2025 • At-Bay • Product Design Lead
AI-Powered Transcription for Broker Meetings
Role
Identified the need and defined the feature end-to-end — from problem framing through product vision, workflow design, and launch. Partnered with PM and engineering to scope the AI transcription pipeline and design how captured data fed into the broader broker intelligence platform.
Outcome
Shifted underwriter behavior around meeting documentation, resulting in:
Richer broker data feeding the platform
Better meeting preparation
Increased visibility for leadership into broker relationships across the team
Team / Stakeholders
Product Manager
Engineering (Tel Aviv)
Underwriting Leadership
An AI transcription tool that captures and structures notes from underwriter-broker meetings — parsing recordings into key insights, follow-ups, summaries, attendees, and escalation flags that feed into the broker intelligence platform.
The Challenge
Broker knowledge lived in individual underwriters' heads. Notes from meetings — when they existed — were sparse, inconsistent, and siloed. Leadership had no visibility into what was being discussed, which relationships were strongest, or where escalations were needed.
Key Decisions
Pushed back on the original manual approach
The PM's requirements doc called for a text field with additional form fields for underwriters to manually capture meeting context. I redirected this before designing anything — based on conversations with underwriters, they were capturing notes on the go or immediately after broker calls, and adding structured fields would feel like extra work. The tool needed to fit their existing behavior, not fight it.
Designed around engineering constraints
Security blockers prevented building transcription directly into the platform for MVP. Rather than waiting, I designed an interim workflow where underwriters recorded notes via Slackbot, which sent the audio to the platform for transcription and structured parsing. This let us ship and validate the concept without waiting for the ideal technical solution.
Structured output over raw transcripts
Rather than dumping a wall of text, the tool parsed transcriptions into key insights, follow-ups, summaries, meeting attendees, and escalation flags — making the data immediately actionable and queryable across the platform.
What informed my decisions
Conversations with underwriters revealed that the problem wasn't unwillingness to take notes — it was that any solution adding steps to their workflow would be ignored. They captured notes on the go, not at a desk filling out forms.
That finding killed the original text-field PM requirement and shifted the design toward passive capture that fit their existing behavior. The engineering and security constraints then shaped the Slackbot MVP — a pragmatic path to validate the concept while the native integration was built.
Results
Reduced friction shifted underwriter behavior — notes went from sparse bullet points to structured, detailed meeting records
New underwriters could get up to speed on broker relationships without starting from scratch
Follow-up accountability became shared across the team rather than siloed with individual underwriters
Captured data fed the broker platform, creating a foundation for the performance insights and strategic planning that came next
Old notes vs new with transcription tool
View other projects
Clinical operations and customer support platform
Customer-facing death claims intake & profile experience
