I led the design of Ask Avo, a medical AI chatbot built in just three weeks by a 3-person team (PO, engineer, and myself). We scaled it from prototype to full product suite integration by analyzing clinician usage data and iterating on workflow insights. Today, Ask Avo contributes ~18–20% of ARR and serves as a catalyst for a more agile, data-driven product culture across the team.
Role
Sole Designer, Data Analysist
Tags
MVP Launch, Rapid Iteration
Timline
May 2024
Build Fast, Learn Fast
Our initial launch was a minimal MVP with just an input field. We released it as a standalone prototype that opened in a separate browser window to quickly test how clinicians would actually use it. From day one, we tracked real-world interactions and used those insights to guide our next iterations.
What the Data Revealed
We instrumented every chat session in Mixpanel. Early data revealed a polarized adoption pattern: many clinicians churned after just one attempt, while a smaller but highly engaged group became heavy users. These users approached the chatbot in a structured prompt style—typing long, structured instructions and folding the output directly into their workflow.
Recognizing this behavior, we decided to support their approach by adding features that made their workflow simpler and more efficient. Our North Star Metric became minimizing the effort to reach a useful answer.
01
Showing Preset Questions
Analyzing query volume revealed recurring patterns like DDX and Summarize. We added preset shortcuts for one-click access and positioned the cursor in placeholders so clinicians could easily fill in missing keywords. This reduced onboarding friction for first-time users and increased conversion.
02
Saving Custom Prompts
While shortcuts covered simple queries, heavy users still typed long prompts. To meet customization needs, we allowed them to edit and reorder presets, and introduced an EHR-style “.phrase” shortcut to trigger saved queries instantly.
03
Adding Patient Context Modal
Clinicians often added patient details to queries, but custom prompts were built for permanent reuse, not session-specific data. We introduced a patient context modal where details could be entered once and applied throughout a session. It quickly became one of the most adopted features, with 64% of chats now including context.
Scaling Revenue, Shaping Culture
Ask Avo was later integrated into our product suite after validating a fast, experiment-driven approach. It now contributes ~18% of ARR and continues to scale, expanding its impact on both revenue and how we work as a company.
