Our Take
Automation without human touchpoints doesn't fail because the technology is immature—it fails because adherence is a trust problem, not a logistics problem, and trust cannot be automated.
Why it matters
Healthcare leaders are deploying AI to solve medication nonadherence, which costs the U.S. $500 billion annually in avoidable expenses. But the evidence is clear: patients abandon therapy when they feel unsupported or unsafe, not because they lack access or information. The vendors pitching full automation are solving the wrong problem.
Do this week
Healthcare leaders: audit your patient engagement solution by asking what happens when a patient is scared and silent—if the answer is 'the system sends another notification,' replace it with a tool that surfaces at-risk patients for human intervention instead.
AI funding in healthcare outpaced patient acceptance by orders of magnitude
Healthcare AI investment reached $18 billion in 2025 and now accounts for 46% of all healthcare funding (company-reported). Vendors are rapidly deploying AI-driven tools for medication adherence, pharmacy workflow optimization, and patient engagement. Yet independent research finds that 98% of users stop engaging with app-based interventions for chronic disease within a short period or drop usage to ineffective levels (per systematic review cited in source).
The disconnect is stark. A global study found that only 51% of respondents trust their healthcare system to deliver optimal care. When patients report comfort with AI in their own care, the margin is even narrower. Healthcare organizations are building faster than patients are willing to follow.
The barriers to adherence are emotional, not operational
The industry has treated medication nonadherence as a logistics problem: eliminate access barriers, simplify dosing schedules, send timely reminders. But research consistently shows the actual barriers are human. A 2025 systematic review found that patients lacking emotional and instrumental support were significantly more likely to stop therapy. Patients discontinue medication because of fear, social isolation, loss of confidence in treatment, or lack of trust in their provider—not because they forgot or couldn't afford it.
These barriers don't respond to better-timed notifications. They require conversation, context, and sustained human connection. Trust in a healthcare provider correlates strongly with improved adherence. A lack of trust-based communication prevents patients from freely discussing concerns, even when they understand their regimen and can pay for it.
The real lesson comes from outside healthcare. Waymo's February Senate testimony revealed that its supposedly autonomous vehicles rely on remote human operators in the Philippines to guide safety-critical decisions. The headline was outsourcing. The substance was this: billions of dollars and years of development later, human judgment could not be engineered out of the equation. Some decisions are too consequential to make without a person in the loop.
Deploy AI to strengthen human touchpoints, not replace them
AI has genuine roles in patient engagement, but not the ones being pitched as primary solutions. AI can identify which patients are at highest risk of disengaging before they do. It can surface emotional and behavioral patterns buried in thousands of interactions, giving care teams intelligence to intervene earlier. It can automate operational work that frees human supporters to focus on what machines cannot do: build trust, listen, and respond to fear.
Healthcare organizations excelling at this are not defined by how much they've automated. They are defined by how deliberately they've protected human touchpoints that drive trust, engagement, and outcomes. They choose technology that strengthens those touchpoints rather than eliminate them.
The harder question for leaders belongs on the receiving end of the next automation pitch: What does this solution do when a patient is scared and doesn't know how to say so? What happens when the technology reaches its limit—who steps in and how quickly? What does engagement look like at six months, not just at launch? In healthcare, nearly every decision qualifies as too consequential to make without a human in the loop.