ElevenLabs Cuts Klarna Support 10X — Now Runs On-Premise

Klarna disclosed a 10X reduction in customer support resolution time through AI automation — and the voice AI agents behind it now run entirely on your own servers, with zero cloud dependency. ElevenLabs, which powered Klarna's support overhaul in February 2026, launched on-premise and on-device deployment in April 2026, eliminating the data-sovereignty objection that had blocked enterprise adoption for years.

What Klarna's 10X Result Means for AI Customer Support Teams

Time to Resolution (TTR — the average time it takes to fully close a customer support ticket from first contact to resolution) is the metric most support leaders live and die by. Klarna — one of Europe's largest "buy now, pay later" financial services companies with over 85 million active users — achieved a 10X reduction in this metric using ElevenLabs Agents (AI-powered voice bots that handle customer calls end-to-end without a human in the loop).

To put that in concrete terms: if your support team previously averaged 10 minutes to resolve a query, a 10X improvement brings it to about 1 minute. At Klarna's scale — tens of millions of monthly interactions — that translates to dramatically lower cost and faster customer outcomes simultaneously.

Revolut, the digital banking platform with over 45 million customers, also deployed ElevenLabs for customer support in the same period. Toyota went a different direction: it built a fan engagement experience featuring an AI-powered version of NFL quarterback Brock Purdy — fully voice-driven, fully real-time, no human agent required.

On-Premise Deployment: The Change That Unlocks Enterprise IT

Until April 2026, ElevenLabs required a cloud connection — meaning every voice call, every piece of audio generated, and every customer interaction passed through ElevenLabs' external servers. For banks, hospitals, law firms, and government agencies, that was a hard blocker. Most enterprise IT security policies prohibit sensitive customer data from leaving the internal network.

The new on-premise deployment option (running the AI software on your own physical servers inside your data center, instead of on third-party cloud infrastructure) removes this barrier entirely. The on-device variant goes even further: models run locally on the end hardware — a laptop, a tablet, or an embedded device — with no internet required at inference time (the moment the AI actually generates and delivers a spoken response).

• On-premise: Your data center, your hardware, full regulatory control
• On-device: Runs fully offline on a single device — no server, no subscription required at runtime
• Cloud (existing): Fastest to deploy, but all audio data routes through ElevenLabs' infrastructure

This positions ElevenLabs directly against OpenAI's voice models (Whisper for transcription, and TTS for text-to-speech — tools that convert text into spoken audio), which remain cloud-only as of April 2026. For enterprise procurement teams evaluating voice AI, the question "will it run inside our environment?" is typically question one. ElevenLabs can now answer yes.

Where ElevenLabs Voice AI Agents Are Already Running in Production

The platform's enterprise footprint expanded significantly across late 2025 and early 2026:

• Harvey — legal AI platform using ElevenLabs to give lawyers real-time global voice capabilities across multilingual client calls (November 2025)
• Deutsche Telekom — powers AI-driven podcast creation inside the Magenta App (partnership launched March 2025)
• TVS Motor Company — deployed multimodal AI agents (agents that handle both voice and text simultaneously) across customer touchpoints in December 2025
• Immobiliare.it — Italy's largest real estate platform built a fully conversational AI property agent in days using ElevenLabs' developer tools
• Liberty Global — accelerating voice AI deployment across European markets (November 2025)
• Ukrainian Public Services — government digitalization use case, giving citizens AI-driven voice access to public services (September 2025)

The breadth here is the real story: legal services, telecom, automotive, real estate, government. ElevenLabs is not a voice cloning novelty — it's running in verticals where latency, accuracy, and data compliance are non-negotiable requirements.

The Open-Source Signal Behind ElevenLabs' AI Automation Push

ElevenLabs recently committed $22,000 to open-source projects through its OSS Engineers Fund (a program that financially supports the public software libraries ElevenLabs itself depends on to function). That's a small absolute number — but it's a deliberate strategic signal.

Over 30 GitHub repositories (publicly shared code collections where developers share and collaborate on software) already implement self-hosted alternatives to ElevenLabs for workflows like converting blog posts into podcast audio episodes. Rather than treating this grassroots competition as a threat, ElevenLabs is funding the ecosystem and releasing its own developer tooling into it.

ElevenLabs UI — its open-source library of audio and voice agent interface components (pre-built visual building blocks developers can drop directly into their apps) — was released in October 2025. A typical developer workflow using these tools looks like this:

# Blog-to-podcast conversion using ElevenLabs
# (pattern used in 30+ public GitHub implementations)
# Step 1: Input — blog post URL or text content
# Step 2: LLM summarizes and scripts the content into a spoken format
# Step 3: ElevenLabs TTS (text-to-speech) generates the audio episode
# Step 4: Output — podcast episode ready to distribute

ElevenLabs raised a $19 million Series A led by Andreessen Horowitz, with participation from Nat Friedman (former GitHub CEO) and Daniel Gross (AI investor). The Voice Translation post — ElevenLabs' most community-discussed product launch — generated 129 upvotes and 51 comments on Hacker News, the highest engagement of any blog post in the company's recent history. With on-device deployment now shipping, the $19M appears to be funding enterprise infrastructure depth rather than headline model improvements.

Three Things to Do Before Your Next Voice AI Evaluation

If your team is evaluating voice AI for customer support, internal productivity, or content production, the ElevenLabs data gives you three concrete actions right now:

1. Use 10X as your benchmark anchor. Klarna's result (10X reduction in Time to Resolution, February 2026) is the most concrete public metric in enterprise voice AI today. When any vendor pitches you, ask them to match it or explain the gap with their own customer data.
2. Re-open evaluations that were blocked on cloud-only restrictions. If your IT or legal team previously ruled out voice AI due to data residency or sovereignty requirements, ElevenLabs' on-premise option is a new answer to that old objection. The conversation is worth revisiting.
3. Start free with the open-source components. ElevenLabs UI components are open-source — you can build and test a voice agent workflow without a paid account. Visit the AI automation guides for step-by-step walkthroughs, or get started here to run your first agent in under an hour.

Related Content — Get Started | Guides | More News