72-Hour Lead Rescue
Dig through the dead and dying leads in the CRM and surface the ten worth a call in the next 72 hours — with the outreach to match.
Digital Dealer 2026 · Detroit
Ultimate AI Hackathon
Run of show — Wednesday, September 23, 2026 · 1:00–5:00 PM. Judged on a decision, not a deliverable.
Frame→Draw→Sanitize→Analyze→Prove→Build→Test→Pitch
Detroit, MIWednesday · Sept 231:00–5:00 PM
Three tracks
One table, one problem
Each table draws a sealed track card at 1:15. Nobody in a store gets to pick which fire they walk into.
Fill Tomorrow's Service Drive
Find the ten-to-twelve service customers most likely to book tomorrow and build the message that gets them on the schedule.
Move the Metal
Pick the one vehicle that needs to move and build the campaign that moves it — audience, angle, and assets.
Table structure
Two independent builds, one pitch
AI literacy is mixed by design — every table has confident users and first-timers. The split exists so the first-timers touch a keyboard.
Up to 10 seated
per table
Two half-teams
of 4–5, mixed experience
One track, one packet
one captain
Two laptops
two independent builds
3:40 reconcile
into a single pitch
Halves don't divide the work — both do all of it, independently, and at 3:40 they argue about which build goes in front of the judges.
Run of show
Eight blocks, 12:30 to 5:00
Goals, activities, and the facilitation calls that keep the day on rails.
- Pre-Event12:30–1:00 · 30 min
Room Set
Zero friction at 1:00. Nobody spends build time on logistics.
Tables numbered. Wi-Fi card and rubric one-pager face-up at every seat. Track cards sealed and held at the front; mystery envelopes sealed, organizer-held, one per table. 12:40: captain and TA briefing, away from the room, ten minutes.
Materials: table numbers, rubric one-pagers, sealed track cards with per-track download QR, mystery envelopes, timer display.
- Block 11:00–1:15 · 15 min
Open and Frame
Everyone understands they're judged on a decision, not a deliverable.
Brent opens. Three beats: the dealership cycle, the fifteen-minute dealer principal test, and why there's no deliverables list.
Facilitation
Say out loud that you're not giving them a deliverables list, and why — a GM hands you the problem, not the spec. Answer the inevitable follow-up once, publicly, then point at the rubric.
Hard stop at 1:15. Running long? Cut to beat three — it's the only one that can't be dropped.
- Block 21:15–1:30 · 15 min
Track Draw and Split
Every table holding data, split into halves, inside 15 minutes.
One person per table draws a sealed track card carrying that track's download QR. The captain splits the table: two halves of 4–5, mixed experience, one laptop each. Both halves get the same packet.
Facilitation
Random draw — say the line: nobody in a store gets to pick which fire they walk into. Deal the deck evenly across tracks; don't reshuffle. Say to the room, not just the captains: “You are not dividing the work. You are both doing all of it. Two independent runs at the same problem, and at 3:40 you'll argue about which one to put in front of the judges.”
This is where the format succeeds or fails. Tables will instinctively divide labor. Watch for it and correct it in the first five minutes.
- Block 31:30–1:40 · 10 min
The Data Governance Moment
Teams notice the customer data in their own files before they paste it anywhere.
One instruction — open your files, look at what you're about to hand a public AI tool, ten minutes, go — then silence. Captains and TAs say nothing for five minutes. At minute five, organizers count and call out the result.
Facilitation
Count half-teams, not tables: “eleven of twenty groups stripped names and contact info before uploading. Nine of you just handed a public tool a full customer list.” Nobody in this room stopped you, and nobody at your store will either. The correction: strip names, emails, phone numbers; keep lead IDs, customer IDs, RO numbers — anything that joins files. Rename the clean file so it's obviously the safe one — the rename is worth points.
Scoring: 5 of the 10 Accuracy & Guardrails points. Say so out loud.
- Block 41:40–2:25 · 45 min
Analyze and Decide
A defensible decision, evidenced from the data, before any customer-facing asset exists.
Both halves work independently to a ranked or selected answer: the ten leads, the ten-to-twelve service customers, the one vehicle. No creative output yet.
Facilitation
Brent and TAs float with one question: what in the data tells you that? Any group that opens a design tool gets redirected — 35 rubric points (problem framing plus AI workflow) are won or lost in this block. No comparing notes across the two halves; the swap depends on two genuinely independent attempts.
