SPEAKER:
Amy Bucher, Chief Behavioral Officer, Lirio
KEY TAKEAWAYS:
• Traditional awareness campaigns fail without addressing capability, opportunity, and motivation barriers
• Agentic AI coordinates multiple behaviors achieving 25-29% clinical goal attainment rates
• Patient barriers evolve requiring adaptive systems beyond fixed content sequences
• Precision nudging generates proprietary behavioral intelligence for competitive advantage
• Modular content architecture enables regulatory-compliant personalization at scale
Pharma companies excel at generating patient awareness but struggle to convert that awareness into sustained therapy adherence—a disconnect that costs the industry billions while compromising patient outcomes. At Pharma Customer Engagement USA in Philadelphia, Amy Bucher, Chief Behavioral Officer at Lirio, challenged this traditional model. "Awareness does not equal action," Bucher explained. "We think about the patient's journey through therapy. We think about traditional DTC campaigns, the advertising spend that we make, and then the cost of non-adherence when the math isn't matching."
Her solution represents a fundamental shift in patient engagement strategy. "We are all about personalizing the way that you talk to patients to activate and empower them by using behavioral science and agentic AI," Bucher said. The week of the conference, Lirio announced its entry into life sciences, bringing a precision nudging platform previously deployed in provider, payer, and renewable energy sectors directly to pharma patient engagement.
The Science of Behavior Change at Scale
Lirio's approach begins with the COM-B behavioral model, a framework that categorizes patient barriers into three domains: Capability (physical and psychological ability), Opportunity (social and physical environment), and Motivation (goals and mental models). This systematic classification enables Bucher's team to move beyond generic education to targeted interventions addressing specific obstacles. The scale of available interventions proves substantial. "There are 93 techniques in the behavior change technique taxonomy," Bucher noted. "That's where my team really gets involved and starts to design out how we do this."
The content creation process follows a structured methodology: identify behavioral determinants through patient research, select appropriate behavior change techniques from the taxonomy, develop nudge assets consisting of microcopy building blocks, conduct user testing with patients and providers, and encode content for AI-driven selection. Bucher's team includes behavioral scientists, content developers, and visual designers who collaborate through workshops to ensure interventions match evidence to patient needs. This disciplined approach proves essential when personalizing at scale—ensuring the right ingredients exist in the content mix even for relatively uncommon barrier profiles.
The distinction matters because patient barriers extend beyond knowledge gaps. A patient prescribed a GLP-1 medication may face injection technique concerns (capability), medication access or social stigma (opportunity), or cost-benefit calculations (motivation). Traditional education addresses only the first category. Bucher described how the technology selects appropriate interventions:
"We're using something called reinforcement learning based on the objective—for us, it might be medication adherence. It is selecting the right bits of microcopy for each individual based first on what they look like, and then increasingly over time, how is this person themselves responding."
This dynamic matching process enables the platform to address capability barriers with technique training, opportunity barriers with access support, and motivation barriers with goal alignment strategies. The system learns which behavioral techniques work for specific patients in specific contexts, then adapts its approach as those patients evolve.
Agentic AI Coordinates Complex Patient Journeys
Bucher distinguished Lirio's technology from the generative AI tools proliferating across healthcare, emphasizing that their reinforcement learning approach optimizes toward specified outcomes rather than generating content. She describes this as agentic or goal-driven AI—a term Meta popularized earlier in 2024—that automatically determines which strategies to prioritize for each individual working toward a shared objective. Using the analogy of coordinating transportation modes, layover durations, and cost optimization to travel from New York to Paris, Bucher illustrated how medication adherence requires similar multi-behavior coordination: scheduling provider visits, refilling prescriptions, mastering injection technique, measuring biometrics at home, and implementing complementary lifestyle changes.
The practical implementation begins at a specific trigger point. "We would typically start nudging somebody around enrollment," Bucher explained. "Something like filling that first prescription—that's typically a signal that it's time to begin nudging somebody." Messaging then adapts based on subsequent behaviors and barriers. The platform integrates with compatible devices including Bluetooth blood pressure monitors, smart scales, and smart medication bottle caps, enabling context-aware nudging that avoids unnecessary reminders when patients have already taken action. A reminder sent to a patient whose smart bottle cap already transmitted medication-taking data that morning represents wasted communication and potential irritation.
The bidirectional communication capability proves particularly valuable for generating behavioral intelligence. "When people text us back or text us, that's information that we can use either to respond and model what we've developed and send them back a response that guides them to the next step, or it's something that we can use to learn from and improve our systems," Bucher said. Patient responses—whether answering questions about medication adherence or texting about barriers—create data the system uses both for immediate personalized responses and for ongoing learning.
This interaction model produces what Bucher calls a Large Behavior Model capturing which behavioral techniques work for specific individuals in specific contexts—insights unavailable through claims data or electronic health records. The system learns not just what patients do, but why they do it and what interventions change their behavior. For pharma companies seeking to understand patient support program effectiveness beyond process metrics, this behavioral data represents a new category of evidence.
