Design Approach

Research

AI Prototype

UI/UX Design

User Testing

Designing AI-Powered Manager Tool

Designing AI-Powered Manager Tool

Attuned is a B2B HR product that helps managers understand their teams motivators and improve work engagement. Its core feature is the Intrinsic Motivator Assessment, which identifies top drivers for each employee and provide managers with guidance on how to align work with those motivators. At the time of this project, Attuned has about 400-500 monthly active users, mostly managers and team leads using the platform to support their daily leadership. In May 2025, I designed new features for AI TalkCoach increasing engagement and conversion rates by 30%.

Problem

About 30-40% of users dropped off after clicking the AI TalkCoach button on the navigation bar.

AI TalkCoach helps managers phrase feedback in ways that align with their team's motivators such as Altruism, Competition, Progress, or Autonomy. Three months after launch, usage had not grown, and the drop-off rate remained stuck in the 30-40% range.

User Research

We interviewed eight managers and uncovered how they struggled with what to type after selecting a team member.

The idea for AI TalkCoach came from company leadership, who were inspired by emerging AI-driven products. The feature was built on the belief that managers often face challenges in communicating effectively with their teams. Customer Success had run training sessions, yet usage numbers stayed flat. I worked with a user researcher to conduct interviews with managers who had tried AI TalkCoach. together we uncovered the following insights:

50% of the managers were unable to understand what AI TalkCoach is and what would happen after selecting a team member's name.

50% of the managers would like AI TalkCoach to consider all the top motivators of a team member rather than just selecting one motivator.

62.5% of the managers would leave in the middle of the conversation because AI TalkCoach asked too many irreverent questions.

75% of the managers would like to adjust the tones of voice of AI TalkCoach messages

37.5% of the managers think the AI-generated message from AI TalkCoach sounds like a robot.

100% of the managers want to use AI TalkCoach for other scenarios that aren't limited to rephrasing negative feedback

50% of the managers wants the AI TalkCoach chat screen to be bigger.

37.5% of the managers have trouble closing the AI TalkCoach screen.

Landscape Analysis

I reviewed leading AI platforms like OpenAI, Microfost Copilot, Claude, and others.

They begin with a blank canvas and rely on suggested task types to give users direction. That pattern helps users know what they can ask the AI to do. I realized AI TalkCoach needed more visible prompts and clearer entry points to guide user behavior.

Conceptualization

The original AI TalkCoach included example prompts, but they were hard to notice.

The navigation bar was crowded and users often overlooked the button. I proposed adding a prominent entry point on the dashboard, replacing a rarely used "Today's Tip" feature. User research told us "Today's Tip" lacked context and relevance. The change would make AI TalkCoach more visible meaningful during users' daily workflows.

AI Prototype

I used an AI-based prototype to test interactions early.

Users focused more on flow than on style elements like color or typography. The prototype allowed them to click anywhere and experience something close to the final behavior. This method helped the team iterate faster.

Moderated Usability Test

All participants preferred the version that included example prompts.

The prompts gave them concrete ideas of what to type next. They asked for more realistic workplace scenarios, such as "encourage more autonomy" or "ask someone to speak up in meetings." They also wanted suggestions that were actionable instead of abstract. They noted that while AI can reformulate their thoughts, they would still express feedback in their own voice. The ideal AI would help them see what they missed, not just tell them exactly what to say.

Iteration

We removed limits on AI TalkCoach so users could ask about any work-related issue, not just feedback rephrasing.

Managers could now ask follow-up questions and customize advice to fit their unique circumstances. I collected common user scenarios from tests and used them to design new example prompts for the updated version. We switched the UI from a pop-up modal to a drawer that opens from the side to give more screen space and improve readability.

Performance

Within three months after the second release, AI TalkCoach usage increased by about 30%

More than half of the users accessed it via the new dashboard entry point. The average number of chats per user rose by 50%.

-12%

In drop-off rate within the first 3 months since the release date

+150%

In average weekly users within the first 3 months since the release date

Performance

Over the past 3 months, there was an increase in market attachment rate by ~2.5%.

Even though the target uplift was about 4%, the ~2.5% in each region was considered to be a success for the new meal upselling strategy. It's also a good start for the future iterations and next steps.

