Who Buys the Products?
The Lessons from Klarna’s AI Reckoning
In February 2024, Klarna CEO Sebastian Siemiatkowski told the world his company’s new AI assistant was doing the work of roughly 700 full-time customer service agents. Resolution times dropped. Customer satisfaction scores held steady. The company said headcount would shrink through natural attrition, from roughly 3,800 employees toward 2,000. Wall Street loved it. Klarna became the poster child for AI-driven efficiency; proof that a “leaner, smarter” company was not just possible but imminent for every knowledge-work employer watching.
Fifteen months later, Klarna walked part of it back. Quality had slipped. Customers wanted a human option, and Klarna began rehiring.
The reversal made headlines as a story about chatbot quality. That’s the small story. The larger story is what the original decision assumed it didn’t need to model. And it’s the same blind spot sitting inside boardrooms making AI staffing decisions right now.
The Layoffs That Don’t Look Like Layoffs
For most of the last three decades, “automation risk” targeted a specific kind of job: repetitive, physical, low-wage. Assembly lines. Call centers. Data entry. The mental model executives carried into every restructuring conversation was that technology displaces the bottom of the org chart.
That model is now wrong, and the people it’s wrong about are not marginal.
They are product managers, Salesforce administrators, senior designers, solutions architects, marketing directors, lawyers, and bankers. People who, eighteen months ago, would have described their jobs as secure; people with mortgages, aging parents, and university-bound kids. In 2026, the distance between “successful professional” and “unexpectedly unemployed” has narrowed further than most leaders are prepared to admit.
This matters because these are not the workers policymakers have spent decades worrying about. They are the ones the system quietly assumed would always be fine; university degrees, white-collar, home-owning, tax-paying. The group least prepared for disruption is now the group most exposed to it, and almost nobody built a contingency plan around that fact.
The Multiplier Nobody Models
Every AI staffing decision I’ve seen modeled treats the layoff as a line-item cost reduction: salary, benefits, severance, done. That math is real, but it’s also incomplete in a specific, predictable way; the same way it was incomplete in every industrial town that lost its anchor employer decades ago.
The people being laid off are also customers, taxpayers, and borrowers. Pull on any one of those threads and the second-order effects show up fast:
Consumer demand contraction. Many of the companies deploying AI to cut headcount are selling directly to the demographic they’re displacing — software subscriptions, financial products, premium retail, travel. The laid-off employee was often also the customer.
Credit tightening. Banks read rising layoffs in a sector as rising default risk. Lending standards tighten — for mortgages, small business loans, auto financing — well before default rates actually move, which slows spending further.
Local service contraction. Massage therapists, independent cafés, boutique gyms, and neighborhood restaurants don’t survive on savings; they survive on discretionary spending from people with steady paychecks. When savings get anxiously jealously preserved instead of used for consumption, these businesses are usually the first to feel it and the last anyone notices.
The Tax Base Nobody’s Budgeted For
There’s a fourth thread, and it’s the one with the worst timing: national tax revenue. Mid-career professional salaries — the product managers, architects, senior analysts — carry disproportionate income tax weight relative to entry-level roles. They’re deep enough into the income curve to be paying real marginal rates, but numerous enough as a group that a wave of mid-tier layoffs shows up in aggregate tax revenues, not just individual hardship statistics.
This is landing at the worst possible moment. Most G7 governments are running large structural deficits, many with multi-year spending commitments already locked in: infrastructure, defense, healthcare, debt servicing on debt itself. Those budgets were built on tax base projections that assumed the professional class kept working, kept earning, kept filing at the same bracket. A government that loses a meaningful slice of its mid-tier income tax base while its spending obligations stay fixed doesn’t get to reduce its deficit; it needs to fund the gap either by more borrowing or by tax increases on a shrinking base of people still earning enough to absorb them. Either path feeds back into the same consumer demand problem the layoffs started.
No CFO models this. It’s not supposed to be their job. But it’s happening because of decisions being made in rooms exactly like theirs.
The Blind Spot in the Boardroom
It is worth reiterating the uncomfortable part. The company doing the cutting is often the same company whose growth depends on the spending power of the people it just cut. Not directly in many cases — but through the same regional economy, the same consumer base, the same credit markets, the same tax-funded infrastructure their business relies on.
Goldman Sachs’s 2023 estimate that generative AI could affect the equivalent of 300 million full-time jobs globally wasn’t a headline about factory floors. It was a headline about the exact roles Klarna targeted first: customer service, administration, mid-level analysis. The IMF’s 2024 assessment that roughly 60% of jobs in advanced economies are exposed to AI, with close to half of those facing negative displacement effects rather than augmentation, tells the same story from a different angle.
