← All projects

Follow-up email agent

An AI agent for post-call follow-ups. 190 people tried it with no mandate and no rollout, and 28 still use it every month.

Role
Builder
Timeline
Launched December 2025
Skills & Tools
  • Glean Agent Builder
  • Python
  • pandas
  • matplotlib
190
People who tried it
959
Total runs
28
Monthly active users
12
Weekly active users
From trying it once to using it every week
Tried it at least once
190
Still active monthly
28
Still active weekly
12

Nobody was told to use this, so the people still running it every week are running it because it beats writing the email themselves.

The problem

Sales and account management reps at HubSpot were losing time drafting follow-up emails after calls. The work meant reviewing the transcript, pulling context from past touchpoints, and shaping all of it into something sendable.

What I built

An agent in Glean that ingests call transcripts and meeting notes, researches internal documentation and customer context, and produces a follow-up draft in the users voice with researched links. Good enough to send with light editing.

I built it outside the scope of my actual role, in my free time.

Results

28 monthly active users and 12 weekly active users, from 190 people who have tried it across 959 runs.

No mandate, no formal rollout, no internal marketing, and no executive sponsor. Every bit of that adoption was word of mouth.

What I learned

Sustained use with no mandate is solid proof that we are solving real problems for real teams. Anyone can drive a spike with a launch email, but holding weekly actives months later means the thing is faster and at least as effective when compared to the alternative.

I dogfooded it from launch and tracked output quality as the underlying model changed underneath it, iterating to keep it effective across versions. Afterall, an agent tuned well against one model version doesn’t always stay tuned on the next.

Next time I would try to engage in a more formal rollout and hold accountability for per-run satisfaction, time savings, and editing needs on the drafts.