You Know Your Data Is Bad. Why Are You Still Letting AI Make the Calls?

Last week, I had dinner with a friend who works in marketing.

Halfway through the meal, he pulled out his phone to show me the AI agent his company had just rolled out. The thing is impressively capable: when lead scores shift, it adjusts the ad spend on its own; when it decides a batch of customers is due for outreach, it sends the campaign. No human sign-off anywhere in the process.

As he demoed it, he marveled: this thing really is good.

I asked him: is the data in your CRM clean?

He froze for a second. Honestly? Not that clean. Duplicate contacts, outdated job titles, company names filled in any which way. But the AI does seem to be humming along?

I said: that's not humming along. That's flooring the gas pedal with your eyes closed.

Later, I went through Validity's State of CRM Data Report 2026. And I realized my friend is not an outlier. Behind him stands a very large crowd.

Here is that crowd in one sentence: knowing full well the data isn't ready — and handing decision-making power to AI anyway.

Numbers at War with Themselves

First, the survey basics: 500 B2B and B2C marketers, all at companies with 100+ employees, spread across the US, the UK, Brazil, Australia, and New Zealand, surveyed in July 2026.

My first reaction after reading: these numbers are at war with themselves.

91% of marketers say that to get the most out of AI, the data has to be ready first. What does that mean? That 91 out of 100 people know: if the data is no good, AI is worthless.

So how many believe their own CRM data has ever actually been ready for AI?

21%.

91 out of 100 know what should be done; only 21 believe they've done it. The remaining 70 keep preaching how important data is while feeding dirty data to AI one spoonful at a time.

Two more numbers. Only 26% of companies dare claim that more than three-quarters of the data in their CRM is accurate and complete. Nearly half admit that data quality has been tormenting them for years.

Between knowing and doing sits an entire CRM.

Dirty Data Used to Stain Reports. Now It Stains Decisions.

Some will say: so the data is dirty — so what? It hasn't been clean in ages; didn't we get by just fine?

Right, we got by. But there is one fundamental difference between then and now.

Back then, what was the worst a dirty dataset could do? A report that looked bad, some manual rework, another round of cleaning. The errors lived on paper. You always had a chance to spot them, a chance to fix them.

And now?

45% of marketers are already using agentic AI — agents that get work done on their own, executing without human review. Two-thirds of companies handed more marketing decisions to these agents over the past year.

What is an agent? The kind of AI that has already sent out your campaign before you've even woken up.

Lead score wrong? It schedules sales follow-ups on the wrong score anyway. Customer data stale? It sends personalized offers to people who left their jobs long ago anyway. Budget miscalculated? It moves the money into the wrong channels anyway.

Dirty data used to stain reports. Now it stains decisions.

And no human eyes review any of it.

Why Do the Higher-Ups Fall Into More Traps?

There's one number in the survey I stared at for a long time.

19% say they often act on the advice AI gives them, and only afterwards start to wonder: is bad data sitting underneath that recommendation? Another 43% say they've hit this occasionally. Add them up: more than 60%.

The interesting part: the ratio climbs with seniority.

Among C-level executives, 78% have stepped in this trap. SVPs and VPs: 92%. Frontline employees? Only 41%.

Why? My guess: the higher the position, the farther it sits from the data.

Frontline people live inside the CRM all day; when the data starts to stink somewhere, their noses catch it first. Executives? They see only the finished dish the AI serves up, plate polished to a shine. Whether the ingredients behind the scenes were rotten — they never get the chance to find out.

Of course, that's my speculation. But one thing is certain: when the data has problems, the higher up a person sits, the more they rely on that one number the AI has processed for them.

And what's it like to take that number into a meeting? Nearly 69% say that revenue, pipeline, or performance numbers reported by them or their teams were challenged on the spot — or walked back afterwards — because the underlying data was wrong. At the executive and director level, that figure approaches 75%.

Worse still: only 28% dare to swear that the performance their CRM shows them is accurate.

In other words, everyone knows bad data is costing them money. 62% say dirty data has cost them renewals, thrown off their forecasts, lost them orders, and sent campaigns to the wrong audiences. But no one can put a number on how much is actually being lost.

Because the ledger you would use to run the numbers is itself a muddle.

At the Root, It's "Everybody's Job Is Nobody's Job"

So why does the data never get better? It can't be that everyone is lazy.

Dig down, and it's an organizational problem.

39% of executives say marketing, IT, and RevOps (revenue operations) cooperate reasonably well and the data is kept in usable shape. Among senior managers and frontline employees, that number drops to 27%.

Looks fine from the top; already falling apart down below.

What about data governance? Only 41% of companies have a dedicated team, or a clearly named owner. At the rest, data is like the living room of a shared apartment: everyone uses it, nobody cleans it. When it gets dirty, everyone frowns; once they've frowned, they go right back to whatever they were doing.

Marketing blames sales for sloppy data entry; sales blames the system for being hard to use; the system side blames requirements that were never made clear. And CRM data, passed between departments like this, slowly rots.

Then, AI arrived. It doesn't care whose responsibility it is; it only knows the data. Whatever you feed it is exactly what it executes.

So What Do You Do?

The survey asked this too: what would make you more confident in your CRM data?

The top answer, chosen by 39%: continuous automated monitoring that spots problems and fixes them in real time.

Note that word: continuous. Not an audit once a year.

Data is alive. Customers change jobs; titles change. The data you just finished cleaning last month starts to go stale again this month. Counting on an annual audit is like counting on a once-a-year physical to judge how your body is doing today.

But AI doesn't wait for your checkup report. Across the entire year while you audit, it makes decisions every single day.

Second: a unified platform that gathers the data into one place. Third: bringing in a third party to validate and enrich the data.

But whichever road you take, the order cannot be shuffled: pour the data foundation first; then invite AI upstairs.

As dinner wound down, my friend said the first thing he'd do back at work: find a clear owner for the CRM data.

I said: for that one line alone, dinner was worth it.

Knowing full well the data is bad and still handing the steering wheel to AI — that isn't trusting AI. That's gambling.

Win the bet, and it's efficiency. Lose the bet, and it's an incident. And the odds need no guessing — they're written in the report: 91 know. 21 do.

Here's wishing you this: before you let AI make decisions for you, make one decision for your data first.