The DMS is the database nobody audits

The offer your CRM wants to send to a lease-ending customer does not come from the CRM. It comes from a join: VIN, deal date, odometer at last visit, payoff, repair order history, contact preference. On a CDK store, most of that originates in the DMS, then gets copied, transformed and occasionally mangled on the way to everything else.

That is the whole argument in one sentence. Offers are not a creativity problem. They are a records problem, and the records live somewhere your marketing department has read-only access to.

Where the equity actually sits

Equity mining is the practice of finding a customer who is ahead on their vehicle and offering them a structured exit before the market does it for them. It sounds like a sales tactic. Mechanically it is a data product:

• Deal date and original structure from the DMS sale record.
• Current odometer from the most recent repair order, not from a guess.
• Payoff from the lender or the deal's finance record.
• Valuation from a guide or auction source.
• Contactability and consent from the CRM and whatever preference flags exist.

Four or five systems have to agree on one vehicle and one person. When they disagree, the offer either goes to nobody or goes to the wrong nobody. Both are expensive, and only one of them is visible in a report.

What dirty looks like on a CDK store

The failure modes are consistent across rooftops:

One household, four records. A husband and wife with shared address and phone are four customers, two with the original VIN, two with a typo. Mileage is split between them, so neither looks like it has equity. The offer that should have gone out in March never fires.

VIN as free text. A character transposed at delivery means service history attaches to a vehicle that does not exist. Retention reporting says the customer defected. The customer is in your drive every six months.

Odometer gaps. If the last recorded mileage is eleven months old, every equity calculation downstream is a projection wearing a fact's clothing.

Consent that was never captured, or never updated. Selling a vehicle does not inherit permission to keep texting the previous owner. Offers to the wrong person are not neutral; they cost trust and, in some states, money.

Source fields that say nothing. If opportunities read website, walk-in and unknown, no campaign can be judged and every next budget is an argument.

These are the shapes these records break in, not a dataset pulled from one store. Treat them as the pattern to test against your own file.

None of these are exotic. Every one of them is fixed inside the system of record, which is precisely why marketing cannot fix them alone.

The contract is a data decision

Worth understanding, because it shapes what you can build. CDK's DMS was not always closed: the company historically ran an open architecture that allowed third-party integrators to pull dealer-originated data, then moved to a closed model in 2015. That shift, and the per-dealer data access fees that came with it, ended up in federal court. Testimony in the litigation put integration fees for one app vendor at roughly $160 per dealer per month before the change and roughly $735 after; dealers were told they could not share log-in credentials to get around it.

The aftermath is public record: CDK agreed to pay $100 million to a dealer class, part of a $129.5 million combined settlement with Reynolds that received final approval in February 2025, and separately a $630 million settlement with a vendor class in January 2025. CDK denied wrongdoing in both.

The practical lesson is not who was right. It is that the terms governing your own data are in a licensing agreement, not in your ambitions. Read the data clause, the export rights, the integration fee schedule and the exit provisions before you promise a marketing team nightly access to anything.

Continuity is part of cleanliness

In June 2024 a ransomware attack took CDK's systems offline for roughly two weeks across an estimated 15,000 North American dealerships. Deals were written on paper. Service was dispatched without parts inventory visibility. One cybersecurity consultancy's estimate put losses in the first two weeks near $605 million, which is a consultancy estimate rather than an audited figure.

Most coverage focused on the outage. The durable damage is the reconciliation debt that followed. Repair orders keyed late, deals funded with incomplete jackets, deposits tracked in a notebook, appointments never logged. Every record written on paper and re-entered in a hurry is a candidate for the errors above, and the equity and retention campaigns that ran on that quarter were quietly wrong.

If you take one thing from the outage: your database needs a practice for reconstruction, not just a backup. Written-down records have to be re-keyed with a second pair of eyes, and the backlog has to be reconciled against lender and OEM records before any offer run.

A triage order that works

1. Deduplicate by household and by VIN, not by phone number. Merge to one surviving record with a documented rule, and log what was merged.
2. Reconcile identifiers. One customer id, one VIN per vehicle, one odometer per visit date. Where two records disagree, the most recent service event wins over the deal, and a human confirms it.
3. Repair contact and consent. Fresh preference flags per channel. Remove sold-vehicle owners from service offers.
4. Map source codes to a closed list. Ten real values beat forty free-text ones.
5. Take an owned copy. Export a nightly extract to your own warehouse so campaign logic does not depend on a vendor's API window or your willingness to pay a fee.
6. Re-key the paper backlog with verification against lender and OEM data.

The short version

The offer engine is downstream of the records, always. Clean identifiers, current mileage, honest consent and an owned copy of your own data are what let an equity list become an appointment instead of a spreadsheet in a shared drive. You cannot mine equity you cannot measure, and you cannot measure a customer who exists four times.

The report page ranks the in-house campaigns that turn those records into work, by appointments per 1,000 delivered, which is the only line on it tied to revenue.