It was revealing a need IKEA hadn’t fully captured before. Now the information flow was also inbound and that changed what happened next. 

​

At that point, IKEA had thousands of customer-service employees whose routine work was increasingly being handled by the chatbot. The obvious response would have been to reduce that workforce.

In its first full year, that business generated about €1.3 billion in revenue. Roughly 3.3% of IKEA’s total revenue at the time. IKEA has said it wants that figure to reach 10% by 2028.  

Think about these numbers for a moment.  

The chatbot saved €13 million by doing the work IKEA already knew it needed to do. 
 
But better understanding its users, by looking at the unanswered questions, pointed IKEA toward a new business that generated €1.3 billion.

That is a very different way of thinking about AI.  

If the business case for a chatbot is simply that it will close more support tickets and reduce call-center costs, the economics can be pretty thin.  

The problem is that we often evaluate AI against the task we gave it.  

We ask: How many tickets did it resolve? How many calls did it eliminate? How much time did we save? Those are useful measures. But they can also cause us to miss what the system is telling us.

A customer support bot is not just an automation tool. Once it is handling thousands or millions of conversations, it is also listening to customers at a scale most companies have never been able to manage.

The questions it answers tell you what customers need help with and most likely confirm what your business already knew. 

More interestingly are the questions it can't answer, because they may tell you what customers want that your product, process or organization isn't currently designed to provide.

That doesn't mean every unanswered question is a new business opportunity. Most aren't. But they can still tell you something the business didn't know before.

They may reveal a broken process, expose a confusing product, show that your documentation is poor, or occasionally point to demand your business isn’t currently designed to serve. 

That changes what we should expect from AI. 

Instead of asking whether the system completed the task, you really ask what the task taught you. Did it confirm something you already knew or is there something potentially new to learn here? 

We tend to measure automation. 

Therefore, we spend much less time measuring discovery and that is a challenge. If you only judge AI by what it automates, you'll optimize for efficiency, but if you also study where it struggles, you may discover entirely new opportunities. 

The challenge is making sure you don't throw that information away and this is where transparency becomes much more than a technical feature. 

If an AI system produces an answer but hides the evidence, the failed attempts, the patterns and the exceptions, you've built an automation tool. Sure that is useful, but it's not very educational. 

The real value comes when people can see what the system is seeing. 

What customers keep asking, where conversations break down, which patterns appear week after week, and which requests don’t fit your existing products or processes. 

At this stage AI stops being just a worker and moves towards becoming a research instrument empowering your business. 

That also changes how we think about trust as we don't trust AI because it never makes mistakes. Mistakes will happen, just like with everything else, but we'll trust it because we can inspect what it's doing, challenge its conclusions and learn from the evidence it surfaces. 

So the interesting problem isn't really whether the AI is right. It's whether the organization notices what the AI is revealing and that still requires people: product teams, operations, customer support, technology, and sometimes finance. Someone has to connect the dots and decide whether the pattern is noise, a process problem or the beginning of a new business. 

That's what makes IKEA's story so interesting. 

The chatbot wasn't the ending, it was the starting point, because the technology didn't create value simply by replacing work. It created value by helping IKEA see its customers more clearly than it could before. 

As more companies move beyond AI experiments and start looking for real economic returns, that distinction becomes increasingly important. 

We've become very good at asking one question: What task can AI automate? 

However, sometimes the better question is: What will AI teach us about our business once it starts doing that task millions of times? 

The former question looks for cost savings and the latter looks for insight. 

The first might save you €13 million. 

The second might show you the next €1.3 billion.