Klarna AI Assistant Handled 700 Agents Worth of Work in First Month

Klarna’s AI assistant handled the equivalent workload of 700 full-time support agents in its first month and matched human customer satisfaction scores while doing it. That result, now sitting alongside similar disclosures from Anthropic Delta Air Lines and four other companies, makes 2026 the year AI agents moved from pilots to operational infrastructure in customer service.

Klarna Cuts Support Costs by $40 Million

Klarna, the buy-now-pay-later provider, launched its AI assistant in February 2024, handling 2.3 million conversations and the equivalent of 700 full-time agents’ workload in its first month, with a projected $40 million profit impact. By Q3 2025, the system’s footprint had grown to the equivalent of 853 full-time employees and $60 million in savings, though notably, Klarna began rehiring human agents in May 2025 after complaints about AI response quality on complex cases, with CEO Sebastian Siemiatkowski acknowledging the AI-only approach had produced ‘lower quality’ outcomes on certain queries.

That result did not come from dropping an AI layer onto existing systems. It followed a foundational effort to standardise customer data architecture and build a reliable tool-calling layer over Klarna’s backend, the kind of infrastructure investment that tends to precede the cases that get cited. Teams expecting similar returns without similar groundwork are likely to be disappointed.

Anthropic Saves 1,700 Hours with Intercom Fin

Anthropic selected Intercom‘s Fin AI Agent for its customer support operations, with a stated priority on safety and reliability. Within just over a month, Fin resolved 58% of roughly 50,000 monthly support interactions, saving Anthropic’s team more than 1,700 hours and freeing agents to handle complex, escalated issues.

Intercom reports that Fin averages a 76% resolution rate across its 12,000-plus customers, with many reaching above 85%, and that it participates in 96% of conversations handled through the platform. Those figures come from Intercom’s own reporting and have not been independently verified.

Octopus Energy Boosts Customer Happiness 18%

Octopus Energy took a narrower approach, one of the earliest documented cases in this space: generative AI drafts email responses, which human agents then review and send. In a widely-cited 2023 BCG study, the AI-assisted emails produced an 18% increase in customer happiness scores compared with fully human-drafted replies. The approach reportedly now covers one-third of customer inquiries, though more recent figures haven’t been independently confirmed.

Delta Air Lines Boosts Satisfaction Scores with Delta Concierge

Delta Air Lines deployed Delta Concierge, an AI-powered assistant in the Fly Delta app, in beta in October 2025 to handle routine self-service tasks: check-in, bag tracking and flight searches. These interactions are high-volume and predictable, which makes them practical targets for automation. Delta reports the tool boosted customer satisfaction scores by 25 points during travel disruptions, though a specific call-centre volume reduction figure hasn’t been independently confirmed.

Everise Tests Voice AI on Its Own IT Help Desk

Customer experience firm Everise took a different tack, piloting voice AI internally on its own global IT help desk, handling employee calls about account access, software issues and telephony problems, before considering it as a client-facing offering. The system contained 65% of calls without human escalation, cut wait times from five to six minutes to zero, and saved 600 staff hours a month. It’s a notable proof point precisely because it’s Everise testing the technology on itself before selling it, though it’s a different kind of deployment than the customer-facing cases above.

Genesys Delivers 9.8x ROI on Its Own Platform

Genesys used its own Genesys Cloud CX platform with agentic AI capabilities on its internal product support operations, a useful test case precisely because the vendor is also the customer. Over three years, the company reports 157,000 cumulative working hours saved and a 9.8x cumulative ROI, built in stages: advanced routing automation cut case escalations by 43%, and the later addition of Agent Copilot delivered a further five-minute drop in average handle time.

Cricut Cuts Wait Times 89% with Zoom Virtual Agent

Cricut, the smart cutting machine company, deployed Zoom Virtual Agent to manage support inquiries. Cricut’s own systems manager reports a 50% self-service containment rate, along with an 89% reduction in call wait times and a 90% cut in call abandonment. Zoom’s own marketing materials cite a higher 87% figure in some places, though that number doesn’t appear in Cricut’s directly-quoted case study.

Across these seven deployments, a consistent structure emerges: a measurable containment or resolution target, infrastructure built before the agent goes live, and human agents repositioned to handle what the AI cannot. The rapid expansion of agent tooling across vendors reflects the same pressure visible in these cases. The Salesforce survey finding that adoption in service organisations rose from 39% to 66% in 12 months tracks with production deployments that carry defined performance expectations, not experimentation. How widely that discipline holds across organisations without Klarna’s or Genesys’s resources is a harder question, and the public evidence is limited to the cases companies choose to publish. For more analysis on enterprise AI strategy, visit our Enterprise AI section.