The Industry Has Normalized Partial Fixes and Called It Data Quality

“Data quality” has become one of those automotive phrases that can mean almost anything. And it often does.

Append some new phone numbers and emails? Data quality. Run a dedupe against the CRM every few months? Data quality. Improve match rates for Meta and Google? Also data quality.

None of those things are bad. Some are extremely valuable. Digital audience activation is where a lot of the money gets made and where results are easiest to see.

But somewhere along the way, the industry, and yes, some vendors, started treating partial fixes as if they solved the larger data-quality problem.

They don’t.

A partial fix can work remarkably well for a while. Think of topping off a tire with a slow leak. The pressure comes back. The car drives fine. But the leak remains, and it will not improve over time.

That is where the danger starts.

The audience matches. Campaign performance improves. More customers become reachable. The dashboard turns green.

So we assume the underlying data must be good.

A 2025 eLEND Solutions survey found that 56% of U.S. dealers experienced regular data gaps or discrepancies across core dealership systems more than a quarter of the time. Almost 70% said their integrations could be improved.

That is the second-order effect we don’t talk about enough.

The first-order effect is positive: better contact data, cleaner audiences, better results.

The second-order effect is false confidence. And false confidence reduces scrutiny.

Meanwhile, one customer exists three different ways. Two people get merged into one. A business becomes a consumer. An ex-spouse becomes the primary customer. An old email outranks a current one.

For a digital audience, some of that mess may be survivable.

For what comes next, it isn’t.

AI changes the stakes.

McKinsey found that even its generative-AI “high performers,” organizations already attributing more than 10% of EBIT to gen AI, still struggle with data integration and governance. 63% of companies surveyed identified inaccurate output as their greatest gen-AI risk. Keep in mind, these are sophisticated, global F1000 organizations with serious data infrastructure.

That should get our attention.

AI does not repair a bad understanding of the customer. It operationalizes it. Then it amplifies it.

Today, one salesperson makes one bad decision from a bad record.

Tomorrow, AI can make that same bad assumption across hundreds or thousands of customers, selecting offers, prioritizing leads, suppressing customers, triggering workflows and personalizing communications before anyone notices.

That is the 100X problem.

Reporting creates the quieter version of the same risk. If identity is wrong, customer counts, attribution, retention, lifetime value, and campaign performance can all be wrong. Or, if you are marketing to partial records, you are reporting on partial records by definition. How can that be ok?

At that point, this is no longer a marketing problem. It is a management problem.

Before declaring your data “clean,” ask:

• Do we know who the customer actually is across our systems?
• Do we know why records are merged, or why they aren’t?
• Do we know which source to trust when systems disagree?
• Would we let AI act autonomously on this data tomorrow?

That last question may be the simplest test of all. We’ve spent years getting better at activating data.

Now we need to get equally serious about whether the data we’re activating is actually true.