, gated membership. Issue #10 put a price on the door: roughly $25,000, the most a champion can sign without needing cover.

This week, the biggest consulting brand in healthcare dropped a number that makes $25K look like pocket change. McKinsey says AI can already perform 16–22% of U.S. outpatient care. Founders are going to paste that stat into every pitch deck by Friday. So let’s put it on trial before you do.

Curb moment: McKinsey just handed every AI founder the biggest TAM slide of the year. The hospital buyer is going to read it as the biggest risk slide of the year.

The Claim: AI Can Already Do 16–22% of U.S. Outpatient Care

In a report dated September 28, 2026, McKinsey analyzed 2024 commercial, Medicare, and Medicaid claims. Its conclusion: 16–22% of all U.S. outpatient claims, roughly 2–3 billion a year and 13–19% of outpatient spending, involve care that "discursive" AI could perform today. Digital Health Wire did the cleanest summary of it. The breakdown matters more than the headline:

11–15% of claims are evaluation and management visits with no or low-complexity care attached.

5–7% of claims are professional interpretation of diagnostic and imaging results.

Almost 60% of the health workforce (11 million of 19 million workers) has a substantial share of core tasks AI can already perform. McKinsey frames this as the first of three vectors: discursive care (intake, triage, guidance, follow-up), then personalized pathways, then autonomous interventions with robotics. The demand case is real: 92.3 million Americans live in primary care shortage areas, and the average appointment wait is 31 days.

Now the line most founders will skip. McKinsey itself says the capabilities exist but many “have not yet been translated into products or deployed at systemwide scale.” And the biggest winner in its model isn’t the health system. It’s the payer.

Curb moment: “AI can do it” and “a hospital will pay you to do it” are two different sentences. Only one of them is in your bank account.

Stress-Test #1: Buyers Already Set the Bar, and It’s 3x

If McKinsey's fifth of outpatient care is a real buying signal, we'd expect hospitals to be loosening the purse strings for AI. They're tightening them. A Bain & Company and KLAS Research survey of 300+ executives, reported by Healthcare Dive on September 24, found that buyers with formal AI ROI thresholds want 3x to 3.9x their money back.

The same survey found something more dangerous for founders. For clinical tasks, more than half of providers would rather have an EHRnative tool with basic or partial functionality than a better third-party product. Translation: "good enough inside Epic" beats "great outside Epic." Providers flip that preference only for governance, risk, and compliance tools.

The report authors put it plainly: early adoption ran on intuition and optimism, and now the bill is coming due.

Curb moment: McKinsey sized the opportunity. The CFO sized the hurdle. Guess whose number decides the deal.

Stress-Test #2: Follow the Money and It Doesn't Lead to the Hospital

Read McKinsey's own payment analysis and it gets uncomfortable fast. AI-powered discursive care threatens low-acuity visit volume for fee-for-service providers. Those organizations would need referrals, follow-up care, and procedures to offset the revenue hit. Value-based care organizations fare better. Payers fare best, with McKinsey suggesting a payer-deployed AI clinician could become the primary way patients access care.

The market is already voting. Per Health Tech Nerds, Devoted Health raised $1.18 billion at a $25 billion valuation and confirmed it is profitable. HTN's Kevin O'Leary calls it the most important privatemarket test of whether an AI-centric approach can bend the cost curve. Devoted is a payer.

So here's the founder trap. If you walk into a fee-for-service hospital pitching "our AI handles 20% of your outpatient visits," you just pitched a 20% cut to their visit revenue. Nobody signs that.

Curb moment: You think you're selling efficiency. A fee-for-service CFO hears "I'm here to shrink your top line."

Stress-Test #3: The AI Winners Are Buying Trust, Not Just Building Models

If raw capability were enough, consumer AI would already own this. Instead, look at who's getting traction. K Health, in a September 28 HTN interview with CEO Allon Bloch, described its shift from a consumer symptom checker to health system partnerships with Mass General Brigham and Hartford HealthCare, backed by a Cedars-Sinai study comparing its AI to physicians.

That's the playbook: capability, then evidence, then a trusted institutional brand on the partnership. McKinsey says the same thing in consultant language. Data availability, not model capability, is the limiting factor, and the winners will be those who can demonstrate consistent, safe, reliable AI care.

And Rock Health dropped the warning label this week: with AI generating a polished GTM plan in an afternoon, it's easy to confuse a snazzy deck with a validated strategy.

Curb moment: AI can write your pitch deck in an afternoon. It can't get you invited to Mass General Brigham.

The Buyer's Side: How Hospitals Actually Read You (and Your Users)

McKinsey wrote for health system CEOs. So this month we asked the buyers directly. Three DHSI guest sessions in September put hospital and practice leaders on the record about how they judge sellers, the information sellers send, and the people who will actually use the product.

They judge your homework in five minutes. Matt Johnson, Chief Administrative Officer of Wake Internal Medicine Consultants (Sept. 24), said he can tell within minutes whether a seller did the work. The tell is the quality of your questions and how well you listen. Nadine Afari, who works across UC Irvine, Rady Children's Health, and Heads Up Partnerships (Sept. 16), was sharper: ask a question answerable from the hospital website and you've told her you didn't prepare. They read fit before features. Kenneth Jones, CEO of HCA Florida Northwest Hospital (Sept. 9), said his first question is how an idea fits work already underway. Pitches outside HCA's core priorities lose traction. And at HCA, purchasing, IT, and supply chain are centralized, so the local CEO influences the deal but doesn't own it.

