Web app · AI sales coaching · B2B SaaS
Back to work
Trust under pressure
Making AI coaching trustworthy before a live sales call
An AI sales coaching platform for sales teams that needed to capture winning strategies and deliver them at the right moment.
+34%conversion-rate lift for pilot teams85%weekly active usage in the pilot6moderated rep tests shaped trust changes
Case snapshot
What I owned, shaped, and proved.
Product Designer owning research synthesis, flows, prototypes, testing, and handoff.
Pre-Call Intelligence, Practice Partner, and Post-Call Analysis.
15 stakeholder interviews, 6 sales-rep tests, and a 3-month pilot.
Product, ML engineering, sales enablement, frontend, and enterprise sales leaders.
Scannable prep, progressive disclosure, calendar-first access, and transparent AI recommendations.
Design judgment
The product worked when it stopped acting like a dashboard and started acting like a coaching loop: prep before the call, practice the hard part, and learn from what happened after.
95% of coaching moments disappeared after the call, and managers could only review a small slice of work.
Generic AI advice felt correct but not useful enough to trust.
Source context, sample size, and top-performer evidence had to be visible before recommendations felt real.
15 stakeholder interviews, 6 rep tests, and a 3-month pilot shaped the final workflow.
Process evidence
The thinking I had to make visible.
Reconstructed artifacts that show what I was trying to understand, where trust felt fragile, and how the direction became safer to build.workflow map
Coaching workflow map
I used this map to understand where a rep was most alone: before the call, inside rehearsal, after the call, and in the manager follow-up.
Before the call
User signalReps had 8 minutes to prepare and no reliable way to borrow top-performer judgment.
Design responseSurface company context, prospect signals, and team-specific talking points in one fast prep view.
Practice
User signalGeneric roleplay felt like training homework, not rehearsal for a live deal.
Design responseUse real objections from lost deals so practice feels tied to the next risky conversation.
After the call
User signalUsers did not trust AI feedback when it sounded confident but contextless.
Design responsePair coaching notes with transcript evidence, source calls, and top-performer comparison.
Manager loop
User signalManagers could review only 5% of calls, so coaching coverage stayed uneven.
Design responseTurn the AI summary into a manager coaching queue instead of another analytics dashboard.
Why it matteredThis structure let coaching appear inside the rhythm of selling instead of asking reps to leave that rhythm.
trust anatomy
Recommendation anatomy
This artifact made the trust problem visible: a recommendation needed a spine, not just a confident sentence.
Source
Recent client calls and team-specific sales moments.
Sample size
Enough context to know whether the pattern was reliable.
Top-performer evidence
What strong reps did differently in similar deals.
Suggested action
A concrete next move the rep could use in the call.
Why it matteredRound 2 testing showed reps trusted recommendations more when source calls, sample size, and top-performer context were visible.
testing synthesis
Round 1 trust breaks to Round 2 changes
This synthesis board turned vague feedback into the exact design changes that made AI feel more earned.
Advice quality
BeforeAsk more discovery questions.
AfterAsk about budget timing and approval process because top reps did this in similar deals.
Practice Partner
BeforeGeneric training prompts felt like an LMS exercise.
AfterReal objections from lost deals made practice feel useful before live calls.
Why it matteredReps described the second version as more useful because practice was tied to real lost-deal objections instead of generic prompts.
decision matrix
Dashboard vs workflow-first coaching
This matrix protected the product from becoming impressive but unused. Reps needed help inside the selling rhythm.
Why it matteredThe pilot reached 85% weekly active usage because the product showed up at moments users already cared about.
Final design
Final Product Screens
Project artifact supplied by Savita for this case-study narrative.
Impact and results
The pilot showed where AI coaching became useful.
Conversion rate increased by 34% across pilot teamsWeekly active usage reached 85% across pilot teams6 moderated sales-rep tests shaped the trust layerAI recommendations became clearer through source context and top-performer evidence
Source: 15 stakeholder interviews, 6 moderated sales-rep tests, and a 3-month pilot with 3 design-partner clients. Conversion lift and weekly usage are pilot signals; ramp and revenue claims are intentionally not used here.