Momentir

Insurance agents spend a meaningful amount of time rewriting consultation notes and logging next steps into their CRM. Notes are usually free-form, while follow-up scheduling lives in a different flow, which makes omissions easy.

The goal for Momentir was not to make AI look autonomous, but to create a workflow where summaries and follow-up drafts appear in a form agents can quickly review and finalize.

I designed an LLM pipeline that takes consultation notes, summarizes the key context, and extracts suggested follow-up schedules and tasks as structured output. Those results were routed through an approval UI rather than written directly into the CRM.

That let the product reduce repetitive data entry without turning model uncertainty into operational risk. The product experience was shaped around “review quickly and confirm,” not “trust the AI blindly.”

In the pilot, 12 insurance agents used the flow with real consultation notes, and user interviews indicated roughly 30% less daily operational effort. The strongest response came from tasks that were repetitive yet easy to miss, such as note summarization and follow-up scheduling.

What mattered in this project was not just adding AI features, but fitting AI into a production workflow where the human still owns the final decision.