The Hidden Infrastructure Problem Behind Automotive AI

The debate is focused on the wrong layer

Every week there is a new AI chatbot, a new AI BDC tool, a new AI assistant, a new AI follow-up system marketed to dealerships. The industry conversation has centered almost entirely on which models vendors use, whether they built their own, and whether they rely on OpenAI, Google, or Anthropic.

Those are interesting conversations. But they are focused on the wrong layer of the problem.

The real limitation in automotive AI right now is not the model. It is the infrastructure that moves data between systems after the AI has done its work. And until that infrastructure changes, better models will only produce better outputs that get thrown away faster.

ADF solved a real problem for its era

More than twenty years ago, the automotive industry solved an important interoperability challenge with ADF-XML, the Auto-Lead Data Format. ADF created a standard way for dealership websites, lead providers, and CRMs to exchange lead data. Before that, every vendor had their own format and integrations were messy and expensive.

ADF gave the industry a common language. And it worked well for the world it was designed for. At the time, a "lead" meant a form submission. Name, email, vehicle of interest. That was enough to get the sales process started.

The problem is that the nature of the interaction has fundamentally changed, and the infrastructure has not kept pace.

Interactions are no longer just forms

A customer today might have a full 15 to 20 message conversation with a dealership's AI assistant before they ever submit a form. In that conversation they share specifics: the exact vehicle they want, a trade-in with a payoff remaining, a target monthly payment, trim comparisons, financing questions, appointment preferences.

By the end of that interaction, there is a detailed picture of who the buyer is, what they want, what they can afford, and when they are ready to act. The AI captured everything.

But when that interaction moves into the dealership technology stack, something happens that should concern everyone building in this space. Most of that context disappears. A rich conversation becomes a basic lead record: name, phone number, vehicle interest. Everything else gets lost in the handoff between systems.

This is what I call context collapse. And it is not anyone's fault. ADF was built for form submissions, not conversations. But the result is that AI systems downstream are operating with only a fraction of the information that was originally captured. Surface-level data produces surface-level intelligence.

The infrastructure gap

What is becoming clear is that automotive retail is shifting from form-based interactions to conversation-based interactions. And that shift exposes a gap in the infrastructure that no amount of model improvement can fix.

The industry has good standards for passing lead data between systems. But there is no standard way to pass rich conversational interaction data across those same systems. Intent signals, conversation context, deal stage, trade-in details, payment targets, behavioral timelines -- none of this has a defined format for moving between vendors.

This is the layer that is missing. Not a better AI model. A better way to move what the AI produces through the systems dealers rely on every day.

What led to the Automotive Commerce Protocol

This is the problem that led me to start building ACP, the Automotive Commerce Protocol. It is an open, JSON-based specification designed to carry richer interaction data between automotive systems while remaining backward compatible with existing ADF pipelines.

The idea is not to replace ADF. ADF solved an important problem. But conversations have outgrown the format that was designed for form submissions. ACP adds the layer that is currently missing: a structured way for systems to exchange intent signals, deal context, trade-in information, appointment data, and full conversation transcripts without breaking existing workflows.

This is one of the core principles behind the AI Operator Framework. Before you deploy AI tools, you have to design the system that carries the data. Structure before tools. If the infrastructure stays shallow, the intelligence built on top of it stays shallow too.

The question worth asking

AI is going to become a permanent part of automotive retail. The models will keep improving. But the biggest challenge may not be the intelligence itself. It may be the systems that move information between the tools dealers rely on every day.

The real question is not "how smart is the AI?" It is: how much of the conversation actually reaches it?

That is the infrastructure problem hiding behind every automotive AI deployment. And it is the problem I am building to solve.