Case study 02 · CDK Global · Voice AI
AIVA: the voice salescompanion
Specified and prototyped a voice assistant for dealership sales reps, from hardware scan to demo-ready build, in eight weeks.
Role
UX and product design lead
Timeline
Apr 14 – Jun 10, 2026
Team
Product, engineering and dealership SMEs
The short version
The problem
Sales reps work on their feet, next to a customer. The question was whether voice belongs in their pocket or only in a demo.
What I owned
Specification and prototype: the AIVA PRD to v0.4, the baseline PRD to v0.2, and a clickable end-to-end flow.
The approach
Four sequenced stages, each producing something the next could be checked against. Mapped to existing platform endpoints instead of inventing new ones.
The result
Demo-ready, with a scope decision that tells engineering what not to build. The intelligence layer and consent strategy are next, which is where the hard part lives.
The long version · for the detail-inclined
02 / 11 · By the numbers
T H E G O A L Prove a voice assistant belongs in the rep's pocket Success meant reaching a stakeholder greenlight with three things in hand: a specification precise enough to build against, a prototype real enough to demo, and a scope decision that told engineering what not to build. 6 dealership roles specified as users 15 feature areas defined in the AIVA PRD 10 Roadster MCP endpoint groups mapped 5 delivery phases scoped, proof of concept to HUD Success was a decision, not a document — the point of every artefact was to make the scope question answerable.
03 / 11
Four stages, run in sequence
04 / 11
What was gathered before any design started
05 / 11
What the reviews actually changed
06 / 11
Four decisions that came out of review
07 / 11
The rep's working surface
08 / 11
The assistant and the manager loop
09 / 11
Beyond the demo path
10 / 11
Fourteen April to ten June
11 / 11