A clickable CRM, rebuilt every round
An automotive retail software company in Australia, with AI already built into its dealer CRM and a walkthrough needed to prove it.
/work/es-command-centreCame in new build
- System
- Dealer CRM command centre, working prototype
- Ask
- A walkthrough to present, with the AI at its best
- Region
- Australia
The piece Three square frames for three days of builds, rising to one block with four lit faces, one for each role.
- 3days26 to 28 August 2026
- 4rolesone decision each
- 3levelsof AI autonomy, on every action
- 1recordbehind sales and aftersales
The ask
A walkthrough the client could present, with its AI mapped against a competitor's and a reason to choose it over the big CRMs.
The real problem
Nobody can check a claim about AI made in words, so the walkthrough had to run and show what the AI did.
The first version put everything on screen and in review felt overwhelming.
From a full screen to one decision
- Idle
- Live
- Strain
- Fail
First versionFour roles facing screens of widgets and an AI that was only described. Each had to work out where to start.
What we builtOne customer record behind four role views, each opening on one decision. Every AI action carries its autonomy tag.
What we built
A working Command Centre for the four people who run a dealership: service advisor, aftersales manager, sales manager and dealer principal. Each view opens on one decision and Why this opens it to the evidence.
Every AI action carries one of three tags: human-led, AI-suggested or AI-operated. An overnight ledger lists what the AI did on its own. Easy mode cuts any role to one card at a time.
How we made it hold
Every round ended in a working build the client could click through, even the round that restructured all four roles around one decision.
Backstage The decisions behind the build, and the hardest problem
Decisions
A working build at the end of every round.
The client reviewed something it could click and the next round started from what it found.
One customer record behind every view.
Sales and aftersales show the same person, which is the story the presentation tells.
A tag on every AI action and an overnight ledger.
Each action says how much the AI may do alone, so a dealer can check the claim.
Ruled out
A walkthrough that describes the AI.
That is what every competitor does and nobody can check a description.
The hardest problem
Changing structure inside three days. The first version showed everything. The next round had to open each of four roles on one decision, with everything else as evidence, and still end in a working build.
The difference
- A Command Centre the client could click through for four roles within three days.
- Every AI action on screen shows its evidence and how much it was allowed to do.
- The client signed off the presentation story on the last day: one customer through sales and aftersales.
