Jessica Lamb has one of the more unusual vantage points in health care AI. As a McKinsey partner leading QuantumBlack's health care and public sector vertical, the firm's Artificial Intelligence center of excellence, she spends her days in conversations with health insurers, hospital systems, and health care tech companies about what AI can actually do, what it can't, and why the gap between those two things is so often a people problem, not a technology problem.
At Blueprint 2026, she brought survey data, a live agentic AI demo, and a framework for thinking about AI adoption that cuts through the noise. Here's what stood out.
Health Care is the Only Industry With Zero Productivity Gains in 25 Years — and That's Why Everyone Is Showing Up
Lamb opened with a chart that should make anyone in the room uncomfortable: health care is the only major US industry that has seen literally no per-person productivity improvement over the last 25 years. Every other sector has captured gains from technology. Health care hasn't.
This matters for AI in two ways. First, it means the industry lacks the practice and muscle memory of incorporating technology into operations and actually seeing the returns. When you haven't done it in 25 years, you're not going to be naturally good at it in year one of generative AI. Second, it's why everyone outside health care is suddenly so interested in the space. The untapped potential is enormous and visible — and that's driving the flood of AI vendors knocking on health plan doors.
"There really is quite a big opportunity to be had. But there's certainly not a wrong place to go — there is certainly a wrong way to do it."
— Jessica Lamb, Partner, McKinsey & QuantumBlack
The Three Failure Modes She Sees Most — and Why Pilots Aren't the Problem
Lamb was direct about why AI hasn't translated to bottom-line impact yet, despite years of piloting. She named three patterns she sees consistently across the industry:
- An imbalance between horizontal and vertical use cases. Horizontal tools — enterprise ChatGPT, Microsoft Copilot, giving everyone access to a general AI assistant — are valuable for building comfort with the technology. But they don't move the bottom line. As Lamb put it, the biggest winners of horizontal AI adoption to date have been dogs, because if you free up 20 minutes in someone's day, they go for a walk. That's not a bottom-line outcome.
- Being reactive rather than strategic — chasing point solutions that come inbound rather than stepping back and asking about how we should fundamentally do this differently. Real value comes from reimagining a workflow end-to-end, not from stitching together a series of tools that each solve a small piece.
- The operating model gap. Everyone is piloting. No one is short on pilots. What's missing is the organizational structure that bridges business and technology to actually implement things at scale and get them rolled out to the people who need them.
"There are a ton of pilots going on. That's not the problem. The problem is having the operating model to actually bring together business and tech to solve the problems that matter — and do it at scale."
— Jessica Lamb, Partner, McKinsey & QuantumBlack
The Survey Data: A Genuine Inflection Point, but the ROI Tracking Isn't There Yet
McKinsey has been running a health care AI survey every few months since ChatGPT launched. The Q4 2025 results showed several genuine inflection points worth noting.
For the first time, a majority of health care organizations now say they have actually implemented — not just piloted — a generative AI solution. And for the first time, providers are outpacing payers in that rollout, substantially. Most of that is driven by AI scribes and revenue cycle management tools, but they're real deployments at scale.
The scope of where people see potential impact is also expanding. Administrative efficiency has always topped the survey. Now clinical productivity, patient and member engagement, and even quality of care are showing up as real areas of perceived potential — a sign that organizations are getting more comfortable with the technology and seeing applications they hadn't previously considered.
Perhaps most tellingly, integration into existing workflows has surpassed risk as the top barrier for the first time ever. Lamb read this as unambiguously positive: you can only worry about integration if you're actually trying to implement something. The field has moved past theorizing about risk and into the harder, messier reality of actually making these things work inside complex legacy environments.
The one honest gap: ROI tracking. Most organizations with deployed AI solutions report that it's going well — but when asked to quantify it, most cannot. The feeling is positive. The evidence base isn't there yet.
"People are saying we definitely think it's positive — but when you ask them to quantify it, most cannot. Which would imply we're not quite at the level of sophistication of tracking that we need to be as an industry."
— Jessica Lamb, Partner, McKinsey & QuantumBlack
Three Eras — and Where the Real Disruption Actually Lives
Lamb's most forward-looking framing described three overlapping eras for AI in health insurance:
- The first is where we are now: the use case and point solution deployment era. Organizations are pressing the easy button — buying things, trying things, running things in parallel. Point solutions get you some of the way there. They don't get you all the way.
- The second era — the one Lamb thinks is coming quickly — is end-to-end domain reimagination. Rather than optimizing a step in a workflow, organizations pick a domain (claims, credentialing, network contracting, care management) and redesign it entirely around what technology makes possible. This is where the real financial impact unlocks.
- The third era is process disruption — and this is the genuinely disruptive part. Once agentic AI can reliably handle the steps of complex claims adjudication, pulling policy documents, calling providers, reviewing itemized bills, generating 835s, the question becomes: do you need the claims process at all? Do you need the full RCM function? What does real-time settlement look like?
Alongside this, Lamb flagged the emergence of AI-native health insurers — organizations being built from scratch with all of today's technology available. They won't build the processes and functions incumbents have accumulated over decades. That structural pressure, she suggested, will accelerate change for everyone else.
Her advice for the room: invest in domain transformation now, even knowing that process disruption is coming. The experience of actually reimagining a domain end to end is the practice your organization needs before the bigger wave arrives — and it delivers real near-term impact in the meantime.
The Ratio That Changes How You Budget for AI
The single most practically useful number Lamb shared: among organizations that are actually getting impact from AI — across industries, not just health care — for every $1 spent on technology, they spend $4 to $5 on adoption, change management, and workforce integration.
This is the ratio the industry is currently ignoring. The technology is not the bottleneck. For most of what health plans and providers want to do with AI, the tech can do it. What can't be done is the organization — the change management, the upskilling, the integration into workflows that people have been running the same way for 20 years.
When asked about the workforce dimension directly — the fear that employees working on AI pilots are effectively training themselves out of a job — Lamb's answer was clear: that fear is legitimate, it's going to happen regardless of what any individual organization does, and the only responsible path is deliberate reskilling and transparent communication about what the new roles actually look like.
"For every dollar they spend on the technology, they spend four to five dollars on adoption and rollout. This is no longer a tech problem. The tech is good enough to do most of the things you want to do. This is a workforce, culture, and integration challenge."
— Jessica Lamb, Partner, McKinsey & QuantumBlack
The implication for anyone building or buying AI solutions in health care: if your implementation plan doesn't account for the 4:1 change management spend, it's probably not going to work — regardless of how good the technology is.
Want to read more about Blueprint 2026? Find out what three of health care’s sharpest minds had to say about fixing provider data.