306% usage growth
High-impact emails, powered by AI
Platform: Similarweb Sales Intelligence Solution
One agent, wired into 3 flows reps live in.
Results First
The Bottom Line
Usage Growth
Agent usage grew 306% over 3 months, driven by a single insight about how top performing users actually worked.
Speed & Quality
Reps using the agent write 3x faster, with better reply rates than manual outreach.
Canvas-Based
An AI agent in a canvas workspace, keeping inputs, research, and the email together and editable.
Workflow Integration
Embedded the agent directly into Gmail, letting users generate an insight rich email in one click, without leaving their inbox.
PERSONALIZED, INSIGHT-DRIVEN OUTREACH
Reps were losing real hours each week to research, digging into a prospect's company, finding an angle, and drafting something personal enough to earn a reply, and the emails still read generic enough to be ignored.
My task was design an agent that owns the research and writes outreach that converts.
FROM BLANK PAGE TO BUILT EMAIl
Rather than a linear chat, I designed the agent to live in a canvas layout, a workspace where the user's inputs, the AI's research, and the resulting email all exist together and stay editable.
The user
Defines the recipient's details - name and company, giving the agent everything it needs to personalize the outreach.
The agent
Searches the recipient's company data, CRM context, and role in real time, then turns it into a tailored chart- one insight that makes the email visually stand out in a crowded inbox.
FINDING WHAT DRIVES ADOPTION
The insight that changed the design
Users who iterated with the agent in chat got better emails and kept coming back. Users who let it run once with no user in the loop didn't like the results and walked away.
We made chat the canvas's primary interaction, with quick prompts pulling users from a passive first draft to active refinement.
Bringing it into Gmail
Once we understood the feature's value, we brought it directly into the user's inbox.
From Gmail, users can now trigger the agent and get a researched, chart backed email in a single click, without leaving thier inbox. Meeting the workflow where it already lived, and driving another significant jump in usage.
My Reflection
Design behavior that works
The biggest lesson from this project was that giving users a fast, passive result isn't always the win it looks like. The agent's first version made it easy to disengage (one click, done) but the data showed that was actually the weaker path. Redesigning the canvas to pull people into iterating, even though it asked more of them, is what made the product both better and stickier. Success here wasn't just about generating an email faster, it was about designing the interaction that got users to a better email as well. It's a pattern I'm watching for in every AI feature now - where the easy path and the right path might not be the same thing.
