AI-powered email assistant for Gmail

Bringing Mindtickle in sellers' flow of work

Platform

B2B, Web, Email extension

My role

Product Designer

Team

1 Product Manager, 1 Engineer, Design manager, TPO

Category

Product Design, UX Research

What I did

Research & data analysis

Interaction design

Stakeholder alignment

AI-assisted ideation

Manali Bapat | Product Designer

Email me

Manali Bapat | Product Designer
Manali Bapat | Product Designer

The challenge

The challenge

Sellers spend a surprising amount of time after every call on two things:

  • Writing the follow-up email

  • Finding the right content to attach.

    Both directly affect whether a deal moves forward.

We already had an extension for Gmail. It could insert assets and create drafts. But reps weren't using it.

0.5

0.5

0.5

shares/seller/month

shares/seller/month

21%

21%

21%

Utilisation of Asset Hub

Utilisation of Asset Hub

Some of the rep behaviours:

  • Copy-pasting email threads into ChatGPT

  • Hunting through Drive folders

  • Saving files locally on their system

What we set out to build

What we set out to build

An AI copilot inside the existing extension that reads the full email thread and handles both jobs at once: drafting the reply and surfacing the right content, without the rep leaving Gmail.

Most B2B products have a tradition of treating delight like a luxury. Something you earn the right to, once the "real" problems are solved. Nice to have. Low priority. Our badge system was a perfect example of this thinking: functional but forgettable. And they were quietly telling every learner who completed a module: we noticed, but we don't really care.

This is the inertia I kept running into: Learners are professionals. They don't need confetti. Keep it serious.

But I think in the past few years, gamification has evolved so much and that is should be domain-agnostic. The dopamine hit of earning something that looks earned is human. And our learners were completing module after module, racking up real effort, only to receive a digital equivalent of a shrug.

Research

Looking at what's happening

Looking at what's happening

Before designing, I pulled a year of Mixpanel data and ran content analysis across 842 prompts from 300 unique users to understand how reps were actually using it.

Most B2B products have a tradition of treating delight like a luxury. Something you earn the right to, once the "real" problems are solved. Nice to have. Low priority. Our badge system was a perfect example of this thinking: functional but forgettable. And they were quietly telling every learner who completed a module: we noticed, but we don't really care.

This is the inertia I kept running into: Learners are professionals. They don't need confetti. Keep it serious.

But I think in the past few years, gamification has evolved so much and that is should be domain-agnostic. The dopamine hit of earning something that looks earned is human. And our learners were completing module after module, racking up real effort, only to receive a digital equivalent of a shrug.

Of the 625 users who clicked to add an asset, only 86 ever opened the copilot panel. Half the people who added an asset never sent a single prompt.

Is AI-Copilot discoverable enough? If yes, is it capable?

When people did engage, they stayed. Engaged users averaged 5.3 prompts per session.

May be the Copilot wasn't being abandoned because it was bad. It could be because it was buried.

71% of all prompts were custom free-text. Suggestion pills, the pre-set shortcuts we offered were largely ignored. "Follow-up" and "Shorten" were the only ones used with any frequency. "Deal Acceleration" got zero clicks.

Reps wanted to put in the effort and write, with copilot help when needed.

Design challenges

Design challenges

Working inside Gmail extension means

  • Limited real estate

  • Living inside someone else's product

  • We were also facilitating two distinct tasks in one surface: drafting and content recommendations without either getting in the way of the other.

Most B2B products have a tradition of treating delight like a luxury. Something you earn the right to, once the "real" problems are solved. Nice to have. Low priority. Our badge system was a perfect example of this thinking: functional but forgettable. And they were quietly telling every learner who completed a module: we noticed, but we don't really care.

This is the inertia I kept running into: Learners are professionals. They don't need confetti. Keep it serious.

But I think in the past few years, gamification has evolved so much and that is should be domain-agnostic. The dopamine hit of earning something that looks earned is human. And our learners were completing module after module, racking up real effort, only to receive a digital equivalent of a shrug.

Iteration

Working fast and together

Working fast and together

Holding on to the previous thought, I went broad with references - games, fitness apps, learning tools, leaderboards.

Most B2B products have a tradition of treating delight like a luxury. Something you earn the right to, once the "real" problems are solved. Nice to have. Low priority. Our badge system was a perfect example of this thinking: functional but forgettable. And they were quietly telling every learner who completed a module: we noticed, but we don't really care.

This is the inertia I kept running into: Learners are professionals. They don't need confetti. Keep it serious.

But I think in the past few years, gamification has evolved so much and that is should be domain-agnostic. The dopamine hit of earning something that looks earned is human. And our learners were completing module after module, racking up real effort, only to receive a digital equivalent of a shrug.

Output generated by Claude. Great as a start - has issues in colors, color consistency, CTA hierarchies and non-thought UX in some places.

Made this low-fi wireframe with Claude Chat - with all UX enhancements and layout changes needed.

Asked Claude Design to implement this in prototype. Work henceforth was aimed at keeping the experience and interface as simple as possible.

There are already so many AI-writing tools, so that cannot be the star offering of ours. Assets are what unique to us. So why not position them at the center and surface AI-recommended assets once extension is launched. 'Draft reply with copilot' could be an option in this step, after the recommended assets.

✅ What worked

  • It saves the steps, time and clicks.

  • Also gives the user value in 1st step itself.

  • And it is consistent with the current extension: letting users add assets via MT directly to email.

⛔️ Cons

Clutters the UI, increases information load

The experience

The experience