Why we built Rooftop (and what every dealer-tech vendor got wrong)

Twenty-five years across ten dealerships, twenty-five vendor logins

I've run F&I desks across Connecticut for 25 years — ten dealerships, six dealer groups, five OEM brands. Today I run F&I at Harte INFINITI in Hartford. The day I sat down and counted the dealer-tech vendors my stores were paying every month, I got to twenty-five before I started skipping ones I wasn't sure about.

Twenty-five logins. Twenty-five renewal cycles. Twenty-five separate "AI features" that 2024 added to slide decks. About $25,000 a month, give or take, in software that didn't talk to itself and didn't talk to our DMS without paid integrations on both ends.

This post is about why none of it was good enough, and why we ended up building our own operating system instead of buying a twenty-sixth thing.

What every dealer-tech vendor got wrong

There are four mistakes I see consistently across the vendors I've used, demoed, and walked out of QSP rooms with. They're not character flaws — most of these companies are run by smart people doing their best inside a structure that pushed them in a particular direction. But they're real, and they compound.

Mistake 1 — Vendor sprawl as a feature, not a bug

Almost every dealer-tech company is built to be a point solution. A widget for trade-in. A widget for window stickers. A widget for service-drive video. A widget for digital retailing.

Every widget needs its own login, its own contract, its own training, its own integration with the DMS. Every widget owns a slice of the customer record and gets defensive about sharing it. The "integration" between two widgets is usually a CSV export and a manual upload, or a paid Zapier-equivalent middleware.

This isn't a bug from the vendor's perspective. The narrower the surface area, the easier it is to charge separately. The more separate things you charge for, the bigger your TAM looks on the next pitch deck. Vendor sprawl is a feature for them. It's a tax on us.

Mistake 2 — Compliance as paperwork

The FTC CARS Rule landed. Most state attorneys general are getting more aggressive about addendum disclosures and doc-fee transparency. Federal CFPB and FTC enforcement has gotten teeth.

What did our existing vendors do? Sent us a one-page PDF with suggested disclosure language. Maybe a checkbox somewhere in their admin panel.

That's not compliance. Compliance is a per-VIN audit trail that ties the version of the disclosure the customer saw to the timestamp they signed at to the IP address they signed from. Compliance is being able to answer the question "what exactly did this customer agree to on this exact date" three years later, without rooting through email screenshots.

We needed software where compliance was enforced at the database level, not bolted on with a checkbox.

Mistake 3 — AI bolted on as theater

In 2024, every vendor I worked with shipped an "AI feature." Nine times out of ten, it was a pattern matcher running on top of an existing rule engine — the same business logic they had in 2019, with a chat box on top, marketed as "agentic."

Real AI in a dealership looks like this: photos drive condition scoring that meaningfully changes the appraisal. Voice agents handle inbound and outbound calls and write back to the same pipeline your team is working. The AI doesn't replace your appraiser, your BDC, or your GM — it gives them better defaults so they spend their time on the deals where human judgment actually matters.

That requires building AI into the spine of the product. Not the surface.

Mistake 4 — The per-VIN tax

Almost every successful month makes you poorer with most dealer-tech vendors. You buy more cars, you trigger more "events," you pay more in per-VIN or per-message fees.

This is structurally insane. The pricing should be flat-per-rooftop. We pay for the capacity, not for the throughput, because the capacity is what costs the vendor money to deliver. If you're growing, you're already paying more in cost-of-goods. The tooling shouldn't tax growth.

What an operator-built OS looks like

We sat down — me and a small product team — and wrote out the principles we wanted for whatever we built next. They came out short:

• One platform, one login, one customer record. Modules ship inside the same product, sharing the same database, the same audit trail, the same customer profile. No CSV exports between modules.
• Per-rooftop pricing. Flat fee per store. The tenth car in a month costs the same as the hundredth.
• Compliance enforced at the database, not at the form. Versioned consent, immutable audit trail, exportable per-VIN packet for any disclosure or signature collected.
• Dealer-owned data. Every record is exportable. The dealer's customer list is the dealer's. We don't resell it, we don't enrich it for someone else's marketing, we don't lock you in by hoarding the data.
• AI in the spine, not the surface. Re-appraisal, voice, lead routing, pricing decisioning — built in, not chat-boxed on.
• Built by operators. Every feature has to survive a desk-up review with a used-car director who has been in the business for 25 years.

What we're shipping first

The first module is AutoCurb — direct-from-consumer acquisition. Three channels (off-street, service drive, in-store trade), one pipeline, full audit trail, ~$2,000 saved per direct unit vs. the auction. It's live at Harte today and rolling to pilot dealers as we build out the bench.

After AutoCurb come three more:

• AutoLabels — FTC CARS-compliant window stickers and addenda, with versioned disclosures pinned to the deal.
• AutoFilm — two-way walk-around video for the service drive and online retailing, with built-in compliance disclosures.
• AutoFrame — studio-grade VDP photos that turn a phone-shot rectangle of asphalt into something a buyer actually wants to click on.

All four ride the same pipeline. All four hit the same audit trail. The dealer logs in once.

Where you can pilot

If you run a one-to-ten-rooftop group, you're who we built this for. The pilot is 60 days, AutoCurb only, no contract, your domain, your branding. We come in with a runbook, you put it in front of your customers, and we live in the dashboard with you while you run it.

If that sounds like the right kind of headache to take on for thirty days, start a pilot. If you want a 20-minute version of this post on a Zoom call, book a demo.

We're not trying to sell you a twenty-sixth thing. We're trying to replace the other twenty-five.