Business leaders modeling these decisions are running a narrow P&L — cost per seat, resolution time, headcount ratio — when the more valuable exercise requires a wider one.
What happens to our own revenue line if this becomes an industry pattern rather than a single company’s edge?
At what point does “efficiency” become a coordination failure, where every company optimizing individually degrades the demand environment collectively?
E.D.G.E.: Managing Your Own Second- and Third-Order Exposure
Establish
What can you actually control? Your own staffing decisions, your own customer diversification, your own balance sheet resilience. What you can’t control: aggregate layoffs across your sector, competitor decisions, or the speed at which credit tightens in response.
Key question: Have you separated the AI staffing decision itself from the demand environment it might help create?
Diagnose
This is where most leaders stop too early — at the cost side of their own AI decision, never turning the lens back onto their revenue side.
Key question to force the diagnosis: What percentage of our revenue comes from customers whose income is directly exposed to AI-driven displacement — in our industry or adjacent ones? A B2B software company selling seat licenses to mid-sized firms, a regional bank with a mortgage book concentrated in professional-class borrowers, a retailer whose margin depends on discretionary spending — each has a different exposure number, and almost none of them have calculated it.
Go
Purposeful action here isn’t “wait and see”. It’s building the contingency plan before demand actually softens, so the response isn’t improvised under pressure:
Stress-test revenue against a 10% and 20% contraction in your most AI-exposed customer segment, the same way you’d stress-test for a rate shock.
Identify which product lines or customer segments are least correlated with mid-tier professional income, and know in advance how quickly you could shift resourcing toward them.
Review lending, credit, and payment-term policies now, not after delinquency data forces your hand; reactive tightening after the fact tends to overshoot and choke off customers who would have recovered.
Evolve
Build the feedback loop before you need it. Track leading indicators of demand deterioration in your exposed segments — subscription downgrades, order size trends, payment-term requests — rather than waiting for the quarterly revenue number to confirm what’s already happened.
The goal isn’t to predict a recession. It’s to know, faster than your competitors, when your own customer base is starting to feel the effects of decisions being made across the wider economy — including, possibly, your own.
A Personal Note: The Analyst Who Wasn’t There
Thirty years ago, I was the junior investment bank analyst doing grunt work at ridiculous hours. I built valuation models late into the night. I reworked PowerPoint decks for the third time because additional feedback from senior bankers needed to be incorporated. None of it felt like important work in the moment.
But looking back, it was the training ground. Building those models taught me how deals actually got structured. Sitting in on the negotiations I was only there to take notes on taught me how senior bankers read a room, when to push and when to concede. A few years later, I was the one leading those deals; because I’d spent years doing the unglamorous work that builds the judgment senior work requires.
Investment banks are now talking seriously about having AI replace the junior analyst function: the modeling, the deck-building, the number-crunching. The cost logic is sound in isolation. But senior bankers don’t arrive fully formed. They’re built, over years, by doing exactly the work that’s being automated away. If a bank can no longer offer that apprenticeship, it faces a third-order effect nobody’s pricing into today’s efficiency gains. Where exactly does its next generation of rainmakers come from? The answer is the market; through bidding wars for the scarce senior talent every firm is now competing for, having each cut its own pipeline. Today’s junior-analyst savings become tomorrow’s senior-banker premium, and the arithmetic that looked so good in this year’s budget quietly reverses in five.
Your Turn: Model the Second-Order P&L
List every role category under consideration for AI substitution — including entry-level roles you think of as “training”, not just cost centers.
Estimate what fraction of your own revenue depends on customers whose income sits in the bracket most exposed to AI displacement.
Ask finance to model a 90-day and 12-month scenario where 20% of that segment reduces discretionary spending on your products specifically.
Ask HR to model where your senior talent comes from in five years if the entry-level rung disappears today.
Decide with all four numbers on the table — not just the one that made this quarter’s business case easy.
What will you do the quarter your demand actually drops — and will you have built that plan before you need it, or after?
© The Uncertainty E.D.G.E. | Published every other Tuesday
If this resonated, I’d love to have you as a free subscriber — and forward this to a leader who needs to see around corners.
The Uncertainty E.D.G.E. is for leaders who are accountable for outcomes they can’t fully control — and want a clearer way to think when certainty won’t come.
Essays and conversations on decision-making under pressure, every other Tuesday. Join me at theuncertaintyedge.com.
If the human side of leadership is what draws you, I also write The Good Human Practice — on inner clarity and character for leaders called on to make those hard decisions under pressure.