Your Data Was Never the Problem. Reaching It Was.

Then ask why. Then which salespeople. Then ask which of the leads you wrote off are back on your website this week, looking at a car still sitting on your lot.

That last one matters most. No CRM report can answer it, because your CRM doesn’t know who’s on your website, your website doesn’t know what’s on your lot, and getting an answer takes all three systems matched to the same shopper.

That’s Foureyes Connect MCP. It’s live, and it puts a dealership’s own unified data inside Claude, ChatGPT and other AI assistants, where anyone at the store can work with it.

It builds, too

Last week I asked it for this:

Build me an app that finds inventory on the lot 45 or more days with leads that went cold, meaning no activity in 14 days. For each car, show who asked about it, and recommend other unsold shoppers who submitted a lead on the same make and model, with the reason we picked each one. Let me filter by store.

What came back was a working application with a URL, running on unified dealership data, and somebody on a sales floor could have opened it the next morning and started calling.

It took about as long to build as it took to describe.

Jeff Ramsey, CMO of Ourisman Automotive Group, did his own version of this:

“We were looking at a solution to score our customer engagement, which would have been $3,000 a month per store. Instead, I built it myself in Claude. Our Foureyes-unified data helped me figure out and fine-tune the metrics that actually matter, and most of what we measure now, I never would have thought to track.”

Why this couldn’t happen until now

In 1999 I watched a talk by Douglas Engelbart that I’ve been chewing on ever since, and the part that stuck was his claim that technology moves along two dimensions. Access is whether you can get to the information. Connectivity is whether it can move. One of them is always the thing holding you back.

Put that next to the theory of constraints and you get the idea behind everything we’ve built, which is that the only work that matters is work on the constraint.

In most dealerships the constraint isn’t people and it isn’t tools. It’s access to your own data. Walk into a store running with remarkable efficiency and those people aren’t smarter than you, they’re operating closer to the ceiling of what their access allows.

AI made writing software nearly free, and that part isn’t news anymore. What was missing is that the model had nothing worth connecting to, and a model with nothing to connect to is a yappy intern. It will write you a beautiful application about no data whatsoever.

Which is where most dealership AI runs aground. The demo looks great. Then a dealer points it at their own store and gets an answer that’s confident and wrong.

The model isn’t the problem.

Underneath it sit four systems that disagree with each other, and this is what we’ve measured at dealers connected to Foureyes.

Point anything at that, however smart it is, and you get nonsense delivered with total confidence.

How automotive got here

In the mid-2010s nearly every industry did the same unglamorous thing, moving data out of the systems that created it and into consolidated cloud platforms like Snowflake, BigQuery and Redshift. Retailers, banks, airlines, software companies, all of them. A whole class of company showed up to do the plumbing, pulling data out of point-of-sale systems, ledgers and CRMs, cleaning it and keeping it flowing.

Nobody knew it at the time, but that was the groundwork for AI. When the models arrived, those industries had data ready for them.

In automotive, that work was effectively banished.

In 2017 Authenticom, one of the independent companies pulling data out of dealer systems, and the two largest DMS providers, CDK Global and Reynolds and Reynolds, got tied up in litigation over who could access dealer data and on what terms. The plumbers left. The capital that funds that kind of infrastructure went to industries where building it didn’t come with a lawsuit attached.

So everyone else spent a decade laying pipe and automotive spent it running like a closed economy. You’ve got plenty of software and very little connecting it, your CRM, DMS, website, inventory and phones each hold a piece of the same customer, none of them agree, and every question about your own business turns into a purchase order.

You’ve been renting answers about your own data.

We stayed and built the plumbing. Website behavior, leads, CRM, DMS, inventory, phone calls and consent records, stitched together one customer at a time across roughly 3,000 US dealerships. We match anonymous website sessions to known shoppers, so a return visit from somebody you wrote off shows up under their name, and we link every phone call to the customer record, including the ones nobody logged, which is where that missing 24% of phone leads goes.

And it stays yours. The same data reaches you in our app, in your own Snowflake, through our API, or inside Claude. We never lock it in.

Consent management is the clearest example of what it costs when nothing connects. The average dealership has nine vendors talking to its customers, so when a customer says stop to one of them, the other eight never hear it.

TCPA class-action filings went from 239 in Q1 2024 to 507 in Q1 2025.

Fewer vendors was never going to fix that. Coordination was, so we built consent as its own product, and it now has more than 100 vendor partners integrated.

That’s also why Connect MCP can act on your data and not just answer questions about it. When it builds a list or sends a message, it goes through the same opt-out gate as every other vendor. Underneath it is our public API, the same one vendors build on, so you can ask a question, build an audience from the answer, check it against consent, and deploy the tool you just made, all through one authenticated connection instead of nine.

MCP is a generic term and anyone can spin up a connector in ten minutes, so ask any connector three things. How many of your systems does it reach? Does it know what counts as sold at your store? Can it act without blowing past an opt-out?

What to build first

For as long as dealership software has existed, somebody else decided what it did. They built for a generic store and sold it finished, because there’s no business in something you can only sell once.

So a whole category of problem never got solved. Anything too small to sell 5,000 times. The odd one, the one that’s yours alone, the one that eats four hours of somebody’s Tuesday afternoon, every Tuesday, forever.

That category just opened, because software got nearly free to write and the data a model needs finally exists in a form it can reach.

When dealers get their hands on this, the instinct is to build something impressive, a dashboard with everything on it, or something to show off at a conference. Build the ugly, specific thing that annoys you every Tuesday afternoon instead. The list somebody rebuilds by hand every morning, the report you gave up asking for. Those are the ones nobody was ever coming to build for you.

Connect MCP is live, and dealers on Foureyes Connect are building with it now. See what it does at foureyes.io/mcp, or get a Connect demo.