Caught in my Nette

52% of organizations now say data quality and availability is the single biggest barrier to using AI. This comes ahead of not having the right skills on the team, regulatory concerns, and plain old resistance to change. Source: PEX Report 2025/26 (PEX Network). In simpler terms, the thing stopping AI from working isn’t the AI. It’s what you’re feeding it.

In a race of AI-readiness, growth teams buy the tool first and ask the data question second, if they ask it at all. That order is backwards, and I think it’s the actual reason so many “we tried AI and it didn’t move the needle” stories exist.

The stat that should worry you more than it does

52% (coincidentally) of marketing teams don’t have clear ownership of their own data strategy. Not just “don’t have enough data” which signals a tracking issue. Data is currently owned by the data, tech or product team and sits in a CDP or an analytics platform, technically collected, functionally unused. Only 6% of teams say they have fully embedded data-driven workflows, and 40% still can’t confidently prove ROI across channels even though the underlying numbers exist somewhere in their stack. Source: Supermetrics 2026 Marketing Data Report.

Meanwhile the pressure to adopt AI is real and getting louder: 80% of marketers feel pressure to adopt it, but haven’t actually embedded it into how they work. That gap in this high pressure, low capability chamber is exactly where a rushed tool purchase happens.

And the tracking layer underneath all of this is messier than most people assume. 95% of enterprises report real integration challenges getting their data sources to talk to each other. On the identity side, only 15% of marketers felt fully ready for a cookieless world as of last year, even though 81% had already adopted some privacy-first measurement approach — meaning most teams are running mixed tracking setups where signal loss is baked in, not a one-time transition problem.

AI doesn’t fix a data problem. It just scales it faster

Agentic AI systems now execute campaign optimizations in milliseconds. That’s usually pitched as the benefit. It’s also the risk. A human running a weekly budget reallocation might notice that Meta and Google Ads disagree on conversion counts before making a call. An AI system will just optimize toward whichever number it was fed, confidently and immediately.

Bad data used to cost you slowly, through a marketer making a slightly-wrong call once a week. Bad data feeding an AI system costs you continuously, at machine speed, without anyone in the loop to catch it. That’s not a reason to avoid AI. It’s a reason the data audit has to come first, not as a nice-to-have cleanup project you get to eventually.

The financial version of this problem is well documented even outside AI: enterprises lose an average of $12.9M a year to poor data quality, 42% of CRM records contain errors or duplicates that distort attribution, and teams waste 15-20% of ad spend on targeting errors and inconsistent campaign structures. Every one of those numbers gets worse, not better, once you point an AI system at the same broken pipes and ask it to move faster. Source: Gartner.

It doesn’t help that the AI governance layer is thin, too: only 43% of organizations have an actual policy in place, and 29% have none at all. So the typical setup right now is unaudited tracking data, feeding a fast-moving AI system, with no policy layer catching either problem before it compounds. Any one of those three gaps on its own is manageable. Stacked together, it’s a risk that honestly none of us know the real impact of right now.

Next step: Run a data and tag audit

Here’s what I’d actually do, and it doesn’t require a new tool. Just an afternoon and some honesty about what you’ll find:

1. Run the reconciliation check. Pull last month’s conversion numbers from Meta, Google Ads, and your analytics platform (GA4 or whatever you’re on) side by side. If they don’t agree within about 5%, you have a tracking problem, and any AI optimizing budget across those platforms is optimizing against noise.
2. Map every tag actually firing. Not the tags you remember installing. Pull the ones live on the site or app right now. Duplicate pixels, tags from a tool you cancelled two years ago, events firing twice, conversions attributed to the wrong action. Most sites accumulate these the way a garage accumulates boxes: nobody removes anything, they just add.
3. Assign an actual marketing owner. Given that 52% of teams have no clear data-strategy ownership, this step alone probably fixes more than any tool would. Someone specific owns whether tracking is accurate, and moves the data towards marketing actionability.
4. Test traceability. Can you trace a data point to the action you’re tracking? If the honest answer involves a multi-day manual reconciliation process, that’s your answer on whether you’re ready to hand real-time decisions to an AI system.
5. Fix before you feed. Whatever the audit turns up (dead tags, duplicate events, unowned data), it needs to get cleaned before it feeds into a personalization engine, predictive scoring model, or agentic optimization tool. Even in pilot stage.

Here’s what that might surface in practice: a mid-size B2C team runs the reconciliation check and finds Google Ads reporting 1,400 conversions for the month while GA4 shows 1,150 for the same campaigns — an 18% gap, well outside that 5% threshold. Digging in, they find a checkout-confirmation tag firing twice on mobile due to a page reload, inflating one platform’s count. Nobody had looked at that tag in over a year.

Note to self: AI seems to have amplified the gap in growth fundamentals and speed of implementation. Good to go back to basics to good old first principle thinking and guiding design thinking frameworks.

---

Want to be alerted once a new post is out? Stay “In the Nette”.