- AI/ML Product / B2B SaaS · Product Design Case Study · 2025-2026
Designing the AI Practice Experience
Led the zero-to-one design of an AI-powered sales practice platform — from public Try Now entry point through live AI simulation to post-session coaching report. Shipped to early access with measurable behavior change.
Role
Lead Product Designer
Team
1 PM · 2 Devs · 1 AI Dev · CEO
Verdict
Shipped — “Behavior-Changing”
- 00 · My Role & Scope
Lead Product Designer
Full end-to-end design ownership from public Try Now entry point through complex real-time AI conversation simulation to the interactive post-session feedback report.
Performance Verdict
SHIPPED — “Behavior-Changing”
Validated via robust quantitative metrics in active early access cohorts. Reps averaged 5x more practice sessions per week compared to traditional coaching methods.
What I Owned
- ✓ Product strategy & market positioning
- ✓ UX/IA, wireframes & high-fidelity flow layouts
- ✓ Scalable dark-mode design system from scratch
- ✓ A/B testing setup for 3 critical product hypotheses
- ✓ In-depth research: 12 sales rep & 6 L&D manager interviews
The Constraints
- ⚠️ No existing design language or pattern library
- ⚠️ Highly unpredictable LLM latency & conversational quality
- ⚠️ Strict 6-month runway to close the first paying customer
- 01 · The Problem
Sales training has not changed in 20 years. Practice is still missing.
“I watch the training video, score 95% on the quiz, then blank completely on my next call.”
GAP 01
Knowledge Does Not Transfer
Reps easily absorb static course content, but freeze and fail to access it during high-pressure, live buyer pushback.
GAP 02
Roleplay Does Not Scale
Live practice requires high-touch 1-on-1 coaching. The average rep receives fewer than two genuine sessions a quarter.
GAP 03
Feedback Arrives Too Late
Quarterly performance reviews flag mistakes made months prior, making behavior correction extremely sluggish.
$15B
Addressable market in sales enablement
87%
Of rep training forgotten within 30 days
- 02 · Field Research
Who we designed for, and what we learned.
Arjun, 28 · Account Executive · Fintech
“I know what I should say. I just freeze when the client pushes back.”
Goals
- • Practice cold calls privately without scrutiny
- • Build robust, automatic objection-handling muscle
- • Get highly specific performance feedback
Frustrations
- • Fear of judgment from manager during practice
- • Generic training content unrelated to local realities
Priya, 42 · L&D Director · Enterprise
“I can build courses all day. Proving behavior change to my VP is where I struggle.”
Goals
- • Deploy highly scalable practice to 200+ reps
- • Clearly prove program ROI and skill improvement
- • Instantly identify skill gaps across the team
Frustrations
- • Zero actual engagement or behavior-change data
- • Traditional coaching creates an extreme time bottleneck
Key Research Insight
Psychological safety is the number 1 barrier. Reps hate roleplaying with managers because admitting failure threatens compensation. AI simulation completely removes fear of judgment.
Research note: 12 in-depth sales rep interviews · 6 enterprise L&D manager interviews · exhaustive competitor analysis · 4 alternative tools audited
- 03 · Design Approach
Three principles that shaped every decision.
01
Commit first, friction second
Get users into the interactive practice loop immediately before asking for any personal details. Auth gates belong at peak motivation, not on entry.
Applied to: Try Now sign-in placement
02
Feedback is coaching, not scoring
Every score or rating must map directly to a behavior the learner can actionably change tomorrow. A simple number without next steps is useless.
Applied to: Post-session feedback report
03
Trust through showing, not telling
Users are highly skeptical of AI roleplay. The simulation interface must cleanly demonstrate conversational depth rather than claiming it on a landing page.
Applied to: Try Now landing hierarchy
- 04 · The Core Loop
From choosing a scenario to receiving a coaching plan.
1
Choose scenario
Browse industry scenarios categorized by distinct sales skill and difficulty level.
2
Meet your persona
Read the contextual background brief, understand target objectives, and constraints.
3
Practice live
Speak directly with the dynamically adaptive AI persona in real time.
4
Get your report
Receive exhaustive timestamped coaching feedback and diagnostic breakdowns.
5
Launch next
Instantly apply one key coaching insight and launch the next adaptive practice session.
The entire loop runs in under 10 minutes. Most active reps complete 3 full sessions per week.
