Where Does Dealership Lead Data Go Inside AI Tools?

CDK Global’s 2024 breach disrupted roughly 15,000 dealerships across North America — and it dominated every twenty-group conversation for months. But that incident, as damaging as it was, involved an external attacker. The quieter risk sitting inside most AI vendor contracts right now is fully legal, leaves no fingerprints, and nobody is talking about it.

When your BDC connects VinSolutions, eLead, or any CRM to an AI lead-response tool, customer data — names, contact details, purchase intent signals, trade-in information — moves off your servers immediately. What happens next depends entirely on what your vendor contract says. Most dealers have never read it.

The Short Answer: Your Data Leaves Your Building the Moment You Hit Send

The instant a lead from Cars.com, CarGurus, or AutoTrader hits your AI tool’s inbox, that record is transmitted to and processed on the vendor’s infrastructure — not yours. “Processed” is the operative word. It means the vendor’s servers read, parse, and act on the data. Whether they also store it, share it, or learn from it is a separate question answered only by your contract.

Most standard SaaS agreements grant the vendor at minimum a license to process data to deliver the service. Beyond that, the terms vary wildly. Some vendors retain data for 90 days after contract termination. Others retain it indefinitely unless you submit a deletion request. A handful say nothing at all — which is not reassuring.

The Three Places Lead Data Actually Goes (Most Dealers Don’t Know #2)

Your lead data typically travels to three distinct locations once it enters an AI vendor’s platform:

• The vendor’s application servers — where the AI reads the lead and generates a response.
• The vendor’s model training pipeline — where anonymized or aggregated versions of your data may be used to improve the AI’s performance across all customers, including your competitors.
• Subprocessors — third-party infrastructure companies the vendor relies on to deliver the service.

Most dealers are aware of the first. Almost none ask about the second or third.

What “Processed” Really Means — and Why It’s Not the Same as “Stored”

Processing means the data was touched — read, parsed, acted on. Storage means the data was written to disk and retained. A vendor can process your lead data in real time without storing it at all, which is the cleanest arrangement. But most vendors do both, because stored data improves product performance over time.

Ask specifically: What is the retention window? Is PII retained, or only anonymized records? What triggers deletion — contract end, your request, or nothing?

Model Training: Is Your Customer Data Teaching a Competitor’s AI?

This is the contrarian paragraph most AI vendors don’t want published: the industry markets “data security” as a compliance feature — SOC 2, GDPR alignment, encryption at rest — but none of those certifications prevent a vendor from legally using your customer data to train their models. A SOC 2-certified vendor can still feed your Vanguard Auto Group leads into a training pipeline that ultimately improves the AI responses sent to shoppers at a competing store across town.

The mechanism is typically anonymization or aggregation, which makes it legal under most current frameworks. The result is that your customer behavior data — what questions they ask, what objections they raise, what offers convert — quietly subsidizes a model that serves your entire competitive market. That is not a breach. It is a business model. And it is standard unless your contract says otherwise.

For a broader look at how AI is reshaping BDC operations this year, see What the 2026 NADA Midyear Data Says About AI in the BDC.

Subprocessors: The Hidden Third Parties in Every AI Vendor Stack

No AI vendor runs entirely on its own infrastructure. The typical stack includes a cloud host (AWS, Google Cloud, or Azure), an SMS or email delivery provider, possibly an enrichment or identity-resolution service, and one or more analytics platforms. Every one of these is a subprocessor with access to your customer records, operating under the vendor’s privacy terms — not yours.

A legitimate vendor will provide a complete, current subprocessor list on request. If they hesitate, that tells you something.

What a Data Processing Agreement (DPA) Must Say to Actually Protect You

A DPA is not a privacy policy. A privacy policy describes how a vendor handles data broadly. A DPA is a bilateral contract that governs your specific relationship. If a vendor cannot or will not sign a DPA, walk away.

A compliant DPA must include:

Clause	What It Must Say
Purpose limitation	Data used only to deliver contracted services
Model training	Explicit opt-out or prohibition
Subprocessors	Named list with change notification requirement
Retention	Specific window; deletion confirmed in writing
Breach notification	Under 72 hours
Jurisdiction	US-based processing only (for most dealers)

Red Flags in AI Vendor Contracts Dealers Should Reject on Sight

• “We may use aggregated or de-identified data to improve our services” — this permits model training on your leads.
• No subprocessor list available, or “available upon request” with no SLA on the response.
• Data retention clause tied to “legitimate business purposes” with no defined window.
• Breach notification language that says “prompt” rather than a specific hour threshold.
• No DPA offered at all — just a standard clickthrough terms of service.

Objection Handled: “We’re a Small Group — Why Would Anyone Target Our Data?”

This is the most common reason dealers skip the DPA conversation. The answer is that your data isn’t being targeted; it’s being aggregated. An AI vendor with 400 dealerships on its platform doesn’t need to single out your store. Your leads — your customers’ purchase intent, trade-in habits, objection patterns — become one node in a dataset that trains a model deployed across every one of their accounts. The risk isn’t targeting. It’s contribution.

Dealerships running Synthevo today, including early customers managing high-volume inbound like Vanguard Auto Group in Sterling, VA, have a signed DPA as a standard part of onboarding — not an add-on negotiated later.

How Synthevo Handles Lead Data (and What We’ll Put in Writing)

Synthevo processes lead data solely to generate and send the response. We do not use customer records to train models across accounts. We maintain a current subprocessor list, and we sign a DPA before any CRM connection is established — whether that CRM is VinSolutions, Reynolds, CDK, or DealerCenter.

Understanding how fast a dealership should respond to a Cars.com lead matters less if the tool you’re using to respond is quietly enriching a competitor’s model with your buyer data.

Questions to Ask Every AI Vendor Before You Connect Your CRM

Before you authorize any CRM or DMS integration, put these questions to your vendor in writing and request written responses:

1. Do you use our dealership’s lead data to train or improve your AI models? If so, can we opt out?
2. Provide a complete list of all current subprocessors who will access our data.
3. What is the data retention window after contract termination, and how do we request deletion?
4. Will you sign our Data Processing Agreement prior to integration?
5. In what jurisdiction is our data stored and processed?

For context on response mechanics that sit downstream of this data question, the comparison of Email vs. SMS for Dealership Leads: Which Gets More Replies? is worth reading alongside your vendor evaluation.

If you want to see exactly how Synthevo handles your lead data — including the DPA we sign before day one — request access to our live demo and we’ll walk through the contract terms alongside the product.