Don't Trust People Who Say "I Don't Trust the Data"

"I don't trust the data" is the most expensive sentence in any meeting.

It costs you the next 30 minutes. It makes at least two people visibly sad. And worst of all, nobody decides anything, because now nobody knows what to believe.

Look, I don't trust a lot of things. I don't really trust McDonald's. I still eat it occasionally. I definitely don't trust myself after a few shots of tequila, and yet, historically, that has happened.

Heck, some people don't trust me. Totally fair. I'm quite unpredictable!

But when someone says they don't trust the data, what they're usually announcing is that they didn't build it, don't own it, and would prefer not to. That's fine. Just say that part out loud instead.

I never say "I don't trust the data" about anything I actually own. I trust my numbers because I did the boring work. By boring work I mean ripping out the three analytics packages sitting crustily in the theme since 2017, quietly double-counting everything, installed by someone who left two reorgs ago. By boring work I mean asking marketing for the fifth time to use the right UTMs, and then also asking the lord for forgiveness for our data sins of the past.

At the same time, I have never expected perfection from any data I've worked with. It's a dashboard, not the Fibonacci sequence. Things are wrong or broken all the time. Cookie consent, GDPR, and whatever EU regulation launched this week don't help in this quest for perfection.

So I don't seek perfection. I'm looking for signal. A direction. A guide good enough to bet the next two weeks on.

I do, however, have real sympathy for anyone who got burned by bad data years ago. It was genuinely hard! Everything needed a custom integration, half of them broke silently, and getting two platforms to agree on just about anything could take a month. Not to mention that getting developer time to integrate anything was only slightly less uncomfortable than waterboarding.

That is not the landscape anymore. The integrations today are generally rock solid. Whatever you use, whatever you want to pipe it into, there's usually a good one natively right out of the box.

And then, because it is 2026 and no blog post escapes this topic, there's AI.

Last week I fed Claude a pile of campaign exports, and it came back proclaiming (very proudly, I might add) that someone had botched the attribution on this whole ad set and that several of the links pointed to the wrong destination entirely.

That someone was me.

Then it did the part that matters. It wrote the mess to memory. Now every time it touches that data it already knows which campaigns are wrong and how to adjust for it. Beyond the slight data shaming, it just moved on with its life and got to work. Now and every time after.

I know hallucination is the common objection here. Totally get it, and if you have trust issues with data, you probably have serious trust (and commitment) issues when it comes to AI. That's understandable. But when you give a model actual source data, I have not seen it happen once. Not one invented number. If you're really unsure, hand it multiple exports, MCPs, APIs, whatever you have, and ask where they disagree. That alone is illuminating, and it will solve most of your data issues.

AI can go further than that. It can tell you what's wrong and how to fix it. I find myself prompting like this a lot these days:

"Open xyz.com, review the code and analytics integration, reconcile this with the actual data from the API, tell me any issues and how to fix them."

Then you paste the response and send it to the developers. Or better yet, push the fix yourself.

So, in closing. "I don't trust the data" used to be a reasonable position. Then it became a hedge. Now it's just a status update of a very important part of the job someone is ignoring.

The numbers were never going to be perfect. They were only ever going to be yours.