The version of this story going around overstates two things, so start with what the sources actually say.
IKEA put Billie, an AI/NLP customer-service chatbot, in front of customer inquiries. Ingka reports it resolved close to half of the inquiries it received between 2021 and 2023, several million interactions, at savings in the low tens of millions of euros. Separately, Ingka says it reskilled around 8,500 remote customer support workers so they could take on more skilled work, including remote interior design, digital sales, relationship building, and complex problem solving.
The retelling usually compresses those into cause and effect and attaches the €1.3 billion figure to the reskilling. That figure is sales through Ingka's remote customer meeting points in FY22, not revenue created by Billie or by the reskilling program.
The 8,500 figure is also softer than the retelling suggests. It refers to workers Ingka says it reskilled as customer support work changed. It does not establish that Billie displaced 8,500 jobs.
What is true, and is the part worth your time: a company automated a large volume of routine work while reskilling thousands of the people doing it for more complex work.
That was a decision. It is not what automation does on its own.
The decision looks different at five people
Every version of this story I have read is written for someone with a workforce plan. Redeploy rather than cut, invest in reskilling, think long term about talent. Reasonable advice for a company with thousands of people in one function and a training budget.
If you have five employees you do not have a function. You have Maria, who does invoicing.
So the IKEA lesson does not transfer as written. What transfers is the underlying question, and at your scale it is sharper, not softer.
What you actually lose
Automate invoicing and you save the hours. You also lose the person who knew where the process was wrong.
That is the part worth sitting with. Your documented process describes the normal path. Maria has spent eleven years quietly handling everything the normal path does not cover.
She knew which customer always pays late but always pays, so chasing them at day 31 damages the relationship for nothing. She knew the one who disputes every third invoice as a negotiating tactic. She knew that purchase orders from one client arrive with the reference number in the wrong field, every time, and she fixed it without mentioning it because it took four seconds.
None of that was in her job description. Some of it is not written down anywhere. Automation encodes the documented process, which means it encodes the version that was already wrong, and you find out where the gaps were when the system starts flagging your best customer as delinquent.
Not all of it belongs in the same place, either. The purchase-order reference in the wrong field is a rule waiting to be written: catch it, fix it, done. Deciding whether a chronically late customer deserves another week depends on a relationship history a rule cannot see. Some exceptions become rules. Others tell you where the automation should stop and hand the decision to a person.
A large company loses this too. It just has more people who half know the same things, and enough slack to rediscover them. You have Maria, and then you do not have Maria.
That is a practical case against cutting, and it has nothing to do with being generous. Rehiring the salary is easy. Rehiring eleven years of exception handling is not.
Where the freed hours go, honestly
At IKEA the answer was a training program and new roles to point people at. At five people it is a conversation about what Maria does on Tuesday instead.
The useful framing is not "what job do we move her into." It is "what valuable work has been sitting undone because nobody had the hours, and which of it can this person reasonably become good at."
Both halves of that matter. Most small businesses have a long list of neglected work: customers who went quiet, the three biggest accounts that only get called when something breaks, pricing nobody has reviewed since 2023, the thing you have described to clients twice and never shipped.
But someone excellent at careful financial operations is not automatically good at account management. IKEA's own account of this is more specific than the retelling: training across remote design, digital sales, relationship building, complex problem solving. Not "the bot does support now, go sell kitchens." Moving someone into work they have no path into is not redeployment, it is handing them a different problem.
And sometimes the arithmetic does not work. You automate 25 hours and find you have ten hours of genuinely valuable work available. That happens, and pretending otherwise is how these decisions get made badly.
If the honest answer is that there is nothing worth moving her into, then this is a cost reduction. Be clear with yourself that it is. Responsible use of this technology does not mean never reducing headcount. It means understanding what you are choosing and not describing the choice as something the software did.
Redeployment is also a design problem
Redeployment fails when it is announced rather than designed.
If Maria's invoicing gets automated and she is told she now owns customer follow-up, with no change to targets, no definition of what good looks like, and no acknowledgment that she has never done this before, it goes badly. She will do the new work in the gaps around old habits, feel evaluated on something nobody explained, and start looking.
The version that works is slower and duller. Name the new work specifically. Say what success looks like in ninety days. Accept that the first month is worse than the status quo.
There is also a governance point here that is easy to miss. Automating the task does not remove ownership of the task. It changes who owns what happens when the automation is wrong. That ownership has to land on a named person, and if it does not land anywhere the project is unfinished. We worked through what that looks like in What AI Governance Actually Means When You Have Five Employees.
Use Maria's judgment, then stop depending on it
Keep her close to the automation at first, because she is the person most likely to recognize a wrong output as wrong.
That is a transition arrangement, not the permanent answer. Leaving detection with Maria forever just moves the single point of failure. If she leaves, the automation runs unwatched and nobody knows what normal looked like.
The progression that works: her judgment tells you what wrong looks like. Some of those observations become written rules and thresholds that run whether she is there or not. Others do not reduce to a rule, and stay a judgment call for whoever inherits her role. A transition period with her watching the outputs is how you find out which is which. After that, the monitoring should stand on its own.
What to actually ask
Before automating anything a specific person currently does, four questions.
What does this person know that is not written down. Write some of it down before the process changes, while they still remember it.
What valuable work has been sitting undone, and can this person plausibly become good at it. If the answer to either half is no, you are making a cost reduction.
Who watches the automation. That establishes ownership.
What happens when it is wrong. That establishes the control, and it is a different question. For invoicing it means wrong amounts, duplicate invoices, a good customer marked delinquent, collection activity that should never have gone out. Somebody has to notice, and something has to happen next.
IKEA had a training function and new roles to point people toward. You have a list of things you have been meaning to get to. The mechanics differ. The choice is the same one, and at your size you are close enough to it that you cannot pretend it was made by the technology.
Where Agent Micho Fits
We build automations for small businesses, and the conversation about what happens to the freed-up hours is part of the scoping rather than an afterthought. If you are looking at automating something a specific person does today, that is the conversation worth having first.
Sources
- Billie's share of inbound inquiries, the interaction volume, the cost savings, and the reskilling of around 8,500 remote customer support workers into remote design, digital sales, relationship building and complex problem solving: Ingka Group newsroom and the Ingka Group FY23 annual summary.
- Sales through Ingka's remote customer meeting points, EUR 1.3 billion in FY22, 3.3% of total sales: reported by Ingka and covered by Reuters' wire report on the figure (via RTÉ). It covers the whole channel, not revenue attributable to the reskilled workers or to Billie specifically.
- The 8,500 figure refers to workers Ingka says it reskilled as customer support work changed. Neither source establishes that Billie displaced 8,500 jobs, which is the claim the popular retelling makes.
- The Maria example is illustrative rather than a specific client, and the tacit-knowledge points are drawn from our own client work.
- Part one of the governance series covers the ownership and incident-response decisions referenced here: What AI Governance Actually Means When You Have Five Employees.
A note on the images in this piece: the hero and the listing illustration are both AI-generated.