Klarna Replaced 700 Agents With AI — Could a Singapore Bank?

Klarna gave the world its cleanest soundbite about AI and work. In early 2024 the Swedish fintech announced that its OpenAI-powered assistant was handling roughly two-thirds of all customer-service chats — work the company equated to about 700 full-time agents — and projected a profit impact close to US$40 million. The message landed like a thunderclap in every boardroom: the machines can do the work of hundreds of people, and they can do it now.

Then came the part most headlines skipped. By 2025, Klarna's own leadership admitted it had pushed too far — that cutting humans out of support had cost something the dashboard didn't measure — and began rehiring people for the cases that needed them. The story everyone remembered was "AI replaced 700 agents." The story that actually matters is what happened next.

For Singapore's banks, insurers and service businesses — where trust is the product and the regulator is watching — that second half is the whole lesson.

A modern customer-experience operations floor where human agents and AI assistants work side by side

What Klarna actually proved

Strip away the noise and Klarna demonstrated three things, all true at once.

First: the routine tier is genuinely automatable. A huge share of customer contacts are repetitive — "where's my refund," "change my address," "reset my password." A capable AI assistant resolves these instantly, in dozens of languages, around the clock. That part of the claim was real, and it isn't going away.

Second: the productivity is enormous. Resolving two-thirds of chats without a human touching them is not a rounding error; it is operating leverage of a kind service businesses have never had. The reclaimed hours are real money.

Third — and this is the part Klarna learned the hard way: the last third is where the value lives. The complex disputes, the distressed customer, the edge case that becomes a complaint, the conversation that decides whether someone stays a customer for ten years — those are not "tier-one tickets." Automate them badly and you don't save money; you quietly destroy it, one churned relationship at a time.

Klarna didn't discover that AI replaces people. It discovered that AI replaces tasks — and that mistaking the two is expensive.

The redesign, not the replacement

The companies winning with AI in customer service aren't the ones that asked "how many agents can we cut?" They're the ones that asked a better question: "if the machine clears the routine 70%, what could our people finally do with the other 30%?"

That reframing is the entire game, and it produces a concrete operating model. Take the customer-service function and decompose it into tasks rather than job titles. Then sort those tasks into three buckets.

A three-bucket framework diagram: tasks machines do better, tasks humans do better, and tasks they do better together

• What machines do better — instant retrieval, status lookups, password resets, FAQ, first-draft responses, routing, summarising a case history. Route these to the AI.
• What humans do better — de-escalating an angry customer, exercising judgment on a goodwill exception, handling a vulnerable client, owning a complaint end-to-end, spotting the fraud pattern that "looks fine" to a model.
• What they do better together — the agent who now handles three times the complex caseload because the AI prepped the context, drafted the options, and handled everything routine before the human ever picked up.

Done properly, you don't end up with a smaller team doing the same job. You end up with the same team doing a higher-value job — the agent becomes a problem-solver and relationship-owner, not a script-reader. That is redesign. Klarna's correction in 2025 was simply the market teaching it the difference.

Could a Singapore bank do the same?

Technically, a great deal of it — and some already are. DBS has spoken publicly about running well over a thousand AI use cases and models across the bank; OCBC and UOB have rolled assistants to staff and customers. The raw capability is here. So the interesting question isn't can a Singapore bank automate tier-one service. It's should it do so the Klarna way — and the answer is no.

Three things make Singapore different.

The regulator is in the room. The Monetary Authority of Singapore's expectations around fairness, accountability and transparency mean a bank cannot simply let a model make consequential decisions unsupervised. A human-in-the-loop isn't a nice-to-have; it's the design constraint.

Trust is the moat. Singaporeans will happily let a bot reset a password — and will switch banks over one badly handled dispute. In a small, high-trust, reputation-dense market, the "last third" of service is disproportionately valuable. Automating it carelessly is a strategic error, not an efficiency win.

The labour model is tripartite. Singapore's whole approach — through institutions like Workforce Singapore, SkillsFuture and e2i — is built around redesigning workers into new roles, not discarding them. A bank that frames AI as "redesign and reskill" rather than "cut" moves with the national grain, qualifies for support, and keeps the goodwill of its people and the public.

A Singapore bank's customer-experience team in a Marina Bay office, blending human advisors with AI tools

The playbook a Singapore service business can actually run

If you lead a bank, insurer, telco or any service-heavy business here, the Klarna saga compresses into five moves.

1. Map tasks, not roles. Pull a month of contacts and tag them: routine, complex, emotional, regulated. You'll usually find 50–70% is genuinely routine. That's your automation surface.
2. Automate the routine — visibly to staff, invisibly to customers. Deploy the assistant on tier-one, but tell your team plainly: this clears your queue so you can own the hard cases. Adoption dies the moment staff believe AI is there to fire them.
3. Redesign the human role upward. Rewrite the job around judgment, complex resolution and relationship ownership. Pay and title should reflect that the role got harder and more valuable, not smaller.
4. Keep the human firmly in the loop where it counts. Disputes, vulnerable customers, anything with regulatory or reputational weight — AI assists, human decides. This is both good governance and good business.
5. Reskill, don't release. Move displaced capacity into the redesigned roles, supported by Singapore's reskilling infrastructure. The reclaimed hours become growth, retention and service quality — not just a one-time cost cut.

This is exactly the kind of redesign we help Singapore businesses run end to end — from finding the real automation surface and building the AI safely (Freemansland), to getting the governance, risk and compliance posture right so a regulator never has a question you can't answer (FMC Collective).

The number that should actually move

Here's the part for anyone allocating capital. The naïve reading of Klarna is "AI cut 700 jobs, banked the savings." The operator's reading is different: AI created enormous operating leverage, and the value showed up — or leaked away — depending entirely on whether the organisation was redesigned to capture it.

A Singapore bank that merely buys AI for its contact centre will show a smaller support team, a bigger software bill, and — if it copied Klarna's first move without its second — a quiet drift in customer satisfaction. A bank that redesigns around AI will show rising service quality, flat-to-lower cost-to-serve, and revenue per employee climbing as the same people handle higher-value work. Same technology. Completely different result on the income statement.

Klarna replaced 700 agents with AI, and then re-learned what every serious service business eventually learns: you don't win by removing the humans. You win by redesigning the work so the humans — and the machines — each do what they're best at.

The Singapore bank that internalises that won't make headlines for cutting 700 jobs. It'll make money for keeping the right ones.