Klarna
genAI customer agent
In 2024, Klarna's AI assistant built with OpenAI handled 2.3 million conversations in one month (two-thirds of support, the load of 700 agents), before Klarna reopened human roles in 2025.
Key points
- A genAI customer agent that takes the service chat by default.
- Built with OpenAI GPT-4, integrated with the app and account data.
- 2.3 million conversations in the first month, the load of 700 agents.
- Resolution in under 2 minutes, evidence level B, mixed signals.
Objective
Absorb the volume of customer service requests without growing the team, and shorten resolution time on simple cases (refunds, disputes, payments, cancellations) while keeping a human for the rest.
The deployment
The assistant lives in the Klarna app and takes over the customer service chat. One month after its February 2024 launch, it had handled 2.3 million conversations, or two-thirds of support chats. It covers 23 markets, converses in more than 35 languages, and closes a request in less than two minutes where an agent used to take eleven. The customer can ask for a human at any time. In 2025, Klarna acknowledged that it had cut headcount too far and reopened premium support roles staffed by humans, while keeping the assistant in production.
Results Proof B
Figures published by Klarna (official press release, T1) and echoed in OpenAI's customer story (T2), concordant. The 40m USD gain is an internal company estimate, not an audited result, which caps at B. The 2025 reversal is documented by an established press source (TechCrunch).
How it works
Inferred typical approachThe internal detail is not public. Here is a proven approach that leads to the same result, to adapt to your stack.
The stack in detail
- llm OpenAI GPT-4 the assistant's language model, deployed with OpenAI as partner
- infra Integration in-house Klarna real-time connection to the customer's account, orders, and payment history to handle their specific case
- outil Chat in-app et routage vers agents humains the assistant takes the chat by default, discloses that it is an AI, and escalates to a human on request or for complex cases
How it runs, concretely
For ops teams-
1Chat intake AI
The customer opens the chat in the app; the assistant takes the request by default and discloses that it is an AI.
-
2Context retrieval AI
The assistant reads the customer's account, orders, and payment history to answer their specific case (refund, dispute, cancellation).
-
3Resolution or escalation AI / customer service
Simple case: the assistant handles and closes it. Complex case or explicit request: handoff to a human agent with the conversation context.
-
4Quality loop data team
The team tracks satisfaction, repeat contacts, and escalation volume to adjust the scope entrusted to the AI.
The resolution rate without escalation and the repeat contact rate. If the repeat rate rises, it means the assistant is closing poorly resolved tickets: that is the signal that triggered the 2025 rebalancing.
How your customers perceive this type of use
Sourced studiesLes consommateurs n'acceptent pas les chatbots par defaut : 64% prefereraient que les entreprises n'utilisent pas d'IA dans leur service client (Gartner, 2024) et pres d'un utilisateur sur cinq du service client par IA n'en retire aucun benefice (Qualtrics, 2025). L'acceptation se construit sur trois conditions mesurees par Salesforce : savoir qu'on parle a une IA, pouvoir escalader vers un humain, comprendre la logique de l'agent.
Acceptance conditions
- Etre informe qu'on parle a une IA et non a un humain (pres de 75% le demandent, Salesforce 2024)
- Un chemin d'escalade clair vers un agent humain (45% plus enclins a utiliser l'agent IA, Salesforce 2024)
- Une logique de l'agent clairement expliquee (44% plus enclins, Salesforce 2024)
Red lines
- Rendre l'humain injoignable : c'est la premiere inquietude des consommateurs sur l'IA dans le service client (Gartner 2024) et 50% craignent que l'IA les coupe du contact humain (Qualtrics 2025)
- Remplacer le service client par l'IA sans alternative : 53% envisageraient de partir chez un concurrent (Gartner 2024)
Sources: Salesforce 2024 · Gartner 2024 · Qualtrics 2025
How to replicate
Inference, not sourcedData prerequisites
- History of labeled support tickets
- Real-time access to the customer's account and transactions
- Up-to-date knowledge base of procedures and policies
Org prerequisites
- Clear handoff rule to the human
- Measurement loop for satisfaction and repeat contacts
- Compliance framework (AI transparency, sensitive data)
Possible stack
- Generative LLM (OpenAI, Anthropic, or open model)
- RAG layer on the knowledge base
- Integration with the ticketing system and core banking
The plan, step by step
- Step 1Isolate the 5 to 10 most frequent request types and update the knowledge baseDeliverable: Documented case scope + up-to-date procedures
- Step 2Define the handoff-to-human rules and transparency (AI disclosure to the customer)Deliverable: Escalation policy validated by customer service and compliance
- Step 3Connect the LLM to account, order, and payment dataDeliverable: Assistant in beta on one market and one language
- Step 4Pilot on a share of traffic and compare against agentsDeliverable: Resolution rate, repeat contacts, and satisfaction measured
- Step 5Scale up market by market while monitoring the repeat contact rate (the warning signal at Klarna)Deliverable: Default assistant with a quality loop in place
First step: Isolate the 5 to 10 most frequent and best-documented request types, and frame a handoff-to-human rule before opening traffic.
The same pattern, proven in other industries
ADT
genAI customer agent
Air India
genAI customer agent (automated customer service)
AT&T
Governed enterprise genAI assistant: an LLM connected to internal knowledge, open to tens of thousands of employees and called at scale by automated systems
Sources
- S1 Klarna AI assistant handles two-thirds of customer service chats in its first month Primary archive pending
- S2 Klarna's AI assistant does the work of 700 full-time agents Interested party archive pending
- S3 Klarna CEO says company will use humans to offer VIP customer service Established press archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.