What AI Doesn't Take Away, But Reveals

TL;DR

• Wrong question. Most companies ask what AI can take over — and save €13M doing so. That's the least valuable application.
• What IKEA discovered. The Billie chatbot saved €13 million, but the 53% of unanswered questions fed a design channel worth €1.3 billion.
• The mechanism. AI removes routine work and thereby reveals what was buried beneath it: the questions and projects that were always "too expensive."
• The better question. Stop asking what AI can take over. Go read your own unanswered questions.

Last week I watched a colleague build a working framework in eight hours on a fifteen-year-old system. Built by a vendor that no longer maintains it. A job that normally takes 160 hours.

That sounds like a productivity trick. But it's something else. This was a project that would otherwise never have started.

Too expensive, too complex, too much risk on a system no one dared touch anymore. It had been on the "someday" list for years. Until AI lowered the threshold so much that someday suddenly became last week.

And that's precisely the point most people miss when they talk about AI. The question isn't what AI takes away from you. The question is what it reveals.

The 53% nobody looked at

In 2021, Ingka Group — IKEA's largest franchisee — launched a chatbot called Billie. Named after the Billy bookcase. Billie did what such bots do: delivery status, returns, stock checks, opening hours. The boring, repetitive work that floods a call centre every day.

And Billie performed well. Between 2021 and 2023, the bot handled roughly 47% of all customer inquiries — around 3.2 million interactions — saving nearly €13 million. This is stated directly in Ingka Group's own press release (opent in nieuw venster), confirmed by Reuters (opent in nieuw venster).

For most companies, the story ends here. A deflection rate of nearly 50% is a fine number for the boardroom. It justifies the investment and delivers the savings. And then comes the calculation nobody says out loud. If the bot does half the work, you don't need half the people anymore.

IKEA didn't make that calculation.

Instead, they looked at the other side. At the 53% that Billie couldn't resolve. And there was no noise there. There was a pattern. Customers weren't just calling to check whether a sofa was in stock. They wanted to know if that sofa would fit in their living room. They asked for help with layouts, colour combinations, storage solutions. They wanted advice.

That wasn't a system failure. That was a market that had remained invisible for years, simply because the people who could have spotted it were too busy with delivery statuses.

Looking inward, or looking outward

Here comes the decision that carries the whole story. IKEA could have cut the freed-up capacity. That would have been the obvious choice. Instead, they retrained 8,500 call-centre employees as remote interior design advisors. No redundancies — a new service: paid design sessions via phone and video.

One nuance I want to add immediately, because it gets systematically misrepresented on LinkedIn. You'll read everywhere that "the chatbot generated €1.3 billion." That's not accurate. The €1.3 billion is the revenue of the entire remote design channel in fiscal year 2022 — roughly 3.3% of Ingka's total turnover. That channel wasn't created by the bot. The bot made visible the demand that feeds the channel. That's a distinction. Anyone who says otherwise is selling a fairy tale.

But even with that nuance, the contrast stands. Looking inward for savings gave IKEA €13 million. Looking outward for opportunities fed a channel worth €1.3 billion. As CIO magazine (opent in nieuw venster) put it sharply: a hundred times more — to the point that the company barely mentions the original savings anymore.

The technology was the least interesting part of this story. A decent chatbot, some language processing, a clean handover to a human. You can build most of it today with off-the-shelf tools. What makes the story work isn't the tech. It's the question someone asked. Not "how much can we save?" but "what does this data tell us about what our customers actually want?"

AI is a measurement tool, not a cost-cut

This is the mechanism I want to leave you with, because it transfers to any company, regardless of scale.

AI removes routine work. That's true. But in doing so, it reveals something that was previously buried under that routine work. The questions you never got around to. The complaints that disappeared into a mailbox. The projects that have been sitting at "too expensive, too complex" for years. They don't disappear because you ignore them. They only become visible when space opens up to see them.

