Retention fell after a redesign of voice assistant flow. Investigate the issue

Problem Statement Description

A voice assistant experience used by healthcare coordinators was recently redesigned to improve hands-free task completion. After launch, retention declined, raising concern that the new flow may be interrupting coordinator workflows, reducing trust, or causing users to abandon voice-based interactions in favor of manual alternatives.

In this RCA interview, you are expected to investigate the retention drop as a product leader. The product context involves time-sensitive healthcare coordination tasks where users may be multitasking, operating in noisy environments, handling sensitive information, and relying on predictable assistant behavior to complete actions accurately.

You should frame the anomaly clearly, define what “retention fell” means, determine whether the issue is real or measurement-related, and identify which user journeys, cohorts, devices, environments, or workflow steps may be contributing to the decline. The investigation should balance user experience, AI/voice recognition quality, accessibility, privacy expectations, and operational reliability.

The experience should consider:

- How retention is defined, measured, and compared before vs. after the redesign

- Whether the decline is isolated to healthcare coordinators or visible across broader voice assistant users

- Segmentation by task type, coordinator role, device, language/accent, environment, and workflow complexity

- Instrumentation checks for event logging, session stitching, attribution, and funnel step changes introduced by the redesign

- Hypotheses around friction in the new voice flow, including question clarity, confirmation steps, error handling, latency, interruptions, or failed hands-free completion

- Evidence needed from quantitative data, session traces, voice intent recognition logs, customer support tickets, and qualitative user feedback

- Immediate mitigations, rollback considerations, and longer-term prevention mechanisms

- Guardrails around patient privacy, responsible AI behavior, accessibility, and trust in healthcare workflows

The goal is to demonstrate how you would lead a structured root-cause investigation, separate correlation from causation, prioritize the most plausible failure modes, and recommend a path to restore retention without compromising safety, accuracy, or user trust.