IAPQB

The work of 700 agents

Klarna's claims for its AI assistant, what it said later about quality, and how firms are asked to measure customer service

IAPQB Report No. 25 · Case file · Published 17 November 2026
International AI Practice Qualification Board (IAPQB)

In one minute

  • In February 2024, Klarna said its OpenAI-powered assistant had handled two-thirds of its customer service chats in its first month and was doing work equivalent to 700 full-time agents [1].
  • In May 2025, its chief executive told Bloomberg that cost "seems to have been a too predominant evaluation factor" and the result "lower quality". Klarna resumed recruiting so customers can always reach a person [2, 3, 4].
  • In 2025, the UK Financial Conduct Authority (FCA) reported that some financial firms relied on contact rates and wait times to monitor customer support [5].
  • The FCA asks firms to track results next to speed, as researchers have done; a 2025 Spanish law requires bots to offer a human operator at any point [5, 6, 7, 8].

1. A familiar situation

A customer service director has one month of data from a new AI assistant. The board report shows chats handled, minutes per chat and the agents those chats would have needed. Every line has improved. No line shows customers who came back with the same problem, gave up, or asked for a person and did not get one.

2. Why it happens

An AI assistant produces activity numbers from day one: chats, minutes per chat, share of contacts. Turning them into staff saved needs a calculation, which Klarna did not publish [1]. Whether problems were solved shows up later and is harder to count. A short chat may mean a solved problem or a customer who gave up.

Speed and quality can move apart. In a study of 5,179 support agents, an AI tool that suggested replies raised issues resolved per hour by 14% on average, yet for the most skilled agents resolution rates and customer satisfaction fell slightly [7]. In 2023, the US Consumer Financial Protection Bureau found financial firms using chatbots "as a cost-effective alternative to human customer service" and warned of "doom loops" with no route to a person [9].

IAPQB view. Klarna's published figures measured mostly activity and cost: chats, minutes, staff equivalents, cost per transaction. Its one outcome figure, a 25% fall in repeat inquiries, and its satisfaction claims came without a method [1, 10]. Klarna's own account of lower quality came in an interview, more than a year after launch [2, 3, 4]. The FCA found a similar reliance on contact rates and wait times at some firms it reviewed [5].

3. Three cases

3.1 Klarna, 2024: two-thirds of chats in the first month

On 27 February 2024, the payments company Klarna published figures for its AI assistant one month after it went live globally. It had held 2.3 million conversations, two-thirds of Klarna's customer service chats, and was "doing the equivalent work of 700 full-time agents". Customers resolved their issues in under 2 minutes (11 before), repeat inquiries had dropped by 25%, and customer satisfaction was on par with human agents. Customers could still choose a live agent. The release did not explain how these figures were worked out [1]. The next day, shares in the French call-centre company Teleperformance fell sharply as the statement revived fears that AI would hurt its business, Bloomberg reported [11].

3.2 Klarna, 2025: "lower quality" and a promise of a person

In an interview published by Bloomberg on 8 May 2025, chief executive Sebastian Siemiatkowski said Klarna's push to cut customer service costs with AI had gone too far [2]. Cost "seems to have been a too predominant evaluation factor", he said, and the result was "lower quality". It was critical, he said, to be clear to customers that "there will be always a human if you want" [3, 4]. Klarna resumed recruiting, starting with a pilot of flexible, remote customer service work [3, 4].

Eleven days later, Klarna's first-quarter results said customer service cost per transaction had fallen by 40% since early 2023, "with no drop in consumer satisfaction following the rollout of our AI assistant". The cost measure divides customer service and operations costs by completed transactions. No satisfaction figure was given [10].

3.3 United Kingdom, 2025: firms counting contacts and wait times

The FCA's Consumer Duty requires financial firms to monitor whether retail customers get the support they need. It surveyed 407 firms in May 2024 (356 replied) and examined 40 in more depth that September. Its findings (March 2025, updated July 2026) said some firms "relied on transactional metrics, such as contact rates and wait times", or depended only on feedback and complaints to spot problems. Of the firms surveyed, 13% said they did no quality assurance on their support channels. Where support was outsourced, progress on oversight was mixed; firms need data to monitor outcomes "beyond just measuring operational efficiency" [5].

