By Brian Silver, EVP of Global Marketing Solutions at TransUnion

At CES this year, AI was everywhere. But the thing that stood out most was not a flashy demo or a new use case. It was a statistic: 60 percent of AI projects will be abandoned this year because companies are not ready from a data standpoint.

With numbers like that, it is hard not to be skeptical. Anyone who has worked in this industry knows marketers love new tools. But when new strategies fall flat, it is rarely the tool’s fault. It is the data.

Even the most advanced systems cannot deliver results if the foundation is weak. When performance stalls and budgets shrink, it is usually because the integration was never set up to succeed. Bad inputs do not become useful simply because a model processed them.

I have seen brands invest in sophisticated AI only to find the results are underwhelming. The reason is simple. They never cleaned their data. They never resolved identities. They never connected systems across their stack. Instead of accelerating growth, the technology spins in place. In some cases, AI becomes a quiet budget drain. It runs, but it does not move the business forward.

One recent study showed that when machine learning models were built on identity-resolved data, precision improved by nearly 20 percent. That meant fewer wasted impressions, more relevance, and better ROI in high-cost channels like catalogs and paid media. You don’t get that kind of lift by layering AI on top of confusion. You get it by giving the system something solid to work with.

Identity work isn’t glamorous. Demos don’t typically leave you on the edge of your seat. But it’s the difference between AI that guesses and AI that performs. When you start with data that reflects real people, with real behaviors and verified connections across channels, your models have something to build on.

The good news? Smarter targeting and better identity signals don’t just improve AI — investing in your data infrastructure can (and does) have knock-on effects across your entire stack. A rising tide lifts all boats. One study found that campaigns targeting audiences with just two or three optimal traits saw performance gains of up to 7x. Meanwhile, campaigns aimed at the wrong audience groups lost as much as 90 percent in return.

That’s the kind of performance gap that no model can patch over.

So if you’re investing in AI, here’s the real question. Not what the model can do. Not how fast it runs. But how well you’ve set it up to succeed. Because if the data going in isn’t right, the outcomes will never be.