Where We Are Seeing Traction

We are seeing AI generate meaningful impact across several areas of healthcare operations. Prior authorization and claims processing — historically among the most labor-intensive workflows in the industry — are being handled end-to-end by multi-agent systems, with reductions in processing time and administrative burden. We are seeing patient-facing contact centers’ wait times fall and costs decline as AI agents handle predictable, high-volume inquiries that don't require human judgment. Ambient scribing has achieved rapid adoption among physicians, with published research documenting reductions in documentation time, increases in patients seen per day, and measurable declines in burnout.⁽¹⁾ In revenue cycle management, agentic systems are beginning to replace manual denials workflows with a growing focus on preventing errors before claims ever reach a payer; 69% of early adopters report fewer denials.⁽²⁾ And in care coordination, AI is enabling organizations to manage patient outreach at a scale and consistency that manual processes cannot match.

A pattern we have observed and worth noting across these examples: much of the labor required in these areas is the result of upstream errors — claims denied because authorization wasn’t secured before service, for example. In our experience, the higher-impact deployments don’t just process that work faster; they move organizations from remediating problems to preventing them, improving the patient experience in the process.

In our view, the common thread across all of these is a shift from AI as a single-task tool to AI as an end-to-end workflow replacement, what some in the field describe as the difference between tools and workforce. Companies moving toward agentic systems, where multiple specialized agents handle discrete steps in sequence with humans engaged only on exceptions, are where we believe the more durable value is being created.

A Framework for Effective Implementation

The companies we believe are executing well on AI in healthcare share a recognizable pattern in their approach. The discipline with which they select, design and deploy matters far more than the size of their engineering team or the sophistication of their models. In our experience, companies that have found measurable impact from AI tend to follow three principles.

1. Focus on workflows, not use cases.

The most common failure mode in healthcare AI is a poorly defined problem. Initiatives that stall tend to start with a technology in search of an application. Initiatives that reach production tend to start with a specific operational or clinical workflow, map it in detail — not to automate it as-is, but to ask how the work itself should be reimagined. In our experience, the stronger initiatives use AI to rethink the job to be done first, then apply the technology to help execute that redesigned workflow at a superior quality, speed and cost profile. We believe the discipline here is to resist what's technically interesting in favor of what's operationally consequential. These initiatives also tend to share a few traits: a clearly defined workflow, a measurable outcome and a credible path to ROI within months rather than years. When evaluating where to start, look for the workflow where the human cost is highest, the process is most repetitive and the outcome is most clearly measurable. Start there, demonstrate the result and build from that foundation. (Read more about Summit’s perspective on disciplined decision-making in the age of AI)

One dimension that is often underappreciated: the person who owns the AI initiative matters as much as the technology itself. In our experience, pilots owned by operations — by the leader accountable for the business outcome — tend to outperform those driven primarily by engineering. When ownership of the outcome and ownership of the initiative are aligned, the criteria for success stay grounded in what actually matters.

2. Build for compliance before you scale.

Healthcare is a regulated environment in ways that meaningfully shape how AI can be deployed. Many healthcare AI deployments encounter compliance issues at the point of scaling because data privacy requirements, audit logging and vendor agreements were treated as a final step rather than addressed from the outset. In our experience, treating compliance as a foundation is one of the more reliable predictors of whether an initiative reaches production and stays there.

3. Be deliberate about build, buy and partner.

Many growth-stage healthcare companies are assembling point solutions across multiple vendors — one for scribing, another for scheduling, another for revenue cycle. Getting AI into production quickly often matters more than getting the architecture perfect from day one, and a multi-vendor approach can be a reasonable way to move fast. The risk emerges when it's left unmanaged. As you build out your AI capabilities, we believe it is worth investing in common infrastructure and data architecture on the back end regardless of how many vendors you use. That foundation serves three aims: it lets you move quickly in a build-and-buy model; it shifts you away from point solutions that patch individual problems toward longer feedback loops that prevent those problems from recurring; and it lets you swap in a better vendor quickly as the market evolves. Think carefully about which partners are strategic — those with whom you are building a long-term capability — versus those serving a near-term need. We believe vendor relationships and integrations designed to be modular will serve you better as the market continues to evolve. Your data are a compounding asset, and your architecture should reflect that.

‍