4. How practitioners handle it now

Speed is reported next to measures of the result. The FCA wants firms to consider "a range of metrics", qualitative and quantitative [5]. Its guidance lists data such as first contact resolution rates, time to resolution, call abandonment rates, satisfaction surveys and net promoter scores [6, para 9.53]. The support study tracked resolution rates, customer surveys, the tone of customers' messages and requests to speak to a manager, which fell by almost 25% with AI help [7].

Satisfaction is measured after the problem is closed, and the numbers are audited. Spain's customer service law, in force since 28 December 2025, requires large companies and providers of basic services to measure customer satisfaction and the causes of dissatisfaction, and bans satisfaction surveys before a query or complaint is resolved [8, art. 18]. Companies must evaluate quality each year, publish the evaluation and its audit, and have an accredited auditor confirm that each measure is accurate to within 5% [8, arts. 21 and 22]. These articles do not apply to the financial sector, which has its own rules [8, art. 2(5)].

A person can be reached from inside the bot. The same law bans automated answering as the only channel. Bots must let customers ask for a human operator at any point, and 95% of such requests must be answered within three minutes on average [8, art. 8]. Companies have twelve months from 28 December 2025 to adapt [8]. From 19 June 2026, an EU directive gives consumers buying financial services at a distance through online tools a right "to request and to obtain human intervention" before the contract, and in justified cases after it [12].

Hard cases go to a person by design. The FCA's guidance says firms that support customers mainly through one channel, such as digital-only, are likely to need "a real-time human interface, such as a phone service" for some non-standard issues, for example fraud [6, para 9.18]. In the FCA review, some firms automatically sent chatbot questions about bereavement to a support representative [5].

5. What to take from this

For customer service staff. Agents see what activity numbers miss: customers who come back with the same problem or ask for a person. The FCA counts feedback from customer-facing staff among the ways firms monitor outcomes [5], and the support study counted requests for a manager [7].

For managers who set targets or buy AI tools. At Klarna, the published figures measured mostly activity and cost [1, 10], and its own account of lower quality came in an interview [3, 4]. The FCA and Spain's law name measures worth agreeing with a vendor before launch: problems resolved, repeat contacts, satisfaction after resolution and time to reach a person [5, 6, 8].

For people who read or report claims about AI. Klarna's 700-agent statement was followed the next day by a sharp fall in a call-centre company's shares; the release did not show the calculation [1, 11]. Three questions follow: how a figure was calculated, which customer outcome it covers, and whether customers could still reach a person.

6. Limits of this report

Only the opening paragraphs of both Bloomberg articles could be read; the interview quotes come from Fortune and CX Dive [2, 3, 4, 11]. Reuters and Financial Times reports were not read. Klarna's figures are its own; the sources do not say how it measured satisfaction or quality, or how its service runs in late 2026. The FCA review does not single out AI [5]. The FCA's Consumer Duty page, updated 23 October 2026, describes the Duty as current [13]. Spain's consolidated law showed no amendments on 16 November 2026, and a constitutional challenge concerns provisions not used here [8]. National laws applying the EU directive were not checked [12]. The study's AI tool advised human agents, while Klarna's assistant answered customers; the figures are from the 2023 working paper [7]. No AI assistant was tested.

References

[1] Klarna (27 February 2024). Klarna AI assistant handles two-thirds of customer service chats in its first month. Klarna press release. https://www.klarna.com/international/press/klarna-ai-assistant-handles-two-thirds-of-customer-service-chats-in-its-first-month/

[2] Bloomberg (8 May 2025). Klarna Slows AI-Driven Job Cuts With Call for Real People. https://www.bloomberg.com/news/articles/2025-05-08/klarna-turns-from-ai-to-real-person-customer-service (only the first two paragraphs could be read; quotes from the interview are taken from [3] and [4])

