Independent
Software Developer
HR Manager
Team Leader
Accountant
Copy Writer
Customer Service
Virtual Assistance
Manila Got the Memo First
When Sebastian Siemiatkowski told the world in early 2024 that Klarna’s AI was doing the work of 700 customer-service agents, Teleperformance’s stock fell 19% to a seven-year low. Eventually it shed 39% from its peak.
Investors read the story they had been trained to read. AI eats offshore. The robots had finally come for Manila.
The trade was wrong.
Six months earlier, Klarna had already moved 750 customer-service roles offshore — to Foundever, the Miami-headquartered BPO, and to Accenture. Ten countries.
The AI was the second leg of a two-leg strategy, not its replacement. Nobody bothered to put that on a chart.
Two legs, one loud story
In August 2023, Klarna transferred 250 customer-service roles from Sweden and Germany to Foundever. In October, it transferred 500 more — customer service to Foundever, financial-crime prevention to Accenture, across ten countries from Australia to the US.
That was the quiet leg.
The loud leg came in early 2024. An OpenAI-powered chatbot that Klarna said was doing the work of 700 full-time agents, handling 2.3 million conversations a month.
The internet ran with it. Teleperformance bled.
What the headlines missed: the offshore base never went anywhere. Foundever was already there. The AI sat on top of it.
The wave nobody priced in
By May 2025, Siemiatkowski admitted the AI-only customer service had produced “lower quality” outcomes. Customer satisfaction had dropped 22%.
Klarna started rehiring humans — directly this time, in an “Uber-style” remote pool, while quietly leaning harder on the existing Foundever contract.
Klarna had company. Sinch research released this month found that 74% of enterprises that deployed live AI customer-communications agents have rolled them back. Among organisations with the most mature governance frameworks, the rollback rate is 81%.
Gartner now forecasts that half of all companies that cut customer-service staff for AI will rehire by 2027 — often under polite new titles like Solution Consultant or Trusted Advisor, to soften the embarrassment.
Air Canada’s chatbot fabricated a refund policy and got the airline held legally liable. Cursor’s chatbot invented company rules. One unnamed firm forced engineers to take phone calls when its AI collapsed.
None of this got the Teleperformance treatment.
The inversion
The CFO case for AI was meant to be cost. That case is now collapsing in slow motion.
Gartner projects that by 2030, the cost per resolution for generative AI customer service will exceed the cost of an offshore human agent.
Token prices are rising. Vendors are pivoting from subsidised growth to profitability. Complex use cases burn more compute.
And the supervisor stack — AI quality reviewers, conversation designers, guardrail engineers — keeps growing. 76% of enterprises now spend more on AI trust and compliance infrastructure than on AI development itself.
To his credit, Siemiatkowski isn’t backing down on the broader thesis. He still maintains that AI halved Klarna’s workforce — 7,000 to 3,000, heading for under 2,000 by 2030 — and that other tech CEOs are “sugarcoating” the AI job impact.
He’s half right.
The headcount math worked. The customer-facing layer didn’t.
That distinction matters. AI did real work in Klarna’s back office.
What it could not do was the gnarly emotional 20% — the angry-call escalation, the empathy ledger.
Offshore didn’t get displaced. It got promoted.
What Manila knew
While Western markets panicked, the Philippine BPO sector quietly logged its best year. IBPAP projects $42 billion in revenue in 2026, 1.97 million jobs, and a 4% employment growth rate — outpacing the global services average. By 2028 the sector expects $59 billion and 2.5 million workers.
Sixty-seven percent of Philippine BPOs have adopted AI tools. As augmentation, not replacement.
Training periods are down 67% — from 90 days to 30. The same AI that Klarna couldn’t make work alone is making Filipino agents better at the work they were already doing.
The trade markets got wrong in 2024 wasn’t a Klarna trade. It was a category trade.
The category was always the hybrid stack: AI for the easy 60%, offshore humans for the judgment 40%.
Klarna ran both legs from day one. It just only told us about one of them.
Teleperformance hasn’t received a public apology. It probably won’t.
The question for your business
Does your AI strategy have a second leg — or just a press release?
