Healthcare organizations are racing to adopt AI while skipping the question that determines whether it works: where in the workflow does it actually belong? The highest-value applications are not diagnostic. They are operational. Agentic systems that reschedule appointments, accelerate revenue-cycle coding, summarize clinical visits, and deliver real-time evidence are producing measurable results. The organizations chasing AI as a clinical decision-maker are chasing what one practitioner calls "big shiny objects."
John Paul Backhouse is Global Head of Data & Advanced Analytics at Sutherland, where he leads Connected Intelligence across healthcare. He is also a Harvard T.H. Chan School of Public Health Fellow and the creator of the Four I's Framework for healthcare AI transformation. Before moving into data leadership, Backhouse spent years as an NHS paramedic specializing in trauma and urban search and rescue.
"AI in the workflow is the true power. Not AI in trying to make clinical decisions," says Backhouse.
Backhouse identifies four operational areas where agentic AI is producing measurable outcomes in healthcare today.
Patients miss appointments for behavioral reasons, not clinical ones. "You can't cancel quick enough, you can't reschedule quick enough, you call a number and the IVR is holding, and you just give up," Backhouse says. A speech-to-speech agentic agent that answers in seconds, holds a natural conversation, and reschedules the appointment removes the friction that caused the no-show in the first place. "The cause of the problem wasn't about data. It was about human behavior and habit. And that's where AI can step in."
Coding, pre-authorization, and claims processing drive revenue, but the traditional RCM process is slow and only affordable at scale for large organizations. Backhouse says smaller providers can now use agentic AI to unlock cash within days through faster validation, coding accuracy, and denial reduction. "Coding equals revenue, revenue equals more investment into the healthcare system."
Backhouse distinguishes current ambient AI from older dictation tools. The new model is a two-way conversation with an agent that takes notes, checks contraindications, documents in structured format, and sends the patient a single email with the visit summary, assessment, actions, and attached educational material. "Before you've even left the room, the summarization is documented to the patient and into the EMR system in one file."
Drawing on his paramedic background, Backhouse describes a stroke assessment where augmented reality glasses recall the patient's previous FAST evaluation and overlay a real-time comparison. "It can play back immediately and record your assessment now and give you the variation, in front of your eyes, while you're talking to the patient." That kind of insight delivery changes the care pathway decision without replacing the clinician's judgment.
Backhouse pushes back hard on organizations that lead with AI adoption before understanding their own readiness.
"Healthcare is renowned for buying something to fix a problem and not fully deploying it," he says. "They'll buy multiple things, half the product is doing one thing and the other half in another product is doing another." AI layered on top of duplicated systems and fragmented data infrastructure inherits those failures.
His prescription is practical: run a maturity assessment using established frameworks, define a problem statement with financial or clinical value, and start small. "Your maturity assessment and preparation shouldn't take much longer than two to four weeks," he says. "If you can save millions, you can fund the next one and the next one by the savings. But if you're investing millions and you've got no output, adoption will fall very quickly."
Backhouse treats AI governance as an ongoing training process, not a one-time policy exercise. Knowledge bases must be curated, continuously updated with the latest research, and scoped to singular subjects to reduce hallucination. "You can over-govern and therefore get nothing out of it," he says. "Governance has to be like onboarding a human being. You give it the rules, the policies and procedures. You expect it to behave. And if it doesn't, it gets adjusted."
Backhouse's VR project for veteran PTSD applies the same philosophy to mental health. The platform uses AI to guide conversations and control therapeutic environments without interfering with the veteran's experience. "Twenty-two veterans a day kill themselves with PTSD," he says. "If I could prevent one by using AI and technology, then I'd be successful." It is the same principle that runs through every use case he describes: AI drives insights and access, not decisions. And the organizations that understand that distinction are the ones producing results.
