What DMEXCO 2026 taught us about AI

DMEXCO 2026 is behind us, and if one question kept coming up in hallway conversations, customer chats, and the “how’s it going” at our booth, it was some version of this: We understand AI’s potential, but how do we make it work for us?

With “Scaling Intelligence” as the event’s theme, it’s hardly surprising that AI dominated the conversation. What stood out was how often people returned to the same problem. Those two days in Cologne, on September 23–24, revealed just how large a gap there is between what most people know AI can do and what they can actually do with AI.

I left thinking less about which tool teams should adopt next and more about what would make the tools they already have useful.

Key takeaways from DMEXCO 2026

• AI needs to earn its place in everyday work: A successful demo is a good starting point, but the real test is whether a team can repeat the result and check it.
• The urgency to adopt AI is running ahead of confidence in using it: Supermetrics’ 2026 Marketing Data Report found that 80% of marketers felt pressure to adopt AI, while only 18% reported high trust in it.
• AI delivers huge benefits if you invest time and energy into setting it up properly: Norwegian marketing agency Layer reduced bespoke reporting tasks from up to 10 hours to 20 minutes with Supermetrics for Claude, but this required several weeks of considered implementation.
• AI visibility is about more than getting mentioned: Brands need to check whether AI answers describe their products accurately and recommend them for appropriate use cases.
• Agentic commerce is changing how brands need to show up: As shoppers delegate research and comparison, brands need to make it clear how their products meet customers’ needs.
• My verdict: AI is a magnifier, not a silver bullet. It requires good data, a cohesive strategy, and well-defined processes in order to deliver dependable results.

Why is getting AI beyond the pilot so difficult?

The challenge is getting AI to work consistently when you’re accountable for the output, not just getting it to work once.

An impressive demo doesn’t necessarily mean an AI tool will deliver consistent results in everyday work. That requires reliable AI-ready marketing data, clear expectations, and someone who’s accountable for the output.

Most DMEXCO attendees had variations of the same story: seeing an impressive demo that completed a normally cumbersome task in minutes, with the promise that you could do the same at the click of a button.

Then you try it yourself. You bring your own data and expect an answer you can use. Suddenly, there’s more checking and troubleshooting than the demo suggested.

One figure came up repeatedly, in different words and on different stages: Gartner reported that at least 50% of generative AI projects were abandoned after proof of concept by the end of 2025. Why? Poor data quality, inadequate risk controls, rising costs, and unclear business value.

The gap between “it worked once” and “it works every week” felt like the real subject of many sessions, whatever their titles said. I don’t think that diminishes AI’s potential. But it does explain the frustration right now among many marketers.

You can see what AI is capable of yet still struggle to make it dependable enough for the work you’re accountable for.

What would make marketers trust AI?

Marketers need confidence that AI understands their business and produces answers they can verify. That confidence takes reliable data, clear instructions, and testing on real work.

The concern I kept hearing at DMEXCO was whether people could trust what AI gave them. Getting an answer was easy. Knowing whether it was good enough to use was another matter.

That distinction matters when a recommendation could change a client’s budget or a report could shape next quarter’s strategy. An answer can sound convincing without being correct. And if marketers have to investigate every claim before acting, much of the promised efficiency disappears.

Our own research shows the tension. Supermetrics’ 2026 Marketing Data Report found that 80% of marketers felt pressure to adopt AI, while only 18% reported high trust in it. Teams are being urged to move faster with tools they’re still learning to rely on.

That was the problem we built our DMEXCO session around, referencing Layer’s experience with Supermetrics for Claude. The Norwegian marketing agency used Supermetrics for Claude to reduce bespoke reporting tasks from up to 10 hours to 20 minutes. But they didn’t just connect AI to Supermetrics and hope for the best. They did the work behind the scenes: experimenting, validating, and purposefully creating a process they could actually trust.

I think the new-hire analogy is useful here. You wouldn’t expect someone to understand an account’s goals, metric definitions, and history on their first day. An AI assistant needs that context too. It also needs a review process that establishes where you can rely on it and where you still need to intervene.

Layer’s time saving is the catchy headline. But for marketers looking to do the same, the preparation is just as useful. The team invested time in checking the output before relying on it. That’s the commitment missing from the promise that AI will save you hours at the click of a button.

Why does AI need a reliable data foundation?

AI needs accurate, connected data to produce useful analysis. Missing information and inconsistent definitions can lead to misleading answers, even when those answers sound convincing.

The case studies that stayed with me at DMEXCO explained the work behind the AI results.

One retailer couldn’t get AI-driven measurement working until the team brought its point-of-sale and digital marketing data together. Connecting those sources gave AI the information it needed to produce useful analysis.

That preparation can get lost when we focus on the headline result. But for marketers hoping to achieve something similar, understanding what had to happen first is essential.

We’ve talked about marketing data quality at Supermetrics for years. That retailer’s experience showed why it still matters: AI can analyze data faster, but gaps and inconsistencies can make their way into its answers too if you’re not careful.

Watch our full talk, “It's not the tech, it's trust: How Layer turned 10 hours of reporting into 20 minutes", below.

What does AI search mean for brand visibility?

Brand visibility now includes appearing in AI-generated answers, and making sure those answers describe your business accurately. Ranking in search results is only part of the picture.

DMEXCO revealed an interesting tension. Teams are trying to make AI work for them. At the same time, they’re working to make their brands visible to AI.

This year, a new category of vendors showed up, promoting tools dedicated to improving, measuring, and reporting AI search visibility. Answer engine optimization (AEO) and generative engine optimization (GEO) are the new search engine optimization (SEO).

In traditional search, you compete for a place among the results people browse. In AI search, you also compete to become part of the answer itself. A prospective customer might read an explanation of your product, or a recommendation for a competitor, before visiting either company’s website.

But getting mentioned is only part of the job. Does the answer understand what your product does? Does it recommend you for an appropriate use case? Or is it repeating an outdated claim that could put someone off?

For me, that makes accuracy just as interesting as visibility. Knowing that your brand appeared tells you little unless you also know what the answer said.

That was one of my main takeaways from DMEXCO. We want AI to understand our businesses well enough to help us work. Increasingly, we also need it to understand our businesses well enough to explain them to potential customers.

Is agentic commerce becoming a reality?

Agentic commerce is moving into practical use, with AI agents helping shoppers research and compare products (and even make purchases). How widely customers will adopt it remains an open question.

At DMEXCO, agentic commerce came up often enough to feel like something marketers need to prepare for. The question is what happens when a customer delegates part of their buying decision to an agent.

That raises a new question for marketers: how do you make the shortlist when an agent does the research? What information do agents prioritize? Your website may explain your product brilliantly to a human visitor. But if an agent can’t establish whether it meets the customer’s needs, you could miss out before that customer ever reaches your site.

I haven’t left Cologne believing that agents are about to completely take over shopping. But I have started thinking that businesses need to begin preparing for what an agentic future might look like.

What we’re taking away from Cologne

Our best moments at DMEXCO weren’t on stage — they were at our booth and in the conversations we had walking the floor. Customers wanted to talk about the same thing everyone else did: not whether AI works, but how to make it trustworthy enough to run their business on.

Showing Supermetrics for Claude and Supermetrics Studio gave us something concrete to discuss. The demos opened conversations about the data behind the answers and how teams would use the output. They acted as proof that AI works when it’s built on the right foundation.

That's the takeaway we're carrying out of Cologne. AI is a magnifier rather than a shortcut. It amplifies teams with good data, documentation, and strategy. But if you lack any of these, then it can turn into a costly distraction.

Ready to put AI to work with your marketing data?

Frequently asked questions