Back in June I posted "30 Days with Jordan: The Receipts" — six car deals, roughly $19,200 in attributed revenue, one AI BDC at the independent store where I run sales and F&I. It was honest the day it went up. And it started aging the minute I hit publish.
That's the flaw in every case study you've ever been pitched: it's a snapshot, chosen by the vendor, frozen at their best angle. Nobody publishes the slow month. So last week we did something I haven't seen another AI company do.
The numbers on our homepage are now live production data. Same source as the dashboard House of Carz opens every morning — not marketing copy that mentions numbers, but the actual totals, pulled from the actual system, refreshed every week. As of this week's refresh, day 78 in production:
(inbound + outbound)
sent by Jordan
for the sales floor
one got away from us
Median response to a new lead: 2.4 minutes, with 86% answered inside the five-minute window where deals are actually won. And here's the part that matters: the refresh isn't curated. If Jordan has a slow week, those totals crawl — in public. There's no version of this where I hide a bad stretch behind a "contact sales" wall. That's the point.
The screenshots were pretty. They were also fake.
Confession: until last week, the "dashboard" on our homepage was a mockup I had built in code. Plausible numbers, invented customers, a tidy little fake waveform. Every software vendor does this — it's so normal that nobody thinks to call it what it is. It never sat right with me.
So we replaced every frame with captures of the real thing. What cycles on the homepage now is the actual morning workplan Jordan builds, a real booking conversation — the text thread and the call transcript — the receipts tab, and the actual House of Carz lot. The trucks in that inventory screenshot are for sale in Rochester right now.
One honest footnote: the customers you see in those frames are fictional stand-ins. We seeded the demo data ourselves, because our customers' privacy isn't marketing material. The interface, the numbers, the inventory — all real.
An AI that knows what it doesn't know
Something else shipped this week, and it started with a real conversation. A customer asked Jordan whether she was pre-approved on her financing application. That is not an AI's question to answer — it's a person's. Jordan already knew to step back and hand the thread to a human. What was missing was the tap on the shoulder: the handoff happened quietly, and a customer who deserved a fast answer from a person got silence instead.
We caught it in a live thread that night and closed the gap before the next morning — if you read "The Night Jordan Went Quiet," you know that's how it works around here. Now, the moment Jordan decides a conversation needs a human — a financing status, a manager question, anything above his pay grade — the assigned rep's phone buzzes with the thread and exactly what the customer needs. He doesn't guess. He doesn't improvise an answer to a question that was never his.
I'd rather tell you that story than pretend the gap never existed, because a vendor who says their AI has no gaps is describing an AI they haven't watched closely enough.
Three questions to ask any AI vendor
- Can I see live numbers? Not a case study from four months ago — numbers that update, from the system, where a bad week would show.
- Are the screenshots real? Ask them directly. If the product shots are mockups, ask why.
- What happens when the AI is out of its depth? The honest answer involves a human getting notified. The dangerous answer is "that doesn't really happen."
Our answers are sitting on the homepage, refreshed weekly. If those numbers ever stop updating, that tells you something too.