Nearly 90 percent of companies now use AI in at least one business function, and 44 percent have already scaled it across their organizations. Yet according to McKinsey’s “State of AI in 2026,” only 37 percent report a measurable impact on EBIT. The study, released in late August, shows little change from the year before. And only about 6 percent qualify as true “high performers”: companies that attribute at least 5 percent of their EBIT to AI and consider that impact significant.

Closing that gap won’t come down to the next, better model. What matters is whether someone is actually accountable for the outcome.

Anyone who has worked on these projects knows the model itself is rarely the biggest headache. More often, it’s the data silos. Customer data lives in the CRM, inventory sits in an ERP system that hardly anyone fully understands, and revenue figures are kept in an Excel spreadsheet maintained by one person on their laptop. Ask an AI model whether a product is still in stock, for example, and it may get two different answers from two different systems — then confidently pick one. The result is exactly what companies were hoping AI would eliminate: errors that someone has to clean up afterward.

Then there’s the question of ownership. IT handles the technology, the business team owns the use case, and management expects results. Yet hardly anyone is responsible for the project as a whole. And once the pilot is over? It ends up in a drawer alongside the three other experiments from last year.

So what do the companies that escape this trap do differently? They start by fixing their data foundation instead of immediately buying another tool. They integrate AI into the places where people already work — directly into the CRM or service desk, rather than adding yet another browser tab no one remembers to open. Most importantly, one person owns the outcome, instead of three departments passing responsibility back and forth.

In the end, whether a pilot turns into real progress comes down to a person. Not the algorithm.

The 6 percent that McKinsey says are actually seeing meaningful returns have something else in common: They don’t spread themselves across dozens of pilots at once. Nearly three-quarters have fundamentally redesigned their workflows around AI rather than simply layering AI onto existing processes. A year earlier, that figure was just 55 percent. They focus on a small number of use cases with real potential and measure success where it ultimately has to show up: on the P&L, not in vague metrics like hours saved.

A similar shift is happening on the customer side. Not every purchase journey starts with a Google search anymore. People ask AI assistants for recommendations, discover products while scrolling through social media, or search directly on marketplaces. Google isn’t going away, but it’s now just one of several entry points.

For brands, that has a very practical consequence: Products now need to be understandable to machines, too. That means complete product data, consistent descriptions, and credible reviews. Without them, the chances of showing up in a recommendation drop considerably.

Those two themes — moving AI from pilot projects into everyday operations and adapting to a new model of digital visibility — ran throughout DMEXCO 2026 in Cologne at the end of September. The theme was “Scaling Intelligence.” More than 40,000 attendees, around 1,000 speakers, and two days that kept circling back to essentially the same question: What does it take for AI to actually work in the real world?

“We need a human in the lead, not just in the loop.”

Verena Gründel, host and Brand & Communications Director at DMEXCO, captured the point at the opening, and, in many ways, the central idea of this piece: “We need a human in the lead, not just in the loop.” A day later, the line got an interesting echo. A jury of marketing executives named “The Return of Experience” the most important trend for 2027, after a year that had been dominated by AI.

Perhaps that’s the best way to frame the entire debate: What matters isn’t who has the best AI. It’s who knows what they want to do with it — and is willing to take responsibility for the outcome.

Missed a session? Most are available in the video-on-demand library. And the newsletter will keep following the topic between events. DMEXCO returns to Cologne on September 22 and 23, 2027.