Klarna and Commonwealth Bank Rehired the People. Both Said the Same Thing About Cost.

Two big companies cut customer-service people for AI. Two big companies then brought people back. Neither story is really about a chatbot being clever or stupid. Both are about the same owner question: is AI worth it once you count the whole job, not just the cheap part?

Klarna became the poster child for replacing support staff with AI. Commonwealth Bank of Australia — CBA, the country's biggest bank — did a smaller version with a voice bot. Together they give a cleaner lesson than any AI failure rate headline: they measured the saving before they had proved the work was covered.

Klarna: "lower quality" was the bill that arrived later

In 2024 Klarna said its chatbot was doing the work of 700 customer-service agents, cutting average resolution time from 11 minutes to two. It also said staff had fallen from about 5,000 to about 3,800. Those are the company's own numbers, reported by Reuters — the kind that make cost per contact look like the whole story.

Then, in May 2025, chief executive Sebastian Siemiatkowski told Bloomberg the company was hiring people again so customers would always have a real-person option. "As cost unfortunately seems to have been a too predominant evaluation factor when organizing this, what you end up having is lower quality," he said. "Really investing in the quality of the human support is the way of the future for us."

By September 2025 he was plainer: "We probably over indexed a little bit on that, and then in the last six months we have been trying to course correct," he told Reuters. Note what Klarna did not do: it did not abandon AI. A spokesperson told Fortune it was "very much still AI-first," with the focus moving from cutting costs toward improving services. The reversal was about order — people were removed before quality had been proved.

Not sure whether AI would pay off in your firm — or land you in the same spot? Talk it through before you cut anything. Talk to a free agent — tell it the one task you'd hand over, and you'll get a straight answer on where AI fits, where it shouldn't touch, and the next step. Free, no signup, and no client files needed.

Commonwealth Bank: 45 roles, one voice bot, and an apology

CBA's version was smaller and faster. In July 2025 it said 45 Customer Service Direct roles would go after it introduced an AI "voice bot." The Finance Sector Union disputed the picture: members said call volumes were rising, with staff offered overtime and team leaders pulled onto the phones. It took the dispute to the Fair Work Commission.

On 21 August 2025 CBA reversed the decision. "CBA's initial assessment that the 45 roles in our Customer Service Direct business were not required did not adequately consider all relevant business considerations and this error meant the roles were not redundant," a spokesperson said. "We have apologised to the employees concerned." Staff could stay, seek redeployment, or leave with a payment.

That word — assessment — is the whole article. CBA did not say the voice bot failed a technical test. It said its assessment was not thorough. The redundancy decision ran ahead of the evidence about demand.

What "was it worth it?" actually has to count

This is why "klarna ai rehiring" belongs with the MIT GenAI Divide report and the wider AI failure stats. MIT's point was not that AI never pays — it was that most organisations saw no measurable profit impact in a short window, especially from tools that didn't fit the workflow. Klarna and CBA show the owner-level version: a tool can handle volume and still not cover the job.

Before AI replaces anyone's hours in a small firm, count four things, in this order:

• Demand, not just handling time. CBA assumed calls would fall. Measure enquiries before and after, including people who give up, call back, or ask for a human. If demand rises, your "saving" is a queue.
• Quality at the hard end. Klarna's bot resolved simple queries fast. The cost showed up when a customer was anxious, angry or confused. Sample the hardest cases separately — that is where trust lives.
• The human option. Both firms ended up restoring one. Build it in from the start: a clear, quick route to a person for high-stakes, sensitive or stuck cases. Ford learned a version of the same lesson with experienced engineers — see Ford rehiring the engineers AI couldn't replace.
• A kill-or-keep date. Run any replacement idea as a time-boxed AI proof of concept: baseline first, one task, a named owner, and a date to keep, shrink or stop it. Our AI use-case scorer and AI ROI page are built for that order.

Notice what is not on that list: headcount. Cutting a role is the last step, after demand, quality and the fallback have held up for a full cycle — not the launch announcement. Still deciding where you stand? The AI readiness assessment is a calmer start than a vendor demo.

Three questions to ask before you copy either firm

You don't need a bank's budget to make — or avoid — a bank-sized mistake. If you're weighing AI for your phones, inbox or chat, take these three questions to the free agent on this site. A real person reads every conversation:

1. "I want AI to handle [this enquiry/task]. Which parts should it never touch alone, and where does a human need to stay in the loop?"
2. "This task takes my team about [X hours a week]. Would AI actually pay off here, or is it cheaper to fix the process first?"
3. "How would I run a 90-day proof of concept for this — what do I measure in week one so I can prove whether it worked?"

Ask the free agent your version of those three → Tell it about one tool you're considering and the piece of work you'd want it to take over. You'll get a straight answer on where AI fits, where it shouldn't touch, and the next step. Free, no client files.

Klarna kept the AI and rehired the people. CBA kept the voice bot and un-cut the roles. That is not AI failing. It's two firms that counted the saving first and the work second — and then said so out loud. A small firm can run that order in reverse.

FAQ

Why is Klarna hiring people again after replacing staff with AI?

Chief executive Sebastian Siemiatkowski told Bloomberg in May 2025 that cost had been "a too predominant evaluation factor" and the result was "lower quality." Klarna began recruiting remote, on-demand human agents so customers always have a human option, while saying it remains "very much still AI-first."

What did Commonwealth Bank get wrong with its AI voice bot?

In July 2025 CBA said 45 Customer Service Direct roles were redundant after introducing an AI voice bot. In August it reversed the decision, apologised, and said its initial assessment "did not adequately consider all relevant business considerations" and the roles were not redundant. The union said call volumes had risen, not fallen.

Does this mean AI customer service isn't worth it?

No. Both firms kept the AI. The lesson is sequencing: prove demand, quality and a human fallback on one task before removing anyone's hours. Count quality and repeat contacts, not just cost per contact or handling time.

How should a small firm test AI before cutting hours?

Run a time-boxed proof of concept: take a baseline, pick one task, name an owner, keep a quick route to a human, and set a keep-or-kill date in advance. Then judge it on demand, hard-case quality and total cost — not on the vendor's demo.

Sources

Bloomberg (8 May 2025); Fortune (9 May 2025); Reuters (10 Sep 2025); ABC News and Finance Sector Union (21 Aug 2025), as linked in the text above.