Materials: visible timer. 15-minute and 5-minute calls.
- Block 52:25–2:40 · 15 min
The Swap Test
Each half finds out whether it built a workflow or got lucky once.
Halves trade prompt sets — the instructions, not the outputs. Five minutes to run the other half's prompts cold on the same data, no questions asked. Five to write down every place it broke. Five back at their own build to fix what was found.
Facilitation
Frame it as rehearsal, not peer review: in an hour I hand you a file you've never seen — let's find out cheap whether that's going to work. Most of you are about to discover you don't have a workflow, you have nine chat messages and context in your head that never made it into writing. Float question: is that gap in their prompt, or in what they assumed you already knew?
Why it works: the swap is internal to the table. No pairing logistics, no odd-number problem, no cross-track mismatch. Scales from 2 tables to 12 unchanged.
- Block 62:40–3:20 · 40 min
Create and Operationalize
The decision becomes usable work product plus a shareable resource guide.
Both halves still building independently — customer-facing outputs, manager-facing outputs, and the resource guide: the one-pager a staff member could pick up and run without anyone standing over them.
Facilitation
Brent's float question: give me three other ways someone in this store could use what you just built. Bonus, not requirement — a team producing three extra use cases in a three-hour event has almost certainly shortchanged the first one. Halves still independent; the reconcile happens at 3:40, not here.
Materials: resource guide prompt or one-page template — no design requirement.
- Block 73:20–3:40 · 20 min
Mystery Case
Prove the process generalizes to data nobody has seen.
Sealed envelope per table — a full CSV matching the original schema, not a single record. Both halves run it through their own process as built. No rebuilding, no starting a fresh chat. Judges observe live.
Facilitation
Judges watch how the group reacts more than what comes out. “We drop this into step two and rerun” wins the section regardless of answer; opening a new chat and re-prompting from scratch fails it regardless of answer. If one half holds and one collapses, that table has the best pitch material in the room — tell judges to reward a group that can explain the difference.
Twenty minutes is enough if the workflow is real and not enough if it isn't. Don't extend it for struggling tables. That's the measurement.
- Block 83:40–4:00 · 20 min
Reconcile and Submit
Two builds become one pitch, submitted before the first team stands.
Three moves, timed. 3:40–3:48: halves present to each other — three minutes each, decision and mystery result; the captain makes the call on which build leads. 3:48–3:56: assemble the pitch. 3:56–4:00: submit — prompts and workflow, outputs, resource guide, mystery result.
Facilitation
The captain's call is a call, not a vote: “You have eight minutes to show each other what you built, then your captain picks. You are not merging two builds into a mush — pick the stronger one and borrow the best piece of the other.” On-screen the entire block: Problem (1) → Data and AI approach (1) → Demo (3) → Business value and guardrails (1) → Mystery result (1).
The tightest block in the day. If anything slips earlier, steal from Block 6, not from here.
4:00–5:00 · Judging
Pitches and scoring
Every table presents, judges score the full 100 independently, totals are averaged, and the winner is announced in the room.
Live scoring app
Judges enter scores here. Totals are averaged and the winner is announced in the room.
Tie-break: Mystery Test → Accuracy & Guardrails → AI Leverage
Format by table count — decided at 3:45, not before
10-min pitch + 2-min judge Q&A
36 min pitching, 24 to deliberate
Spend the room you have on judge Q&A — it's the best content of the day and doubles as the proof check: show me the prompt that produced that.
7-min pitch, 1-min transitions, no Q&A
48 min at six, 12 to tabulate
Track semifinals inside Block 8: each judge takes one track, four-minute table-side pitches, advances one. Three finalists on stage at 4:00, 8 min each plus 2 of Q&A.
30 min pitching, 30 to deliberate
Each judge names their track's standout when introducing their finalist — thirty seconds of recognition per track so the tables that didn't advance still get heard.
Posted in the room
Standing rules
- 1
Public tools only. No dealer-only platforms, no paid automotive products. Tool count earns nothing.
- 2
No private dealership data. introduced into any track.
- 3
Show the workflow. Not just the finished deck. Prompts and inputs are part of the submission.
- 4
Vendors support. Vendors do not pitch — not at the table, not in the hallway, not during judging.
- 5
Rubric is on the table. It tells you what's measured. It does not tell you what to make.