AI Sorts, Humans Sell: The IKEA Math Almost Nobody Copies
Most companies use AI to cut headcount. The math is simple: automate inquiries, downsize the team, record the savings. It adds up — and yet it is still the worse of two possible calculations.
IKEA made the better calculation. It starts with the exact same technology and ends up not at a cost center, but at a sales channel. The difference comes down to a single decision: What happens to the inquiries that the AI cannot answer?
The IKEA Calculation
In 2021, Ingka Group, the largest IKEA franchisee, introduced the AI chatbot Billie. By 2023, it resolved around 47 percent of customer inquiries — roughly 3.2 million interactions. The simple, repetitive ones: delivery status, returns, opening hours, spare parts.
That left 53 percent. And this is where the two calculations diverge.
Under the conventional view, this 53 percent is the residual workload that you automate away with the next model generation. IKEA looked inside and saw something else: people who wanted to know how to furnish a kids' room, whether the kitchen fits the floor plan, which sofa belongs in the space. Not service inquiries. Buying intent that required consultation.
What the chatbot couldn't resolve wasn't a flaw in the system. It was a list of moments when a customer wanted a human — and was ready to buy because of it.
So IKEA reskilled 8,500 call center employees as interior design advisors. Today, they provide consultations by phone and video, by appointment, sometimes for a fee. The remote sales channel hosting these consultations generated around 1.3 billion euros in fiscal year 2022 — about 3.3 percent of total group revenue, with a stated target of 10 percent by 2028. The 1.3 billion represents the revenue of the entire channel, not just from the reskilling effort alone; nevertheless, it demonstrates the magnitude of the opportunity.
The same AI. The same 53 percent. Two entirely different outcomes — depending on whether you treat unresolved inquiries as leftover workload or as pipeline.
The New Division of Labor
What IKEA built is not a chatbot with a human fallback. It is a division of labor that I consider the most important organizational concept in customer service this decade:
AI sorts. Humans sell.
AI handles what is repetitive — fast, 24/7, with zero wait time. And it does a second thing that is missing in most implementations: it recognizes when an inquiry is not a service request, but a revenue opportunity. And it brings that opportunity to the person who can turn it into something real.
| Inquiry | Handled by | Outcome |
|---|---|---|
| "Where is my delivery?" | AI | Resolved in seconds, no human needed |
| "How do I return this?" | AI | Resolved, return label generated |
| "Does this kitchen fit my floor plan?" | AI recognizes → Human advises | Planning appointment via video, four-figure cart value |
| "I'm furnishing a kids' room and don't know where to start" | AI recognizes → Human advises | Consultation appointment, complete room solution instead of a single item |
| "I'm dissatisfied and thinking about canceling" | AI recognizes → Human saves | Relationship preserved, follow-on purchase possible |
The first two rows are what almost everyone builds. The last three are what generate revenue — and they only work if the AI doesn't log the inquiry as "unresolved," but treats it as a handoff.
The Handoff Is the Revenue Moment
This is the point where most customer service AI projects give away their value. The customer has just asked a question that reveals buying intent. The AI determines that it isn't responsible. And then what happens? A phone number. A contact form. "A representative will get back to you."
That is the most expensive break in the entire process. The customer was engaged in dialogue, had formulated their request, was ready — and gets dropped into a queue where their intent cools off.
The handoff needs to look different. Four steps that turn an unresolved inquiry into a confirmed appointment:
From Chatbot Dialogue to a Human with a Human Touch
Identify
Assign
Book
Prepare
Anyone who builds these four steps has turned their chatbot into a sales engine. Anyone who only builds the first one has created a more expensive version of the holding queue.
Why the Human Stands at the End—Not the Next Model Generation
The objection is obvious: What goes to a human today, AI will solve tomorrow. Kitchen planning via image generator, mortgage comparison via agent. I believe this misjudges what customers are actually paying for in these moments.
They aren't paying for information. They already have that. They pay for three things tied to a human being: Judgment—someone who says what they would do in their shoes. Trust—someone who puts their name behind the recommendation. Accountability—someone who is still there afterward. The higher the amount and the more personal the decision, the more weight this carries. A kitchen, a mortgage, a retirement plan—no one signs off on those with a system.
In these moments, AI will get better at recognizing that a human is needed and better at finding the right one. It won't replace them. It will keep them busy—with the conversations that drive revenue, rather than the calls checking delivery statuses.
