A historical announcement, not an independent audit
This is an independent editorial analysis of a third-party deployment, not independent verification of its outcomes. It does not describe XIVTech work or establish the system's current state.
In its 27 February 2024 announcement about its AI assistant, Klarna reported 2.3 million conversations and two-thirds of customer-service chats during the assistant's first month. Those are historical claims attributed to Klarna.
OpenAI's Klarna deployment page, undated and accessed on 30 October 2026, is another deployment-party account. Two parties describing the same deployment do not constitute independent outcome verification.
Throughput does not establish answer quality
Klarna's announcement reported customer satisfaction on a par with human agents. That claim concerns a different dimension from conversation volume: handling a large share of chats does not itself establish correct answers or successful resolutions.
Interpreting the satisfaction comparison requires response rates, survey timing and the populations being compared. Customers who complete a survey may differ from those who abandon a session or seek help elsewhere.
The cited accounts do not provide per-language quality estimates or a sampling design sufficient to establish that an aggregate result applies evenly across languages and task types.
Repeat inquiries need a defined denominator
Klarna also reported 25% fewer repeat inquiries in its February 2024 announcement. To interpret that figure, a buyer needs the definition of a repeat inquiry, the follow-up window, the comparison period and the denominator.
Does a repeat mean another message, another conversation about the same issue, or any later contact from the same customer? Are contacts through another channel linked? Different definitions can produce different measures.