Role
Lead Product Designer
Team
Product Manager
AI Engineers
Front-end Engineers
@2026
What is Highspot?
Highspot is an AI-powered platform that helps sales teams work smarter by bringing content, training, coaching, and buyer engagement into one place—so teams can learn faster, find the right materials, practice conversations, and close more deals.
Impact
+15%
Pipeline
created
+14%
Average deal size
+16%
Win
rate
+20%
Reps attaining quota
+20%
Content
ROI
What is AI Role Play?
AI Role Play is a training tool that lets sales teams create and practice realistic sales conversations with AI personas, helping reps build skills, and product pitching in a safe environment before talking to real customers.
My Role
Own AI Role Play feature from concept to launch and updates.
Designed end-to-end flows for enablement and reps.
Added features like screen sharing and multi-persona scenario.
Worked closely with PM and engineering to ship iteratively.
Built prototypes to test and communicate ideas.
What my teammates say about me.
My Impact
Delivered AI Role Play from concept to all GA in 3 months.
Enabled 1,000+ enterprise training use cases in 1 weeks of launching and still counting, with more than 60% completion rate.
Contribute to 10% of Highspot profit after launching 2 months.
Contributed to Highspot’s Gartner Magic Quadrant Leader positioning by launching the AI Role Play capability
What our customers say about the product.
Who am I designing for?
Role 1
Rachel Kim
Sales Enablement Manager
"I want to build trainings to help team sell more effectively, with measurable impact."
Role 2
Alissa Hilton
Sales Rep
"I want to improve my relevant skills and readiness, so I can close more deals."
What did I do?
Manager Flow
Rep Flow
Designed a low-friction, one-click entry to start role play with the right context.
Take me there
Created a realistic, immersive call experience that simulates real customer conversations.
Take me there
Designed a feedback page with clear next steps for reps to improve future conversations.
Take me there
How do I know my decisions are successful?
How to balance user needs with business goal?
From our initial desk research, we mapped out general pain points that PMs often face. We now aim to understand which challenges Microsoft PMs encounter most in their day-to-day workflows and tools.
Rachel Kim
Sales Enablement Manager
Company:
B2B SaaS Company (1500 - 2000 Sales reps)
Team:
Data Cloud Service
Responsibility:
Build onboarding programs for new reps
Launch training for new product releases
Coach managers to coach effectively
Track rep progresses and skill gaps
Prove training impact to leadership
Goals:
Faster ramp for new hires
Higher training completion
Consistent manager coaching
More deals won by reps
Programs scaling without more headcount
Alissa Hilton
Sales Rep
Company:
B2B SaaS Company (1500 - 2000 Sales reps)
Team:
Data Cloud Service
Responsibility:
Run discovery calls
2. Complete training assigned by manager
3. Close net-new and expansion deals
4. Maintain accurate deal data in CRM
Goals:
Improved call quality (discovery, objection handling)
Apply training in real conversations
Higher quota attainment
Win rate for net-new and expansion deals
What Rachel Needs:
A scalable way to verify that reps can apply training in real conversations before customer calls.
What Alissa Needs:
To improve real-world readiness to adapt when conversations go off-script.
Business Goal - Enablement Loop
Enablement Manager
Creates Role Play
Justify purchase with goals
Sales Rep
Practices Role Play
Drives Adoption
Enablement Manager
Tracks Team Readiness
Proves training impact
Sales Rep
Applies feedback
Increase revenue
How to create a Role Play?
From our initial desk research, we mapped out general pain points that PMs often face. We now aim to understand which challenges Microsoft PMs encounter most in their day-to-day workflows and tools.
1
Background
I started by mapping the required inputs into a step-by-step form for managers to create role plays:
a
Persona
AI buyer the rep will interact with
b
Conversation Background
Context the AI buyers know
c
Objection or Concern
Key concerns or challenges to simulate
d
Rubric
How the performance will be assessed
e
Selected Content for Presentation
Materials the rep should use
It doesn't look that intimidating until we are in Rachel's shoes:
What Rachel knows
How they look and sound?
What Rachel needs
How many personas?
What company? What roles? What Personalities?
What Rachel needs
What topics should be covered at this deal stage?
What common objections the buyers will raise?
What materials should use to practice?
What Rachel knows
What criteria define success?
Which skills should be demonstrated?
What is the conversation goal?
After all, Rachel feels:
High Cognitive Load
C1
“I kept scroll up to reference information I input earlier..”
C3
“Too many text inputs..I need to reference messaging, MEDDICC, and objections all at once.”
Time Consuming
C2
“Creating a role play feels like building it from scratch every time.”
C4
“By the time I finish creating one scenario, I could’ve coached 3 reps live.”
Difficult to Scale and Reuse
C3
“Every new training initiative feels like starting from zero.”
C2
“I basically have to duplicate and edit everything manually....”
Let's help Rachel feel better!
Problem Framing:
How might we help Rachel translate revenue-critical training goals into scalable, measurable practice that drive rep readiness with less friction and cognitive load?
Hypothesis:
With the right structured inputs, AI can transform a high-level training goal into a scalable, ready-to-run practice scenario — reducing setup time while preserving strategic intent.
Success Metric
Time to create a role-play
Number of Role Play created
Adoption of AI-generated Role Play
2
Turning unstructured information into a scalable patter.
Cost and feasibility were still unclear, so I used rapid AI prototyping to determine: What information must come from Rachel vs what the AI can generate, so I can inform the UX architecture.
Step 1: Testing the feasibility of the hypothesis, and here are my exploration:
Inputs tested
Tested 5 input variations to see how input structure affects scenario quality.
Output evaluated
Copied output from 5 variations and paste it in the system to generate Role Play scenario.