IKEA's chatbot wasn't an answer machine. It was a listening instrument. Every question it couldn't answer was free market research. The most honest customer feedback a company ever gets — because nobody fills in a survey, but everyone asks a question.

And that's precisely why "what can AI take over from me?" is the wrong question. Anyone who only asks that ends up with a more efficient company and an empty agenda. The better question is: what will my people do with the time that's freed up? Because that time isn't a void to cut away. It's capacity to reinvest in work that was previously out of reach.

The difference between these two attitudes is easiest to see side by side:

You see the same pattern everywhere

I've worked in business for over thirty-five years, and I've watched more technology waves wash through than I care to count. Always the same pattern. A new technology arrives, and the first thing companies want to do with it is save money. Not grow. Save money.

With AI, exactly the same thing is happening right now, across every industry simultaneously. Listen to the conversations around you. It's about relieving phone pressure, letting a bot handle simple queries, how many FTEs you'll need going forward. All inward-facing. All the 47%. Almost nobody starts the conversation at the question of what emerges from beneath that routine work once you remove it.

Yet that's where the real numbers are. McKinsey estimates in a recent report on AI in insurance (opent in nieuw venster) that generative AI can unlock $50 to $70 billion in revenue in that single sector. The biggest impact isn't on the operations you cut. It's on marketing, sales and customer contact. On the work you can finally do well once there's room for it.

And what applies to insurance applies to virtually every service provider. Companies that only use AI to cut costs leave most of the value on the table. That's not a criticism. It's a mental model we all carry: seeing a new technology as a cost-cut first, and only much later as a growth engine.

The second face: the projects that stayed on the shelf

So far this has been about the customer side. But the same mechanism applies to your own organisation — and that's where it gets genuinely concrete for most mid-sized companies.

Back to that eight-hour framework. The remarkable thing wasn't the speed. It was that the project existed at all. A rebuild on a fifteen-year-old system was simply too risky and too expensive to start. Especially on a tool the builder no longer maintains. The business case broke down every time on the hours. 160 hours for something you weren't even sure would work — no SME director approves that.

At 8 hours, that equation looks completely different. The project shifts from "someday, when we have time and budget" to "this week." And that's not time saved. That's a project that went from impossible to possible.

I see this in my own practice more and more. I build the foundation and design of such a tool myself using vibe coding, without a formal development background. Not because I want to replace developers — quite the opposite. The business can see the direction early and course-correct. That means the developer can build the real framework much faster afterwards. The vibe coder and the developer aren't competitors. They reinforce each other. The fact that building becomes accessible to everyone doesn't mean everyone has something to say — the real distinction isn't in who can build, but in who knows what's worth building.

And this is where the SME reader should recognise themselves. You don't need 8,500 employees to leverage this mechanism. Many mid-sized companies have a solid IT department, even if a C-level AI function is often absent. That department is probably sitting on a backlog full of projects that have been labelled "too complex" or "too expensive" for years. Precisely those projects are now coming within reach. Not because your people are becoming redundant, but because they finally get to the work that matters. Cheaper production rarely leads to less work — it leads to more things being built, as I wrote earlier about the golden age of the programmer.

Stop asking what AI takes over

This is the thinking error most companies fall into. They treat AI as a cost line that must pay for itself by replacing something or someone. They look inward, find their €13 million in savings, and stop there.

The companies that will win over the next decade ask a different question. Not "what does this technology replace?" but "what does this technology make possible?" That distinction seems small. It's the difference between a company that shrinks and a company that grows.

You don't need a billion to see this. Maybe your opportunity is worth €50,000 a year. Or one customer segment you're not serving yet. Or that one system that's been rattling for years. But it's almost certainly there. The only reason you haven't seen it yet is that your people are too busy with the work AI is about to take off their hands.

So the concrete step for tomorrow isn't a tool or a subscription. It's a reading assignment. Go read your own 53%. The questions your customers keep asking that you never had a good answer to. The projects that have been on "someday" for years. The complaints you dismiss because there's no time.

Your margin is there. Not in what AI takes away. In what it finally makes visible.