Almost everyone read the IKEA story backwards this week. It crossed my LinkedIn feed again with the usual headline: “IKEA automated support and fired no one.” Nice, but it hides the part that actually helps a marketing team.

So here is the thesis, aimed at you if you run marketing or growth: the metric you use to judge your chatbot is the wrong one. You are proud of the share the bot resolves on its own. The value is in what the bot does not resolve, because that is a demand map written in your own customers’ words. IKEA read it. Most teams delete it.

The number everyone quotes, and the one that counts

The facts first, because they are real and verifiable. Ingka Group, IKEA’s largest franchisee, launched a chatbot named Billie in 2021. Between 2021 and 2023 it handled about 3.2 million interactions and resolved roughly 47% of inquiries, saving around €13 million. CIO documented it and PYMNTS covered it.

That 47% is the number that fills the slides. Containment rate, savings, efficiency. Fine, but it is the boring part.

The number that changes the business is the other one: the 53% Billie could not resolve. And the point is not that the bot failed. It is what kind of inquiries those were. They were not shipping or returns questions the bot should have caught. They were people asking for help planning a room, combining furniture, solving an awkward space. Demand for a service IKEA did not yet sell, arriving on its own, every day, through the support channel.

Marketing pays a fortune for surveys and focus groups to guess at that. IKEA had it for free in its own chatbot logs.

Unresolved demand is your best market research

This is the part I want you to keep, because it applies whether or not you run a chatbot the size of IKEA’s.

Every conversation your customer starts and your system cannot close is a statement of intent. It tells you what they wanted, in their words, at the exact moment they wanted it. It is the cleanest demand signal there is, without the bias of a survey where people answer what they think you want to hear.

Most teams’ reflex is to treat those conversations as an efficiency problem: “we need to train the bot to answer that too.” Sometimes, yes. But sometimes what the bot fails to answer is not a programming gap, it is a catalog gap. And that gap, read in time, is a new product.

IKEA made the right read. Instead of asking “how do we get the bot to answer the 53%?” it asked “what is that 53% asking us for that we do not sell today?” The answer was a premium design advisory service. And rather than fire the 8,500 agents the bot had freed from routine work, it reskilled them to deliver it, over phone and video.

That service booked about €1.3 billion in 2022, 3.3% of Ingka’s revenue, on track for 10% by 2028. AI did not replace the people. It moved the people toward the work that actually grows the business, and turned a cost center into a service brand along the way.

It is the flip side of what I wrote yesterday about Meta: same technology, opposite leadership decision, opposite result. Meta used AI as an excuse to cut and burned its team’s trust. IKEA used it to discover a business and moved its people up a level. The model was the same. What changed was the question they asked.

Why marketing is the natural owner of this signal

Here is something worth claiming for your function. Unresolved-conversation data usually ends up buried in support or operations, treated as a ticket dashboard. But it is not a support problem. It is demand intelligence, and that is marketing’s turf.

Whoever reads those conversations at scale can see patterns no campaign hands you: which segment asks for what, in what language, at what stage. That feeds messaging, shapes offers, and, as with IKEA, sometimes defines a whole product. The difference between a team that sees “tickets the bot did not close” and one that sees “the next service my customers are asking for” is entirely a matter of view, not tools.

I made the same point from another angle when I explained why it pays to build the system instead of buying a chatbot: a bot that only contains inquiries is a savings tool. A system that also turns every failed conversation into a demand signal is a growth asset. Same AI, different intent behind it.

What we do at IQ Source

When we help a marketing team bring AI into its customer relationship, we do not measure success by how much the bot contains. We measure it by how much the company learns from what the bot does not contain.

In practice that means instrumenting conversations so unresolved demand does not vanish into a support board, but reaches marketing clustered and readable. It means deciding, with that data, where it pays to move a person up to a higher-value service instead of deleting the role. And it means designing that new service when the signal justifies it, the way IKEA did, instead of leaving the opportunity buried in logs nobody reads.

There is not always a €1.3 billion service hiding in your chats. But there is almost always more than a saving, and you will never find it if your only metric is the containment rate.

Next time someone shows you the share the bot resolves on its own, ask a different question: show me the other share, the one it failed, and what it is asking us for. There, not in the savings, is the business.

Turn your customer conversations into demand signal