How Klarna's AI Assistant Did the Work of 700 Agents in 30 Days — and the Part Everyone Skips

Teardown — a real, public example, analyzed. We didn't build this; we're breaking down what happened and what a normal business should take from it. Every number below is cited.

When Klarna flipped on its AI assistant on February 27, 2024, the results it reported in the first month were the kind of thing that makes a CFO sit up:

• 2.3 million conversations — about two-thirds of Klarna's customer service chats.
• The equivalent work of 700 full-time agents.
• Average resolution time dropped from 11 minutes to under 2.
• A 25% drop in repeat inquiries (it resolved things correctly the first time).
• Live in 23 markets, 24/7, across 35+ languages, with an estimated $40M profit improvement for 2024.

Those are Klarna's own figures, published with OpenAI (Klarna press release, OpenAI).

The part everyone skips

Here's what most "Klarna replaced 700 agents!" posts leave out: in May 2025, Klarna's CEO told Bloomberg the company had cut too deep on humans and was reopening hiring for premium support roles (CX Dive).

That's not a gotcha — it's the actual lesson. The AI didn't fail. The scope was wrong. The win was real for the high-volume, repetitive 70% of chats; the mistake was assuming it covered the messy, high-stakes 30% too.

What a normal business should take from this

You don't have Klarna's 2.3M monthly chats. You don't need them to win here. The transferable insight:

1. The repetitive majority of your support is automatable today — order status, "where's my refund," appointment changes, the same five questions. That's where the 11-min → 2-min drop lives.
2. Speed compounds. Sub-2-minute resolution isn't a vanity metric; it's the difference between a customer who stays and one who churns mid-frustration.
3. Keep humans on the 30%. The model that wins is AI-first, human-escalated — not human-replaced.

How you'd build the Klarna upside (at your scale)

• Voice Agents answer and resolve the repetitive inbound calls 24/7 — the phone equivalent of what Klarna did in chat — and hand off cleanly when it's genuinely complex.
• AI Agents handle the multi-step "look it up, check the policy, take the action" tickets that used to bounce between tabs.
• Call & Voice Analytics tells you which 30% to keep human on — the objections, the churn signals, the calls the AI shouldn't have taken.

You adopt one, measure the resolution-time and repeat-inquiry numbers on your data, and scale from there — without the overcorrection Klarna had to walk back.

Sources: Klarna, OpenAI, CX Dive.