AI & ML
77% use AI, and 5% make decisions based on it.
A look at the 2026 Dutch E-Commerce Monitor, and what it has to say to us on this side of the border.
That contrast between 77% and 5% sums up the current state of AI in marketing. Almost everyone uses it, but almost no one lets it drive their decisions. The rest are stuck using standalone tools.
The third figure really drives the point home. Of the companies that use AI, 84% use it for content creation: text, images, ads, and translations. That’s the entry point—visible and accessible. But it doesn’t change the way you make decisions.
You can do what you've already done faster. The engine is roaring, but the car isn't moving.
The Gap Between Adoption and Integration
The real question isn't whether companies use AI, but what they do with it. The E-commerce Monitor identifies four phases:
Among growing companies, 47% are already in phase 2 or 3. Among stagnating companies, that figure is 27%. In phase 1—the stage of standalone tools—growing and declining companies are represented in roughly equal numbers.
Source: E-commerce Monitor 2026 (525 Dutch e-commerce companies)The difference, therefore, does not arise during the experimentation phase, but in the step that follows. The biggest gap between the two groups lies, in the researchers’ words, in “data and analyses from their own systems.”
Why Most Pilot Projects Get Stuck
If the technology is available and the willingness is there, why does AI remain limited to faster text writing at so many companies? The answer is not what most people expect. It’s not the technology that’s the barrier, nor is it the budget. 65% of companies cite a lack of knowledge and skills as the biggest obstacle, and only 23% rate their own team’s AI knowledge as strong.
But there’s a deeper layer beneath it all. A model designed to predict customer behavior or determine the next best action requires something that content creation doesn’t: reliable, consistent data about the entire customer. And that’s exactly where the problem lies. The data is scattered across the point-of-sale system, the online store, the CRM, and the email tool—four silos that don’t communicate with each other.
An AI model applied to fragmented data doesn't fix the fragmentation. It just makes it visible.
The prediction is off, the model is viewed with suspicion, and the pilot project fizzles out. Not because AI doesn’t work, but because the foundation is missing. It’s like a powerful engine in a car without wheels: the technology is brilliant, but there’s nothing to transfer that power to the road.
First, the foundation: from isolated interactions to a unified view of the customer
Here lies the solution, and while it’s less spectacular than a new model, it’s far more decisive. The companies that are successfully implementing AI in production first consolidated their customer data in one place.
One place. One language. One customer view.
That’s exactly where we start. Not with the model, but with the foundation beneath it. Through our MIP approach (the Master Data and Single Customer View initiative), we bring all customer data together in one place: in-store purchases, online behavior, email interactions, and service contacts.
We standardize that data so that “active,” “customer,” and “value” mean the same thing everywhere, and we use it to build a single, cohesive view of the customer. With that, you can piece together every interaction a customer has with your brand into a story rather than isolated signals: who bought what, through which channel, after which email, and what their service experience was like afterward.
That’s where the insights come from that a model needs to predict who is at risk of churning, which customer group to target at full price, and which actions are most effective at which moments.
Only on that foundation will AI become something more than just a faster text generator. Then it will inform decisions, rather than simply producing content.
So the transition from Phase 1 to Phase 2 is not an AI step. It’s a data step. The question isn’t “which model are we going to run,” but “what are we going to run it on.”
Not the demo, but the integration
These days, anyone can put on an impressive AI demo—a model that does exactly what it’s supposed to do in a controlled environment. The demo is easy. The integration is difficult, because it requires the underlying data to be accurate.
That’s why the most accurate figure in the entire Monitor may well be that 5%. The difference between 77% and 5% isn’t a matter of better models, but of a better foundation. So don’t start with the model. Start by asking whether your data is ready to build on.