The secret to leadership, as the saying goes, is to find a parade and get in front of it. Getting in front of the artificial intelligence (AI) parade — whether that’s AI broadly or generative AI — is certainly a necessary step, but not a sufficient one. Do you know where you will lead this parade? Executives must have a clear sense of direction.
It is likely that you have already taken some steps on your AI journey — crowning an AI czar, experimenting with some prioritized use cases, or choosing an ambient scribe vendor from more than 60 candidates. It is just as possible that your organization is not rushing into things.
Wherever you are on the adoption curve, we believe there are three fundamental choices critical to long-term success of deploying AI solutions. These choices are upstream of all the other decisions surrounding AI.
Choice 1: Efficiency vs. Capability
The full potential of AI will only be unlocked when it is used to advance strategic objectives, rather than simply chasing one use case after another. Healthcare organizations compete to provide better cures, care, and coverage and to do it more efficiently and effectively than the competition. AI has already proven to be efficient in certain tasks — such as the ability to capture clinician-patient conversations, create structured notes and add them to the patient's medical record, and, in a very near future, engage with the insurance company's AI to determine the cost of different treatment options before the visit is over. Enabling better and quick scheduling, both for staff and patients , is another efficiency area where AI excels.
Efficiency is often the first strategic objective that healthcare organizations seek to impact with AI. But disappointment quickly ensues when benefits fail to materialize. AI will not thrive if it is fed bad data and grafted onto erratic workflows. “If you layer AI on rotten processes, you end up with rotten results,” David Lubarsky, M.D., CEO of UC Davis Health noted at the 2024 Oliver Wyman Health Innovation Summit . UC Davis works hard to ensure that it redesigns workflows simultaneous to deploying new technologies, he said. Organizations that fail to standardize operations or that have highly bespoke processes have struggled to get efficiencies out of AI.
The other strategic objective that AI can help organizations meet is improving outcomes by doing healthcare better than any human. AI has shown tremendous potential in multiple areas, including diagnosing illness, predicting health outcomes, recommending an optimal course of treatment, and hastening discovery of new treatments. Dan Hendrycks, an AI pioneer and Executive and Research Director of the Center for AI Safety, describes it this way: “If an AI is better than all humans in all senses of the task, it is Pareto superhuman at the task.”
Evaluating and maximizing superhuman AI is hard for organizations that do not routinely engage in experiment design, the peer review process, clinical trials, data monetization, and technology transfer. Simply put, they don’t have the muscle memory to normalize these activities.
But even organizations that are expert at inventing and testing breakthrough medical technology must be disciplined in their rollout. It’s important to not outpace the high level of trust that’s required from regulators, employees, consumers, and others. Getting buy-in from key stakeholders is a linchpin when introducing any new technology, especially one as disruptive as AI.
Choice 2: Deep vs. Wide
Two AI adoption approaches emerged during sessions and conversations at this year’s Health Innovation Summit. They are very different from each other and yet effective for the organizations that had chosen them. On one hand, Lidia Fonseca, Chief Digital and Technology Officer, Executive Vice President at Pfizer, described the pharmaceutical giant’s approach of pushing AI literacy and adoption — all the way to asking every department head to come up with two AI use cases. Pfizer reports using AI in areas as diverse as discovery, manufacturing, and marketing. At the other end of the spectrum was Kaiser Permanente’s AI chief, Daniel Yang, MD. The integrated delivery system chose to begin implementing just one AI opportunity — but to make it a showcase in employee engagement, safety, precision and learning. The opportunity — AI transcription — has now been rolled out, very carefully, to all of their facilities. Importantly, though, the process it not about IT chasing use cases – rather it is about advancing the operational and strategic goals of the organization.
Which approach is right for you? The key is to exercise control in the right section of the innovation funnel. Your people will experiment with the possibilities of AI whether you like it or not, especially if your organization prizes innovation. Curious team members will look for ways to predict complications, detect fraudulent claims or avoid running out of personal protective equipment. Ideas should be welcomed. However, experience with health technology shows the many risks of uncontrolled implementation such as departments contracting with their own vendors and building their own data environments and employees inadvertently creating security risks that expose sensitive data. AI brings an additional bouquet of risks, with its challenges of alignment, explainability, and bias. Keeping one finger on the fast forward button and another on pause is critical.
Choice 3: Purpose-built vs. One-Stop-Shop
The third choice that healthcare organizations should make early is the choice between different types of AI offerings. Most solutions start with a generic model. Some spend significant time and effort to train and customize the model to perform particularly well in a given clinical subspecialty or functional area — let’s call them purpose-built. However, while such a solution is being purpose-built, a new one-stop-shop frontier model comes out with the potential to outperform many purpose-built solutions out of the box — not just in being more effective or efficient, but in having brand-new capabilities. For example, the latest crop of frontier models is agentic — capable of using other tools autonomously or semi-autonomously. For areas where you try to stand apart from the competition — such as eminence or member experience — you might choose a purpose-built model or even build your own. For overall upskilling and productivity of your people, one-stop-shop is a better choice. And of course, you can cede this decision to one of your major vendors that provides an enterprise platform such as claims or EHR — and let them worry about it.
Making the right choices for your organization
Pursuing efficiency or capability, depth or breadth, going general or specific — leadership teams should begin weighing those decisions now. But choices are rarely binary — most healthcare organizations will end up with hybrids, dialing in the right combination based on the unique strengths and strategic priorities of their organization and market demands. Below are some likely combinations of choices:
- Community health system or local health plan: These organizations would benefit from using the efficiency lens while taking the opportunity to address process gaps in areas like billing. Such an organization could go broad on digital upskilling but should wait for its major technology partner to build in the AI modules by default to minimize the risk and the disruption. In fact, major EHRs, customer relationship management systems, and claims platforms are baking AI into their offerings. At the same time, trying to partner with a large frontier model or experiment with multiple point solutions is likely to be beyond this organization's partner management bandwidth.
- A multiregional hospital network or health plan: Using the efficiency lens until AI regulations are sufficiently clear to allow for a major augmentation of diagnosis and treatment safely is recommended here as well. Go broad with efficiency applications and deep with any capability enhancements. Work with specialized players who have a proven track record.
- Technology-savvy academic medical centers, national provider and payer networks: Since it is easier for them to scale solutions, these organizations could use their experiences in technology transfer, data monetization, and innovation to explore capability enhancements in multiple clinical specialties and business units areas . Such an organization should evaluate a strategic relationship with a major frontier model while continuing to evaluate (and incubate) specialized vendors.
Across industries, we know that leaders are balancing between managing for the here and now and looking for ways to future-proof their organizations. That mindset is critically important when it comes to such transformative technologies as AI. To be successful, leaders must: start with their overarching business strategy, play to your organization’s strengths, and build trust faster than you are building AI.