A recent white paper from McKinsey & Co. argues that many health systems are investing in AI tools without redesigning the workflows those tools are meant to improve. The consulting firm suggests CEOs lead efforts to connect AI investments directly to business functions such as supply chain and rewire them completely. Healthcare Innovation spoke with McKinsey partner Jessica Lamb about how leaders can ensure that AI tools are improving operations, access, and financial performance.

Lamb leads the healthcare and public sector vertical of QuantumBlack (AI by McKinsey), where she focuses on helping clients leverage advanced analytics techniques, such as machine learning and generative AI, to achieve their goals.

Healthcare Innovation: A few months ago I interviewed Keith Figlioli, who's a venture capitalist at LRVHealth. He told me he had dinner with a bunch of healthcare CEOs and every one of them was thinking about AI and saying that it's not just an IT thing. Keith said that what may distinguish the winners and losers in this space is whether you have an enlightened CEO who's driving a huge portion of this agenda. Is that something that your white paper is stressing — that the CEO needs to take a bigger role in AI? 

Lamb: It is absolutely critical. You need this to be a business-backed process, not a tech-led process. The process of rewiring, which is how you get the impact from these things, requires it to be business-led and technology-enabled, and that starts with the CEO. When we talk about rewiring or re-imagining a domain, the level of change is very big, and that's the kind of thing that oftentimes you only do if you have real CEO backing. I think that is going to be the difference between what we have seen previously, which is a lot of AI point solutions being adopted, and getting to some of the real impact that we know is possible.

HCI: Some big health systems and academic medical health systems are appointing chief AI officers. How does that role fit in? 

Lamb: Someone needs to both drive the technology and help enable the operating model of bringing the technologists together with the folks leading the various areas of business. We often see a model where there's a chief AI officer. Whatever the title, they are essentially transformation officers who are bringing together the right technical skill set with the business and change management to make sure that the level of change we're talking about actually happens.

HCI: Your paper mentions there's a risk of mistaking activity for progress. We interview some health IT leaders who talk about pilot graveyards, because they just launch endless pilots that never really lead to sustained change. What are some ways to avoid that in the AI space?

Lamb: I think one key is prioritizing a few projects, not a lot. Especially if you are capacity-constrained or a smaller system, that could be prioritizing one domain. I would always prioritize one domain over doing a smattering of pilots across the organization. You'll get more impact every time from that. For me, it starts with actually having the prioritization, picking something and having the right mindset and change orientation to think about tackling that prioritized area in a fundamentally different way. If you think about how we do things in healthcare today, largely it has come to be over a long period of time, and maybe it is not the most logical or efficient if you were to think about doing it in a different way from scratch today, so that's the push that I would make.

HCI: I think the white paper recommended that perhaps revenue cycle or supply chain would be good places to start.

Lamb: They are very logical places for people to start. They’re pretty self-contained and have very clear ROI. They also don't touch as much on clinical decisions, which is very helpful, especially if you are just getting started on your journey. There are many ways that AI can and will impact the clinical space as well, but if an organization is early in their journey, it makes sense to get started in a non-clinical area.

HCI: If a CEO has clinicians who are chomping at the bit to do AI projects that they say are going either reduce physician burnout or actually have an impact on patient outcomes, should he or she say, “Hold your horses. We’re going to do revenue cycle first. Wait until next year”? 

Lamb: No. I think if there are proven individual algorithms that can be integrated into the existing workflow, and the business leader is ready to integrate it, that's fine. But that's not going to get you to the big impact and does not take the place of doing the top-down prioritization — picking some areas and really holistically re-imagining them.

HCI: Do CEOs also need to be able to discern which AI projects to scale up and others that aren’t working that should perhaps be sunsetted? 

Lamb: Absolutely. Although I would say if you have made the switch and you are going with a much more top-down domain prioritization view, you won't have hundreds of these pilots going on that you have to decide to scale up or down.

HCI: To redesign the workflows, your paper recommends standing up cross-functional development pods for each domain. Could you describe why that's important, and what they would look like?

Lamb: The most important thing with these pods is that they bring together the business leaders and the technical folks. They need to have both together working hand in hand. This is not a business unit saying they have a problem and throwing it over the fence to to the technical folks. This is people from the business and people who have the technical and design experience actually working together hand in hand all day, every day led by what we call a product owner, ideally from the business, who knows the details of how these processes work today, knows the people, and knows how to make sure that what is being built is going to solve the needs of their customer. You need to have the mindset that you are actually building this process. You’re doing this work in service of solving someone's real problem, not just building technology.

HCI: What should the CEOs understand about the data infrastructure and the governance and setting guard rails around AI? How involved should they be in that, and what do they have to take into consideration there?

Lamb: I think the guard rails need to be set very clearly up front. That should be a process that the CEO is involved in. What do we stand for? What are we comfortable using AI for, and not? What is our philosophy and ethos around AI? That should be very top down. Then they need to make clear that it is everyone's job to ensure that what they are doing is aligned with those guard rails. Of course, you need a whole governance process to monitor and enable and help make the hard calls on certain things.

HCI: The white paper notes that in the early phases of implementation, the potential exists to revert to legacy processes and the status quo at the first sign of obstacles or discomfort. What can leaders do to combat that tendency?

Lamb: I think one is making sure that the ambition is set high enough. Because if you can reach the goal given to you with more incremental approaches, you're not going to rewire, you're not going to go through it. It’s hard, right? So you're only going to push to that level of innovation and rewiring if you feel like you have to in order to get to the ultimate goal. Creating a safe space to experiment with a focus on the ultimate goal is very important. Some things are going to work, some things are not. You cannot stick with something forever just because you started doing it. It has to be OK and not seen as a failure to move on to the next idea.

As the shift happens toward more holistic, domain-oriented work and away from the point solutions and the pilots, I think we're going to start to see real impact, and there's going to be competitive differentiation that starts to come into play. It's going to be really hard to catch up. You don't have to be on the cutting edge. You don't have to be the leading innovator in the space. But if you haven't even dabbled and you haven't tried it out yourself, the learning curve is actually quite steep, so it's going to take a little while to catch up.