Clinical Outcomes Across Therapeutic Categories
Lirio's pilot programs demonstrate impact on clinical outcomes rather than just process metrics like message open rates or click-through percentages. Results from an 18-month diabetes care pilot illustrated the platform's ability to re-engage disengaged patients.
"We were able to get 60% of this population to come in for at least one PCP visit after years," Bucher reported. "These were people on value-based care contracts, so their organization is accountable for their outcomes."
These patients hadn't visited their primary care provider despite being covered under arrangements where their health outcomes affected organizational reimbursement—and despite sophisticated provider organizations trying and failing to reach them through other channels.
Vaccination programs showed similarly strong results against rigorous benchmarks. "We're seeing a 17.3% lift over active control," Bucher said. "Active control in this case, by the way, means people get standard pharmacy outreach. Our particular retail pharmacy partner is quite digitally savvy. Their normal outreach is pretty dang good." The comparison wasn't against basic reminder messages but against digitally sophisticated pharmacy outreach already considered effective. Beating that benchmark by nearly one-fifth suggests the behavioral science approach delivers meaningful incremental value beyond current best practices.
The most compelling evidence for pharma applications comes from chronic condition management pilots. "Today, with our hypertension and diabetes pilots, 25% of patients with hypertension and 29% with diabetes are reaching their clinical goals or getting to target," Bucher shared. These clinical goal attainment rates demonstrate that precision nudging doesn't simply increase patient engagement—it drives the biological outcomes that matter to payers, providers, and patients.
The percentage of patients showing significant improvements in these metrics has proven substantial across multiple pilots. For pharma companies navigating outcomes-based contracting and formulary negotiations, this evidence suggests patient support programs built on behavioral AI could strengthen value propositions beyond the molecule itself. Demonstrating that a therapy plus support program achieves superior real-world outcomes compared to therapy alone creates differentiation in competitive therapeutic categories.
Regulatory Realities and Market Implications
The regulatory complexity of pharma patient communication represents a significant implementation barrier for innovative engagement technologies. Many digital health vendors underestimate the rigor of medical-legal-regulatory review processes, creating friction that prevents scaling despite promising pilot results. Bucher addressed this directly when questioned about pharma-specific challenges. "Despite having—we do build our content—we're able to provide fairly manageable selections of content that are appropriately referenced and organized for our review cycles," Bucher explained.
"We have an expert content development team to work through healthcare and legal and medical review processes." The company's structured approach anticipates regulatory requirements: manageable content selections with appropriate references, expert teams experienced with healthcare, legal, and medical review workflows, and proactive identification of regulatory considerations to avoid surprises during approval cycles.
This MLR-ready architecture proves particularly important for personalization at scale, where content volume could otherwise overwhelm traditional review processes. The company holds HITRUST certification and SOC 2 Type II compliance for data privacy and security—table stakes for handling patient health information but evidence of infrastructure investment beyond the AI algorithms themselves. Understanding what requires careful regulatory consideration versus what can move quickly through review cycles represents institutional knowledge that technology-focused vendors often lack.
Bucher identified chronic condition management—particularly diabetes, hypertension, weight loss, and chronic kidney disease—as strong pharma use cases given their pharma components, lifestyle behavior interactions, and adherence challenges. "I think chronic condition management—so diabetes, hypertension, weight loss, and chronic kidney disease—are all really good for multiple reasons," Bucher said. "So first of all, obviously they have a pharmaceutical component to them."
Cancer care screening and treatment pathway adherence represent additional opportunities where the platform has demonstrated effectiveness in non-pharma contexts. The patient journey in these therapeutic areas involves multiple coordinated behaviors sustained over months or years, creating ideal conditions for behavioral AI that adapts as patient needs evolve.
As pharma companies seek to differentiate patient services in competitive therapeutic categories and build evidence for outcomes-based contracting, behavioral AI platforms that coordinate complex patient journeys while navigating regulatory requirements may shift from innovation experiments to commercial imperatives. The question becomes not whether to personalize patient support, but how to do so with both clinical rigor and compliance discipline.
Traditional DTC campaigns generate awareness efficiently, but the adherence gap—the space between awareness and sustained action—represents both the industry's greatest challenge and its greatest opportunity for impact. Closing that gap requires understanding not just what patients should do, but why they aren't doing it, and how to support them through barriers that change over time. That's the promise of precision nudging powered by behavioral science and agentic AI.
To get you highlights of Pharma Customer Engagement USA faster, we are using generative AI technology to summarise the transcripts of the sessions. If you have any feedback about the summary, please contact lucy.fisher@thomsonreuters.com.
Discover more on this topic at Pharma USA 2026 (March 17-18, Philadelphia) - North America's largest cross-functional pharma gathering. Visit the website here.
