Five Ways Healthcare Organizations Should Be Working with AI for Better Results

A clinical trial study conducted in late 2024 showed that artificial intelligence (AI) chatbots outperformed human doctors for diagnosing illnesses, even when doctors were aided by an AI chatbot. 

This outcome seems pretty surprising – but there are some takeaways from this study on how humans and AI can work better together in a healthcare setting.  

Here are five key ways health organizations can collaborate better with AI. 

1. AI Must Work with Human Judgment – Not Fight It

When doctors are shown AI predictions but don’t trust them or are overwhelmed by alerts, accuracy suffers. Doctors revert to intuition – even if it’s wrong. AI should instead be used to present insights as supporting evidence, not directives. Executives should also show the “why” behind every recommendation – using citations, trends or similar cases, instead of overwhelming providers with suggestions, so users can build trust gradually. I also suggest using progressive disclosure: Give quick summaries first and allow users to dig deeper only if needed. 

2. Different Tasks Call for Different Divisions of Labor

AI excels at analyzing structured data and identifying patterns. Humans are better at interpreting context and nuances, such as social factors or patient preferences. Instead, let AI do what it’s good at: surfacing overlooked diagnoses, highlighting gaps or identifying population-level trends. Let humans validate, refine and act – either inside their electronic health records (EHRs) or care coordination workflows.

Encourage a “you do, I’ll check” model. For example, Keebler Health drafts a plan or risk insight, and the provider validates and finalizes it. 

3. Trust is Earned

Doctors don’t fully embrace AI unless they repeatedly see that it improves outcomes. Organizations should implement feedback loops so users can confirm when an insight helped – and get better recommendations over time. Showcase clear wins during onboarding. For example, Keebler Health will identify that the AI platform flagged three patients with unaddressed congestive heart failure, with two patients now in care plans. 

4. Start Simple, Then Scale

Models that offload low-risk or routine tasks to AI work best. Examples include AI reading normal X-rays or suggesting initial coding opportunities. It’s important to emphasize early wins during pilots: Start with basic recapture opportunities, quality flags or “rising risk” alerts. Allow organizations to “turn on” more advanced use cases (e.g., longitudinal cost prediction, attribution shift modeling) once trust and utility are built. 

5. Make AI Feel Like a Colleague, Not a Competitor

Doctors are wary when AI feels like it’s there to judge them or replace their clinical intuition. Instead, organizations should use a tone that feels supportive: “Here’s what we found” or “You may want to take a second look” rather than “Missed diagnosis!” It’s also a good idea to build in chat-based interfaces for Accountable Care Organization leaders and clinicians to give them the opportunity to ask questions, such as “Why do you think this patient is high risk?” or “What changed in Dr. Jones’ RAF trend last quarter?”

Healthcare lags behind the fintech and consumer goods segments by a few decades as related to the use of technologies in day-to-day process improvements. Middlemen solutions within the healthcare ecosystem often leverage personnel to solve these issues. With the introduction of large language models (ChatGPT, bots, AI agents, transformers), healthcare can quickly close the gap technically and find significant efficiency gains. There will need to be attention paid to the tension between process evolution and safeguards, but as we see in all other sectors, winners are borne from the efforts to glean optimal outcomes given these challenges. We expect practitioners to adopt these solutions en masse in the next few years. Where will you stand (as a patient, provider or caregiver) on this technology’s adoption journey?