[3] Ivanova, I. (9 May 2025). As Klarna flips from AI-first to hiring people again, a new landmark survey reveals most AI projects fail to deliver. Fortune. https://fortune.com/2025/05/09/klarna-ai-humans-return-on-investment/

[4] Doerer, K. (9 May 2025). Klarna changes its AI tune and again recruits humans for customer service. CX Dive (Informa TechTarget). https://www.customerexperiencedive.com/news/klarna-reinvests-human-talent-customer-service-AI-chatbot/747586/

[5] Financial Conduct Authority (7 March 2025; last updated 10 July 2026). Consumer support outcome: good practices and areas for improvement. https://www.fca.org.uk/publications/good-and-poor-practice/consumer-support-outcome-good-practices-areas-improvement

[6] Financial Conduct Authority (July 2022). FG22/5 Final non-Handbook Guidance for firms on the Consumer Duty [paragraphs 9.18 and 9.53]. https://www.fca.org.uk/publication/finalised-guidance/fg22-5.pdf

[7] Brynjolfsson, E., Li, D., & Raymond, L. R. (2023). Generative AI at Work. NBER Working Paper No. 31161 (April 2023, revised November 2023). https://www.nber.org/papers/w31161. Published as: The Quarterly Journal of Economics, 140(2), 889-942 (2025). https://doi.org/10.1093/qje/qjae044

[8] Jefatura del Estado, Spain (26 December 2025). Ley 10/2025, de 26 de diciembre, por la que se regulan los servicios de atención a la clientela [Law 10/2025 regulating customer service; articles 2, 8, 18, 21 and 22, single transitional provision and eighth final provision]. Boletín Oficial del Estado no. 312, 27 December 2025, BOE-A-2025-26698. https://www.boe.es/eli/es/l/2025/12/26/10 (consolidated text checked at https://www.boe.es/eli/es/l/2025/12/26/10/con on 16 November 2026)

[9] Consumer Financial Protection Bureau (6 June 2023). Chatbots in consumer finance. https://www.consumerfinance.gov/data-research/research-reports/chatbots-in-consumer-finance/chatbots-in-consumer-finance/

[10] Klarna (19 May 2025). First Quarter 2025 Results [Q1'25 earnings release]. https://s205.q4cdn.com/644747736/files/doc_financials/2025/q1/Q1-25-Klarna-Earnings-Release.pdf

[11] Bloomberg (28 February 2024). Call-Center Firm Sinks on Klarna Claims AI Is Doing Agents’ Jobs. https://www.bloomberg.com/news/articles/2024-02-28/teleperformance-sinks-as-klarna-fuels-worries-over-impact-of-ai (only the first two paragraphs could be read)

[12] European Parliament and Council of the European Union (22 November 2023). Directive (EU) 2023/2673 of the European Parliament and of the Council of 22 November 2023 amending Directive 2011/83/EU as regards financial services contracts concluded at a distance and repealing Directive 2002/65/EC [Article 1, inserting Article 16d(3) into Directive 2011/83/EU; Article 2(1)]. EUR-Lex. https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:32023L2673

[13] Financial Conduct Authority (first published 24 February 2026; last updated 23 October 2026). About the Consumer Duty. https://www.fca.org.uk/firms/consumer-duty/about

How to cite

IAPQB (2026). The work of 700 agents: Klarna's claims for its AI assistant, what it said later about quality, and how firms are asked to measure customer service. IAPQB Report No. 25. International AI Practice Qualification Board. https://iapqb.org/research/025-klarna-ai-assistant.html

About this report

Prepared by IAPQB. Each fact comes from the sources listed, and each source was opened and checked on 16 November 2026; anything that could not be checked was left out. AI tools were used to search and to draft, and IAPQB checked every claim against its source and is responsible for the text. IAPQB runs an AI work qualification; this report does not promote it. Corrections: contact@iapqb.org or an issue at github.com/iapqb/iapqb.github.io. Licence: CC BY 4.0.

Archived with the series on Zenodo: doi.org/10.5281/zenodo.24248199 · Markdown sources: github.com/iapqb/research