Multi-agent architecture for patient engagement across conversational channels

Health systems and populations have expanded rapidly, but the number of human caregivers including doctors, nurses and allied staff has not scaled proportionally. As a result many clinicians are overwhelmed by paperwork, administrative duties, continuous form-filling, record updates, billing and bureaucracy. They often feel less like healers and more like clerks constantly pushing data instead of delivering care. Traditional digital intake forms and static patient-engagement tools attempt to reduce this burden but frequently fall short because of low adherence and fragmented data.This study evaluates the impact of shifting from passive enrollment through static intake forms to a dynamic multi-modal conversational framework, Hana, an integrated conversational health-agent system that leverages AI to conduct multi-turn interactions with patients. We used a two-phase design with a multinational cohort of 1000 patients. Phase 1 established baseline engagement with a detailed digital intake form plus standard templated reminders delivered via an electronic health record system. We quantified non-completion rates, fragmented engagement (time spent across multiple sessions), and data sufficiency obtained via the passive static intake method. One week later Phase 2 transitioned these same patients to the Hana conversational flow. The system deployed SMS scheduling and up to three short intake calls (or chat/voice-based interactions) performing multi-turn dialogues to clarify missing information. Hana’s architecture used a Data Science agent to analyze Phase 1 data so that the Health Coach agent could avoid redundant questions, thereby adhering to the principle of low user burden (P3).Results show that Hana’s interactive conversational approach significantly increased data completeness and patient adherence compared to the static intake baseline. The final pre-assessment reports generated by a specialized Domain Expert agent, which integrated and reasoned over diverse multimodal personal health data collected across both phases, were rated by clinical experts as superior in both comprehensiveness (93% preference) and clinical significance with high usefulness for clinical care. This work demonstrates that a specialized integrated AI architecture that proactively engages patients in conversation not only automates data collection but dramatically improves data quality. It addresses a critical bottleneck in digital patient intake and supports the potential for conversational health-agent systems to complement, not replace, final assessment by the consulting therapist or psychologist.