Deutsche Telekom
Deutsche Telekom's Frag Magenta assistant now uses an LLM; it held almost ten million dialogues in 2024
Deutsche Telekom's customer-service assistant Frag Magenta now uses generative AI to understand customers' requests. The company says it held almost ten million customer dialogues in 2024 and solves more than a third of requests immediately.
Key figures
Copied exactly from the sources. Copy any figure with its citation.
almost ten million
Customer dialogues held by Frag Magenta in 2024 (German: fast zehn Millionen Kundendialoge)1
more than a third
Share of requests the chatbot solves immediately (German: Über ein Drittel der Anliegen)1
over 380
Types of request the chatbot can help with1
about 600,000
Dialogues per month, as stated in June 20252
What it took
Cost, team size and timeline were not disclosed in the sources.
The problem
A telecom operator gets a constant flow of service requests: bills, contracts, faults and moves. Telekom says customers expect answers around the clock, and that a digital assistant takes routine work off its service staff.
What they built
Frag Magenta ("Ask Magenta"), Telekom's digital assistant. Customers use it on the website, in the MeinMagenta app, and on WhatsApp and Apple Business Chat. On the service hotline it takes over some calls, for example about a landline fault or a bill, and understands natural speech, including spoken account numbers.
How they did it
Telekom says the assistant now makes wide use of generative AI to catch nuances and give answers based on context. A June 2025 Telekom release says it uses a large language model, which lets customers change payment and customer data and contracts in the chat. Telekom does not name the model or its vendor.
Results
Telekom reports almost ten million customer dialogues in 2024, with more than a third of requests solved at once. The assistant covers over 380 types of request. In June 2025 Telekom put the volume at about 600,000 dialogues a month. The trade magazine CHIP ranked it first among telecom assistants, with an overall grade of 1.56 according to Telekom. Unresolved requests go to human advisers.
Lessons for your project
Upgrade an existing assistant rather than start over. Telekom added a large language model to a bot it already ran, which widened what it could do, such as changing contract and payment data.
Report resolution, not only volume. Telekom publishes the share of requests solved at once next to dialogue counts.
Keep a human route. Requests the bot cannot solve go to human advisers.
Be careful what you credit to generative AI. Telekom's figures cover the whole assistant, which predates the LLM.
Caveats
- All figures are Telekom's own. The independent CHIP ranking is cited by Telekom, and its method was not read.
- Frag Magenta existed before generative AI. The dialogue and resolution figures describe the whole assistant, so they cannot be credited to the LLM alone.
- The sources say the assistant relies widely on generative AI (a large language model) but give no start date for that, no vendor and no model. The year shown is the first source date (2025) that documents LLM use. Because the 2024 dialogue count predates that, the title does not credit the 2024 volume to the LLM.
- The voice portal Frag Magenta Voice is reported as available on the mobile hotline since autumn 2023, and the sources do not say whether it uses generative AI.
- No cost, savings or team figures are published in the sources read.
Sources
- Unser Chatbot hilft rund um die Uhr
Deutsche Telekom newsroom · 2025-09-22 · Company blog · telekom.com
- Frag Magenta bester Chatbot der Branche
Deutsche Telekom newsroom · 2025-06-13 · Press release · telekom.com
Summarized in our own words from the sources above. Spotted an error? See corrections.