Fleet Coaching
Voice strategy lives downstream of incidents.
Strength: Dashcam coaching at fleet scaleGap: No telematics-triggered callsSamsara, Motive
AI Voice Agent
A configuration and monitoring interface for voice agents
Product designer owner
(End-to-end)
Geotab, Telus X Bitstrapped, Cox Automotive
Dec 2025 →
Feb 2026
First voice agent in Geotab
Overview
Designing Geotab’s first AI voice agent admin platform
The AI Voice Agent contacts drivers automatically after each job, captures structured responses, and synchronizes the data to a connected CRM. I designed the admin platform that supports this system, where fleet managers configure agents, author scripts, define triggers, and monitor performance.
Reduction in safety events
Saved per incident
Calls processed in early beta
Framing
Context
Voice calls are how operations get caught up
From a single technician completing a job to thousands of drivers across 45k+ fleet customers and 160 countries, follow-up calls are how managers verify, log, and coach.
Problem
Manual follow-up breaks down at scale
Field teams rely on clear follow-up information, but post-job reporting is often delayed inconsistent or incomplete. As operations scale, managers have less visibility into what happened on site and fewer reliable ways to respond after the job is finished.
Form skipped
Manager calls
CRM still empty
Requirements
The constraints that shaped the experience
16
requirements
3
stakeholder sign-offs
Configuration
Trigger rule builder
Script editor
Group assignment
Validation preview
Call experience
AI insights mid-call
3 language support
Driver coaching
Transcript capture
Monitoring
Call log dashboard
Failed call actions
Per-agent metrics
Filterable by driver
Integrations
CRM write-back
Driver record sync
Auth & permissions
Webhook events
Research
Discovery
From validation to production
Trust was the entire product
The admin platform had to make trusted calling feel like a core setup decision.
Driver event
The rules engine was the bridge
The design challenge was making that logic easy to configure.
Two archetypes, one product
The flow had to support both modals without splitting the product.
Competitive Analysis
Solution
Structure
A phone call that replaces a form
Pain points were translated into a flow that gave the team and product a shared scope.
New onboarding + Returning users
Design
Delivery
Tradeoffs
01 Real-time did not mean faster clarity
Purposed
Real-time transcription promised faster awareness, but the latency and incomplete sentences made it difficult to trust.
"Job, completed but there's…"
"Yeah I think so, I need…"
[transcribing...]
I reframed the experience around post-call transcripts and structured summaries, giving users clear answers within seconds of each call.
Job
Flags issues
Engine issue
Actions required
Follow-up
Purposed
Step 1 of 7
Step 2 of 7
Step 3 of 7
Returning admins update individual settings far more often than they complete onboarding setup. I kept the structure and for first-time users balancing onboarding support with faster repeat workflows.
Drivers
Scripts
Ace insights
Purposed
I reduced the scope to a validation preview showing the trigger, script, and sample output delivering the core confidence signal in one sprint.
Agent: Hi John, this is Voice Agent. Let's complete your post-job checklist.
Show Full Transcript
Handoff
What designing for voice taught me
Automation only works when people trust it
Clear previews, validation, and failure states build confidence
Scripts that read well can sound robotic.
Listening at every iteration became part of the design process.
What we left out shaped the product.
Defending a focused scope was as important as adding features.
Scroll to view chat