Findings
Minimal inputs produced generic, low-value conversations
Overly detailed prompts made buyers rigid and predictable
Specific, structured paragraphs improves output quality
Rubrics are essential for detailed, actionable, and scorable feedback
The best performing input variation is:
Unstructured information >
Structured input >
Scalable Generation Pattern
Step 2: Designing information architecture based on the my exploration:
Unstructured information >
Structured input >
Scalable Generation Pattern
Step 3: Mapping the steps on the X-axis and the input on the Y-axis into a consistent UI pattern:
Unstructured information >
Structured input >
Scalable Generation Pattern
3
Designing the North Star creation experience
Exploring what the ideal future experience could look like.
3
Translating information architecture into intuitive interfaces using the design system
Grounded the design in how data flows and how users naturally interact.
Step 1
Entering prompt
4
Vibe-coding the demo to align the team on what’s next
As shown, admins can generate a role play, but have no way to preview or validate it before publishing.
Because of that, I introduced the concept of an Edit and Preview state, starting in Figma.
a
Edit
allows users to finalize details of the Role play.
b
Preview
allows users to chat with personas to test the quality and see if it works as expected.
To clearly demonstrate the concept, I vibe-coded a prototype to show it in motion:
After seeing the prototype, the team felt much more confident in the idea, and my PM went ahead to gauge the cost of this.
4
Impact of the my design
We pulled the data from Amplitude to company the activities of the AI-generated flow shown above,
This shows strong adoption and validates both the design direction and our team’s hypothesis — that structured AI generation makes role play creation faster, easier, and more scalable.
We also decided to merge the AI-generated flow with the manual creation flow. This allowed us to reuse the same structure, which reduced engineering cost while keeping the experience consistent.
How to practice a Role Play?
From our initial desk research, we mapped out general pain points that PMs often face. We now aim to understand which challenges Microsoft PMs encounter most in their day-to-day workflows and tools.
1
Background
Participants
20 sales reps (SDR, AE, AM) — US-based
Method
Structured Survey
Topics
Before: How reps prepare for new messaging and calls
During: What makes practice feel realistic and usable
After: What feedback helps them improve and feel ready
Findings
Knowing the goal helps reps set expectations and start with confidence, so clear context must be provided upfront.
Heavy setup discourages practice, so role plays must be pre-assigned with one-click entry.
Because the product is sold across multiple regions, multi-language support is needed
Scores alone are not helpful, so feedback must show observable behaviors and clear next steps.
Hidden criteria reduce trust, so rubrics must be visible and tied to evaluation.
Confidence builds through repetition, so the flow must enable retry without friction.
Translating the research findings into what Alissa wants:
2
Before Role Play
Now we know what information Alissa needs. The question is — how do we present it?
The challenge is to design a role play item page that stays consistent from one to multiple personas, with a clear hierarchy and alignment to the design system. I explored a few directions:
And this is what I eventually landed on:
Scalable
Works seamlessly across single and multiple personas
3 persona
2 persona
1 persona
Consistent
Handles edge cases gracefully, even when information is missing
3
During Role Play
Pre-Call Setup
Help reps feel prepared before the call with the key information they need, plus a simple setup layer to adjust language and preview themselves before getting started.
In-Call Experience
Context On Demand
Fully animating them is costly, so I explored lighter design solutions to reduce tension and make the experience feel more comfortable.
I prototyped simple avatar states, thinking, listening, and speaking, using Claude and Adobe After Effects, to match users’ mental model of AI.
We kept the existing layout and adopted the new behaviors, making the experience feel more interactive and less intimidating without adding technical complexity.
4
After Role Play
In Highspot, we already have a consistent assessment framework that reps and managers are familiar with. It follows a clear structure: context (what’s being assessed), highlights (overall performance and score), and reasoning (a breakdown of why that score was given).
A standardized system used across training products:
to score performance (1–5 scale)
to evaluate content delivery and knowledge
to surface skill & knowledge breakdown
to aggregate data for reporting
Strengths:
Consistent scoring across features
Centralized analytics & reporting
Familiar to reps and managers
Scalable across programs
This is how the framework appears in role play. Hover to explore the different ways this UI falls short:
So following the framework, I redesigned 'highlight' and 'reasoning' sections to better fulfill user needs.
What I updated
Separated skills from presentation
Grouped strengths and areas to improve
Tied feedback to specific timestamps
Highlight (Before)
Reasoning (Before)
What I updated
Tied feedback to specific timestamps
Replaced long reflection text with concise bullet points
Introduced recommended actions to guide next steps
Let's see what the whole experience look like:
What did I learn through out the process?
This was my second time designing a chat-oriented human-AI productivity tool. Since it was an end-to-end experience, I gained many new insights.
1
Power of UX Writing
I used to spend most of my effort mapping user flows into user inputs and system outputs, but I’ve learned that wording matters just as much, especially in conversational design. A single word can change how users perceive and respond.
2
Not to Overcommunicate
It’s easy to assume we should guide users step by step, but what they need, when they need it, how and where it’s presented must be carefully considered. I learned to turn these concerns into interview questions, and user feedback directly shaped my design decisions.
3
Importance Studying User Mental Model
In UX, gap analysis often focuses on what existing products lack, but it’s equally important to understand users’ mental models. With AI-powered tools especially, users develop familiar interaction patterns, and aligning with these reduces cognitive load.
4
Designing with AI
Before designing, I need to define the level of automation we want to provide. In collaborative tools, it’s critical that users feel in control—‘I’m using AI to achieve this,’ not ‘AI is doing it for me.’ Transparency and confirmation inputs help reinforce that sense of ownership.