- 05 · A/B Testing & Data
Three experiments that shaped the product.
CTA Copy
+23% LIFT
“Practice Now” vs “Start Roleplay”
Winner: Start Roleplay. The word "Roleplay" signals a structured professional activity with a defined beginning and end, significantly reducing open-ended performance anxiety.
Sign-In Gate Placement
18% ABANDON
Where to place the Auth Wall?
On landing: 67% abandonment. After scenario: 34% abandonment. At configuration: 18% abandonment — winner, gating at peak motivation.
Simulation Screen Chrome
+26PT IMMERSION
Immersive V2 vs Ambient V1
Removed the heavy, brand-forward gradient header for a clean conversation-first layout with a minimal context bar. Resulted in +26pt felt immersion and +18pt "felt heard" scores.
- 06 · Coaching Feedback
Feedback that coaches, not just scores.
We redesigned the post-session report around one question: what can the learner do differently tomorrow?
1
Score with diagnosis
Conversion probability indicators mapped across 4 core sub-dimensions: Knowledge, Competency, Confidence, and Persuasion.
2
Personality archetype
Reps are matched with dynamic named sales archetypes (e.g. Product Explainer, Trusted Advisor) outlining immediate growth vectors.
3
Critical Moments
Highly actionable, timestamped conversational pivots complete with actual transcripts, contextual AI corrections, and projected score gain.
4
60-Sec Coaching Plan
Exactly 2 highly customized, bite-sized tactical drills to complete before hopping on their very next live customer call.
- 07 · Edge Cases & System Resilience
Designing for the moments AI breaks.
Every failure state was treated as a product moment — an opportunity to reassure, recover, or convert.
Decision
Every AI failure state was designed as a trust-building moment. Instead of hiding errors, we made recovery visible and coached users through uncertainty.
Welcome to Trovex.ai
No Courses Assigned Yet — once a course is assigned to you, it will appear here. Your learning journey begins here.
No Match Found
We couldn't find anything that matches your search. Please try different keywords or adjust your filters.
No Practice Courses Yet
Once practice material is assigned, it will appear here.
No Courses Submitted
You haven't submitted any courses yet. Start learning and submit your progress here.
Trade-off
We accepted higher latency on feedback generation to ensure accuracy. Users tolerated 2-3 second delays when the output quality was visibly better.
- 08 · Impact & Outcomes
What shipped, what moved.
45%
Improvement in sales rep performance
100+
Companies using Trovex to train sales teams
4.8★
G2 rating with 50+ reviews
5x
Increase in practice frequency per rep per week
What Shipped
- ✓ AI Sales Roleplay simulation end-to-end
- ✓ AI powered Live Mentor (in-call coaching)
- ✓ Post-Call Analysis & Coaching Reports
- ✓ Interactive Course Creation Wizard
- ✓ Leaderboards & Team Performance Tracking
Still in Backlog
- ○ Mobile-responsive conversation interface
- ○ AI Post-Sales Roleplays module
- ○ Advanced compliance scoring dashboard
- ○ AI Real Call Scoring integration
“Trovex has truly transformed our sales team's performance. With its AI-based role play simulator, our sales reps have honed their skills, tackled realistic scenarios, and received valuable feedback.”
- 09 · Reflection
What I would do differently.
💡 What Worked
Leading with highly interactive simulation screens in early stakeholder reviews anchored our collective design vision. The “commit first, friction second” onboarding structure yielded massive conversion wins.
⚠️ What I Missed
Underinvested in the empty state experience for users who run through all available free trials. This dead-end screen accounted for 34% of all non-converting visitors.
🚀 What is Next
Designing a highly shareable performance summary “badge card” that empowers free tier users to organically showcase high scores on social platforms.
“The best AI product design problem I have worked on was not the AI — it was the human on the other side of it.”
- Key Takeaways
- Designing for AI means designing for uncertainty — every failure state is a trust-building opportunity.
- Behavior change requires practice, not information. The shift from quiz-based to conversation-based training drove 5× more engagement.
- Building foundations first costs weeks but saves months — the bottom-up architecture enabled consistent scaling across all product surfaces.
- More Case Studies
Explore the full collection of work
Building something ambitious?
Building something ambitious? I'd love to help turn complex ideas into products people genuinely enjoy using.