+2.8%

In Market attachment rate in the US within the first 3 months resulted in over $54k in net revenue

+2.5%

In Market attachment rate in Australia within the first 3 months resulted in over $45k in net revenue

AI Prototype

I used an AI-based prototype to test interactions early.

Users focused more on flow than on style elements like color or typography. The prototype allowed them to click anywhere and experience something close to the final behavior. This method helped the team iterate faster.

Moderated Usability Test

All participants preferred the version that included example prompts.

The prompts gave them concrete ideas of what to type next. They asked for more realistic workplace scenarios, such as "encourage more autonomy" or "ask someone to speak up in meetings." They also wanted suggestions that were actionable instead of abstract. They noted that while AI can reformulate their thoughts, they would still express feedback in their own voice. The ideal AI would help them see what they missed, not just tell them exactly what to say.

Landscape Analysis

I reviewed leading AI platforms like OpenAI, Microfost Copilot, Claude, and others.

They begin with a blank canvas and rely on suggested task types to give users direction. That pattern helps users know what they can ask the AI to do. I realized AI TalkCoach needed more visible prompts and clearer entry points to guide user behavior.

User Research

We interviewed eight managers and uncovered how they struggled with what to type after selecting a team member.

The idea for AI TalkCoach came from company leadership, who were inspired by emerging AI-driven products. The feature was built on the belief that managers often face challenges in communicating effectively with their teams. Customer Success had run training sessions, yet usage numbers stayed flat. I worked with a user researcher to conduct interviews with managers who had tried AI TalkCoach. together we uncovered the following insights:

50% of the managers were unable to understand what AI TalkCoach is and what would happen after selecting a team member's name.

50% of the managers would like AI TalkCoach to consider all the top motivators of a team member rather than just selecting one motivator.

62.5% of the managers would leave in the middle of the conversation because AI TalkCoach asked too many irreverent questions.

75% of the managers would like to adjust the tones of voice of AI TalkCoach messages

37.5% of the managers think the AI-generated message from AI TalkCoach sounds like a robot.

100% of the managers want to use AI TalkCoach for other scenarios that aren't limited to rephrasing negative feedback

50% of the managers wants the AI TalkCoach chat screen to be bigger.

37.5% of the managers have trouble closing the AI TalkCoach screen.

Conceptualization

The original AI TalkCoach included example prompts, but they were hard to notice.

The navigation bar was crowded and users often overlooked the button. I proposed adding a prominent entry point on the dashboard, replacing a rarely used "Today's Tip" feature. User research told us "Today's Tip" lacked context and relevance. The change would make AI TalkCoach more visible meaningful during users' daily workflows.

AI Prototype

I used an AI-based prototype to test interactions early.

Users focused more on flow than on style elements like color or typography. The prototype allowed them to click anywhere and experience something close to the final behavior. This method helped the team iterate faster.

Landscape Analysis

I reviewed leading AI platforms like OpenAI, Microfost Copilot, Claude, and others.

They begin with a blank canvas and rely on suggested task types to give users direction. That pattern helps users know what they can ask the AI to do. I realized AI TalkCoach needed more visible prompts and clearer entry points to guide user behavior.

Moderated Usability Test

All participants preferred the version that included example prompts.

The prompts gave them concrete ideas of what to type next. They asked for more realistic workplace scenarios, such as "encourage more autonomy" or "ask someone to speak up in meetings." They also wanted suggestions that were actionable instead of abstract. They noted that while AI can reformulate their thoughts, they would still express feedback in their own voice. The ideal AI would help them see what they missed, not just tell them exactly what to say.

Iteration

We removed limits on AI TalkCoach so users could ask about any work-related issue, not just feedback rephrasing.

Managers could now ask follow-up questions and customize advice to fit their unique circumstances. I collected common user scenarios from tests and used them to design new example prompts for the updated version. We switched the UI from a pop-up modal to a drawer that opens from the side to give more screen space and improve readability.

Performance

Within three months after the second release, AI TalkCoach usage increased by about 30%

More than half of the users accessed it via the new dashboard entry point. The average number of chats per user rose by 50%.

-12%

In drop-off rate within the first 3 months since the release date

+150%

In average weekly users within the first 3 months since the release date

About me

About me