The organizations seeing real results start with the business outcome, not the technology. Rather than asking where they can use AI, they identify a specific problem they want to solve, whether that's reducing appointment no-shows, automating scheduling, speeding up prior authorization status updates, or improving patient engagement.

Three things separate organizations making meaningful progress.

First, they treat integration as essential. AI that cannot read and write to the EHR, scheduling system, and contact center platform can answer questions, but it cannot fully resolve them.

Second, they take ownership of change management. Staff, patients, and compliance teams should be involved early in the process, not just brought in when it's time to go live.

Third, they view launch as a starting point, not a finish line. Organizations need to continually review conversations, understand what isn't working, and identify opportunities to improve.

My advice is to start with high-volume, lower-complexity opportunities where AI can deliver personalized, human-like interactions. Learn what works, adjust what doesn't, and build from there.

This comes down to understanding the difference between efficiency and effectiveness. Speed and volume are easy to measure, so organizations often default to those metrics. But the better question is whether the patient's issue was actually resolved.

For example, an organization might report that 80% of interactions were successfully handled through AI. But if 60% of those patients call back within seven days for the same reason, the organization hasn't really solved the problem. It has simply moved the work somewhere else while potentially increasing patient frustration.

Organizations should look beyond self-containment rates and measure first-contact resolution, repeat contact rates, patient satisfaction by journey, and downstream outcomes such as fewer abandoned calls.

Ultimately, the real win is redeploying staff to complete high-empathy work, like care coordination and patient outreach, that wasn't getting done before. The goal isn't simply to reduce the number of interactions reaching employees. It's to free up capacity so they can focus on work that creates greater value for patients.

Healthcare is fundamentally different from other industries because we're taking care of people. We don't want to create a barrier between patients and the human support they need. The key is understanding when an interaction should transition from AI to a person.

I look at three categories of triggers:

• Content: Certain topics should route to a person by design. Clinical symptoms, billing disputes, bad news, end-of-life situations, and anything involving patient safety are examples where human involvement is critical.
• Signals: Organizations should pay attention to changes in patient sentiment, repeated questions, requests to speak with someone, and low AI confidence. AI can recognize intent and sentiment, including when a patient's tone suggests frustration or that they need additional support.
• Context: The patient's broader circumstances matter. High-risk patients, complex cases, or individuals who have already tried multiple times to resolve an issue should not be placed in an endless automated loop.

Just as important as knowing when to escalate is making sure the handoff works. Patients should never have to repeat information they've already provided. AI should pass the conversation history, patient intent, and completed verification steps to the human agent so that person can pick up where the interaction left off.

It should also work in both directions. Human feedback should help continuously improve AI, creating a partnership where technology and people learn from one another.

Healthcare has to balance efficiency, finances, and experience at the same time. All three are equally important, and I don't believe they conflict with AI when the technology is applied in the right places.

The goal should be to remove friction and administrative burden, whether that's long hold times, phone tag, confusing portals, or disconnected systems that force patients to navigate multiple departments to resolve one issue.

But AI shouldn't create more distance between an organization and the patients who need care. It should make the experience more connected and personalized.

That requires ongoing oversight rather than treating implementation as a one-time project. Organizations should regularly evaluate accuracy, tone, empathy, escalation decisions, and overall interaction quality. They also need appropriate governance, security, compliance, and human supervision.

Frontline employees are particularly important because they're often the first to recognize where something isn't working. Their feedback, along with patient input and analytics, should continuously inform how AI is monitored and optimized.

Context should drive deployment. Organizations have a tremendous amount of information about their patients, including their history, previous interactions, preferences, and the circumstances surrounding their care. That information can help determine the appropriate level of AI involvement.

I look at three approaches based on the complexity and emotional intensity of the interaction.

Low-stakes, routine transactions, such as changing an address, may be fully automated. More involved requests may benefit from a hybrid approach, where AI assists but a human remains available. High-stakes or emotionally complex situations should be human-led, with AI supporting the employee behind the scenes.

For example, AI can help prepare a human agent by gathering patient history, identifying intent, suggesting next steps, and handling administrative work. That allows the employee to be fully present during a conversation that requires empathy and judgment.

Patients should also feel confident that they can reach a person when they need one. When organizations establish that trust, patients may be more comfortable using AI for routine interactions because they know human support is available when it matters.

I expect AI to shift from simply answering questions to completing tasks. That could include booking and rescheduling appointments, verifying coverage, following up after visits, and proactively reaching out before problems occur.

At the same time, AI will increasingly support human employees in real time, and the distinction between automated and human-supported interactions will continue to evolve.

For healthcare leaders, preparation starts now. Organizations need to build strong data integration, establish governance and compliance practices, and invest in their people.

Frontline staff will increasingly shift from handling routine transactions to managing exceptions and building relationships that require judgment, training, and a deeper understanding of the patient journey.

People aren't going away. Organizations that invest in both their technology infrastructure and their employees will be better positioned to adapt as AI capabilities continue to advance.

Use AI to give people more time for human moments, not to remove them.

Every AI deployment decision should be evaluated against one question: Does this make it easier for the patient to get the care and answers they need?

Are patients having to repeat themselves? Are they being transferred from one interaction to another without resolution? Are we actually making their journey easier, or are we making it more complicated?

The goal is to improve the patient journey, not create additional obstacles. It's not about getting rid of human interaction. It's about bringing people and technology together in a way that allows each to do what they do best.

If we're making the patient journey easier, we're on the right path. If we're making it more complicated, we need to change course.