The AI redesign gap in healthcare

Health system administrators have high hopes for AI. Nearly two-thirds expect to achieve meaningful, quantifiable returns from their AI initiatives within three years, and they’re putting big money toward the technology. More than 60% of administrators dedicate between 5% and 25% of their IT spend to AI.

Despite this heavy investment, only 40% of administrators report being able to clearly measure the return on their AI investments. Cost, quality, and productivity improvements are currently the top metrics used to track ROI.

However, actual gains have been uneven. Approximately 15% of hospital administrators report capturing cost savings between 10% and 20% from AI initiatives, but most have realized thinner results (see Figure 1).

Figure 1

Three issues are limiting AI returns. First, most health systems are adopting AI without doing the difficult work of changing underlying workflows. More than half of administrators say their organizations are adding on tools to solve isolated problems, while only 10% have redesigned core workflows end-to-end with AI embedded. 

Even fewer have fully linked AI-driven workflow changes to workforce planning (see Figure 2).

Figure 2

Second, the data foundations needed to effectively deploy AI are still underdeveloped in most organizations. Only 15% of clinicians report having access to the data they need to deliver optimal care, while the remaining majority say several data gaps affect their ability to do so (see Figure 3). Fewer than one-third of organizations have a well-defined and resourced technology roadmap for closing those gaps. In most cases, new AI capabilities are being layered onto shaky data foundations.

Figure 3

Lastly, multiple barriers to value creation remain unaddressed. Regarding the top barriers preventing AI from delivering frontline value, clinicians point to a lack of trust in AI’s accuracy and reliability, unclear accountability when something goes wrong, and a lack of clinician involvement in AI decisions.

Governance and clinician engagement are just as critical to success as deeply redesigned workflows and clean data foundations. Nearly half of the organizations actively trying to scale find it slower and harder than anticipated (see Figure 4). They’ll need to address each of the above-mentioned barriers to move from proofs of concept and pilots to full-scale AI deployment.

Figure 4