The Same Model in Banking, Insurance, and Retail
Furniture planning is a textbook example because the need for advice is obvious. The mechanics are the same everywhere—only the handoff points differ:
| Industry | AI Solves | AI Detects and Routes | Human Appointment Type |
|---|---|---|---|
| Banking | Card block, account balance, payment status, address change | "We are thinking about buying a house," "What do I do with the inheritance?", "Is amortization worth it?" | Financing consultation, investment consultation, retirement consultation—with the relevant specialist |
| Insurance | Policy document, claims status, premium invoice | "We're having a baby," "I'm starting my own business," "Is my coverage sufficient after the renovation?" | Life event consultation, coverage review, retirement analysis |
| Retail | Delivery, return, availability, store hours | "I'm looking for an outfit for my daughter's wedding," "We're remodeling the bathroom," "Which watch for a 50th birthday?" | Styling appointment, planning consultation, private presentation |
In every row, the third column contains a sentence that is filed today in chatbot logs as “unresolved.” In every row, it is one of the most valuable inquiries the company received that day.
What this means for people
IKEA didn't hire new advisors. It upskilled the people who were already sitting in the call center. This isn't a coincidence; it's the core of the business case: These people already knew the company, the products, and the customers. What changed was the task — not the knowledge.
I believe three things from this are transferable to any organization with service touchpoints:
- AI eases work; it doesn't automatically make people redundant. Whether freed-up capacity is cut or redirected is a leadership decision, not a technical consequence.
- Service employees are not a cost center. They are the only people in the company who hear daily what customers really want — and, with proper training, the only ones who can turn that into revenue.
- Reskilling needs a home. Advisory services need their own structure: appointment types, calendars, unit economics. Anyone who trains service reps to become advisors and then leaves them in the same queue has only changed their job title.
The question that decides everything
Is your customer service managed as a cost center or as a channel? As long as it is a cost center, every AI initiative there will be measured by headcount saved — and every handoff to a human looks like a step backward. Only when deals closed from escalated inquiries are attributed to the channel does the evaluation shift. Then, the handoff is no longer a failure of AI, but its very purpose.
What you should measure
Automation rate, handling time, cost per contact — all three measure the cost side. Four metrics measure the other side:
| Metric | What it answers |
|---|---|
| Identified advisory requests per week | How many revenue opportunities the AI flags as such in the first place — the number that almost no one knows today |
| Share converted into a confirmed appointment during dialogue | Whether the handoff works or ends in a telephone number |
| Close rate per advisory call from the AI channel | Whether the human ultimately generates revenue or just takes on work |
| Revenue per escalated inquiry | The number that turns the chatbot into a sales instrument — and service into a revenue channel |
The second row is where things most often get stuck in practice. In almost all chatbot implementations I’ve seen, this value is close to zero — not because customers don't want it, but because the bridge was never built.
Takeaway
Most companies use AI to get rid of inquiries. IKEA used it to bring valuable inquiries to the people who turn them into revenue — 47 percent for the AI, 53 percent for 8,500 reskilled advisors, 1.3 billion euros in channel revenue. The underlying division of labor is transferable to any industry with advisory touchpoints: AI sorts, identifies buyer intent, and hands off — not to a queue, but to a confirmed appointment with the right person who enters the conversation prepared. That’s where the revenue is generated that no chatbot can achieve alone. And the handoff isn't a failure of AI. It is its purpose.
Sources
Key metrics on chatbot resolution rate (around 47 percent, 3.2 million interactions, 2021–2023) and the reskilling of 8,500 employees as interior design advisors: Ingka Group, newsroom press release "AI and Remote Selling bring IKEA design expertise to the many" (ingka.com).
Revenue from the remote sales channel of around 1.3 billion euros in fiscal year 2022, representing 3.3 percent of group revenue, as well as the target of 10 percent by 2028: reported in multiple sources, including robertjuliansmith.com and fluentsupport.com. This figure refers to the revenue of the entire channel, not the isolated effect of the reskilling.
The routing model (Identify, Assign, Book, Prepare) and the handoff points for banking, insurance, and retail are based on Calenso/jrni project experience with chatbot and appointment integrations in enterprise environments (2024–2026).
The inspiration for this article came from LinkedIn posts about the IKEA case, including by Markus Brunold. Contextualization and adaptation are original additions.
Pass this to your team
Three things a leader can implement this month after reading this article:
- Read the unresolved chatbot requests from the past four weeks. Don't count them—read them. Mark every single one that contains buying intent. This list is your pipeline, and it's already sitting in your system.
- Build the bridge. Where a phone number or a form appears today, there should be a bookable meeting—with the right meeting type, assigned to the right person, carrying the context from the conversation. This is the moment AI transforms from a cost-cutter into a sales tool.
- Introduce revenue per routed request as a metric. As long as handing off to a human is considered an automation failure, it will be minimized. As soon as it drives revenue, it will be built.