In February 2024, Klarna's support team watched its average resolution time fall from 11 minutes to under two, after switching on an OpenAI-built assistant to handle its live chats.
That drop is the number every fintech AI case study since has been measured against. The story didn't stop there, though.
Fifteen months later, Klarna was rehiring some of the human agents it had leaned away from, and its own leadership said so on the record.
Most fintechs planning something similar treat it as one piece of a broader digital transformation roadmap, not a chatbot bolted onto a support inbox.
What This Fintech AI Case Study Actually Showed
Klarna's own announcement of the launch, made in February 2024, put a real number behind the headline: the assistant handled 2.3 million conversations in its first month.
That's work Klarna said would otherwise have needed roughly 700 full-time agents. It ran across dozens of markets, in dozens of languages.
Klarna projected the assistant would add $40 million to its 2024 profit.
The resolution-time drop made the headlines, but the repeat-contact rate mattered more day to day: 25% fewer customers had to come back a second time about the same issue.
| Metric | Before the assistant | After the assistant |
|---|---|---|
| Average resolution time | 11 minutes | Under 2 minutes |
| Repeat contacts on the same issue | Baseline | Down 25% |
| Share of chats automated | 0% | Two-thirds within 30 days |
The Reversal Nobody Highlights
By May 2025, the picture had shifted. Complex disputes, fraud claims, and hardship cases were where the AI's resolution quality slipped, and in fintech, a confidently wrong answer about fees or payment terms creates real compliance exposure.
CEO Sebastian Siemiatkowski told Bloomberg the company had cut human support too aggressively. Customer Experience Dive reported his line that "really investing in the quality of human support is the way of the future for us."
Klarna began rehiring for premium and complex-case roles while keeping the assistant as the front line for high-volume, simple queries. A company spokesperson put the new split plainly: AI handles speed, and people handle the moments that need judgment.
A Bigger, Slower-Burn Example: Bank of America's Erica
Photo by Calvin Seng on Pexels
Erica has been running for years, long enough to be judged on more than a launch-month press release.
Bank of America's own August 2025 update put total interactions past three billion since Erica launched back in 2018.
Adoption has broadly kept pace, with tens of millions of people now using it every month.
More than 98% of users say they get the information they came for, and a companion version built for employees cut calls to Bank of America's internal IT service desk in half.
That employee tool was built the way a company would build any other internal app or SaaS tool, not bolted on as an afterthought.
Where AI Support Still Needs a Human Backstop
Line up Klarna and Erica, and the real difference sits in scope and time, not the underlying technology. Erica took years to earn its resolution rate. Klarna tried to move an entire support queue in one launch and had to walk part of it back within a year.
Scope an AI and automation project for support the way Erica's team seems to have. Start with cases that have a clear, checkable answer: balances, order status, return eligibility.
Disputes, exceptions, and anything touching compliance are exactly where Klarna's numbers slipped. Those are worth routing to a person on purpose, rather than by accident.
The build-versus-buy call matters too. A narrow, well-scoped assistant is a different project from the no-code versus custom development decision fintechs face for their broader platform, and it's worth making that call on purpose rather than defaulting to whichever chat widget was easiest to install.
Cover photo by Tima Miroshnichenko on Pexels
Sources
- Klarna AI assistant handles two-thirds of customer service chats in its first month — Klarna / PR Newswire
- Klarna changes its AI tune and again recruits humans for customer service — Customer Experience Dive