They treat your follow-up as a test. "Just checking in" and "bumping this up" go straight to the no pile, per both Nadine and Matt. Every contact should answer a question, give something useful, or set a next step. Nadine's standard: a one-pager within 48 hours addressing every question raised and what you learned.

They know the champion isn't the checkbook. Nadine's rule is to map power, influence, and money. An enthusiastic clinician is an advocate. The Value Analysis Committee, IT security, and a CFO judging value after reimbursement are the decision. Kenneth added that cybersecurity clearance can take longer than the purchase decision itself.

They buy for users, not just leaders. This is the part McKinsey's workforce math skips. Nadine said the number one thing buyers look at is workflow, and that a leadership-approved pilot can quietly fade if you don't win the education team, pharmacists, and everyone in the daily process. Matt's practice brings physician owners, office managers, IT, nursing, and front desk into decisions. His rule for any tool: a real problem solved with ROI, without added staff burden or extra clicks.

They want a small, safe door in. Matt's best AI purchase, an AI call center, started on after-hours and weekend calls only, came with a 60-day free trial, and expanded from there. That's Issue #10's $25K door in action. His practice has no digital health budget, yet "if you solve a real problem, we'll find a way to buy it."

Curb moment: Buyers aren't reading your AI benchmark. They're reading whether you studied their workflow, respected their staff, and made their week easier or harder.

The Provocation

Here's where I push back on McKinsey, because that's the job. The 16–22% number is real analysis. But it measures what AI can do, not what any buyer has agreed to pay for. Those are two different markets. McKinsey's own sidebar admits the capabilities haven't been productized or deployed at systemwide scale. That gap, between capability and purchase, is exactly where founders live and die.

The uncomfortable extension: McKinsey's report is less a sales tailwind for founders than a sorting mechanism. Big systems and top academic medical centers, per McKinsey, have the money, data, and brand to build their own. Payers are moving to orchestrate care themselves. Epic is the default for clinical tasks. That leaves founders three viable lanes: sell what the giants can't build, sell where the incentives already reward it (value-based orgs and payers), or plug into someone else's trusted front door. And the thread from Issues #7 through #10 holds. The fifth of outpatient care that AI can do will be captured by whoever buyers already trust to touch their patients, their workflow, and their staff. Capability is the entry ticket. Trust is still the price of admission.

Curb moment: McKinsey proved AI can do the work. It didn't prove anyone will let you do it.

Commercial Corner: What This Means for Your Sales Motion

Never lead with the McKinsey TAM. Lead with one workflow, one user group, one measurable outcome. "AI can do 20% of outpatient care" is a consultant's slide. "We cut after-hours call backlog for your front desk in 60 days" is a purchase order.

Match your pitch to how the buyer gets paid. To a fee-forservice system, frame AI as freeing clinician capacity for highervalue referrals and procedures, never as reducing visits. To value-based organizations and payers, sell earlier detection, closed care gaps, and lower cost of care. Same product, opposite story.

Show the math at 3x before they ask. Build the ROI model with the buyer, not for them, and bring frontline staff into it to surface costs the first estimate misses. If you can't credibly show 3x, shrink the scope until you can.

Have your "why not Epic" answer ready. Most providers will take basic EHR-native functionality over a better third-party tool for clinical work. Win by integrating with the EHR, owning governance and risk, or doing what the native tool can't.

Sell to the users before the signature. Name every person in the daily workflow (nurses, front desk, pharmacists, education team) and show each one what gets easier. A pilot leadership approved but users never adopted is a slow-motion lost deal.

Earn the door with a narrow start. After-hours only. One clinic. One service line. A defined trial window. Then expand. That's how AI gets bought under the $25K ceiling from Issue #10.

Treat every follow-up as a deliverable. One-pager within 48 hours. Every question answered. Zero "just checking in."

Curb moment: Your AI doesn't need to do 20% of outpatient care. It needs to do 100% of one painful job for one team that's tired of doing it.

SigCom Interpretation

Five issues in, the throughline keeps getting louder: digital health commercialization isn't a technology problem. It's a trust-andincentives problem dressed up as one. Issue #7: trust forms before the RFP. Issue #8: four committees filter that trust. Issue #9: the trust layer becomes paid infrastructure. Issue #10: the door has a $25K price tag. Issue #11: even when the capability is proven at national scale, the buyer still asks the same three questions. Does it fit how we get paid? Does it fit how my people work? Do I trust you enough to let you in? McKinsey just told the world the AI wave is here. Good. That means every competitor is about to show up with the same stat. The founders who win this cycle won't be the ones quoting the biggest number. They'll be the ones who walk in already knowing the buyer's workflow, the buyer's incentives, and the name of the front-desk lead whose week they're about to fix.

Curb moment: When everybody has the same AI, the differentiator goes back to being the oldest one in sales: who did their homework, and who do they trust.

Poll: What's Really Stalling Your AI Deal?

The CFO wants ROI I can't prove yet

"We'll wait and see what Epic builds"

Leadership said yes, but end users never adopted

IT security or governance review is eating the timeline

Vote, and tell me in the comments which one cost you the most time this year.

The Lab Note

Next issue: one new voice, one new claim, tested the same way. If you're a hospital buyer: when an AI vendor quotes McKinsey in the first meeting, does that help or hurt their credibility with you? On or off the record.

If you're a founder: are you pitching your AI to fee-for-service systems as capacity or as volume reduction? And what happened when you did?

Drop a comment or send a DM.