OVERVIEW

One of the core values Sybill AI offers is to draft follow up emails for account executives after every sales meeting, which is a typical part of the sales process. Despite a huge time-saver, few users engaged with this feature. The team decided to understand why more people are not using this feature. We aim to boost activation, adoption, and user satisfaction.

I led the team in diagnosing and solving the problem, significantly improved user experience that resulted in higher adoption and happier customers.

• My role
• Team
• Duration

IMPACT

<aside> 🎉 14% increase in feature activation Weekly unique users

</aside>

<aside> 🎉 Over half of the emails sent with minimal or no edits The new experience produces more personal, context-rich emails through a streamlined flow that makes sending effortless

</aside>

<aside> 🎉 27% increase in feature adoption Daily # of total events

</aside>

<aside> 🎉

Increased AE workflow efficiency

Reduced time taken on writing emails from 15-20min → 2-5 min

</aside>

DISCOVERY

User needs are not fully met

To diagnose the reason of the low adoption, I gathered feedback through multi channels

• 6 one-on-one interviews talking to customers & internal sales team
• Understand user feedback and support tickets about emails with CS manager
• Watched 30+ user session recordings to observe user behavior

The old version

From user research, I identified not only what was breaking in the current flow but also the broader email tasks AEs wanted AI to support. After syncing with the PM, we prioritized fixing the most immediate drop-off issues first—improving the UI/UX and refining AI output—before exploring additional email types and workflows once we had stronger validation.

Other themes & problems discovered from research

Issues that we wanted to tackle first:

<aside> 🤨 Inconsistent AI output quality Great for first-call follow-ups, but too generic for later deal stages —causing drop-off

</aside>

<aside> 😵 Automation was broken Users had to copy-paste into their email app, which slowed them down

</aside>

<aside> 😕 No easy access to editing

Users typically edit AI output 1–2 times before sending, but LLM-powered editing was hard to access and confusing in app

</aside>