Deutsche Telekom
Natural-language self-care service chatbot, extended to voice and then rebuilt on an LLM (Frag Magenta 1BOT) to cover hundreds of contact reasons
Deutsche Telekom has run Frag Magenta since 2016, a customer service chatbot on chat and voice covering more than 380 contact reasons, which handled over 4 million dialogues in 2022 and resolves more than a third of requests immediately.
Key points
- Natural-language self-care service chatbot, extended to voice, across 380+ reasons.
- Rasa (NLU) foundation then a rebuild on the multi-agent LLM platform Frag Magenta 1BOT.
- More than 4 million dialogues in 2022, over a third resolved immediately.
- Evidence B, confirmed status.
Objective
Absorb as many routine requests as possible in self-care (billing, moving, outages, offers) before they reach an advisor, on both chat and phone.
The deployment
Frag Magenta is Deutsche Telekom's service chatbot, in production since fall 2016. The customer states the request in natural language and the assistant handles more than 380 different reasons, from moving to billing to outages and new offers. Since 2020 it also answers by voice on the hotline, and since fall 2023 Frag Magenta Voice relies on Rasa's voice capabilities. In 2022, Frag Magenta ran more than four million customer dialogues and resolves more than a third of requests immediately, with the rest handed to an advisor. Deutsche Telekom also launched Frag Magenta 1BOT, a rebuild on a multi-agent LLM platform, to move past the limits of scripted NLU and extend coverage. The stack is designed for the group's European customer base.
Results Proof B
Figures published by Deutsche Telekom on its own site (4 million dialogues in 2022, more than a third resolved immediately, 380+ reasons) plus press reporting quoting a group executive on the chat and voice scaling strategy. Primary brand source plus established press: solid on volume and deflection; the retention impact in points is not isolated.
How it works
Documented architectureThe stack in detail
- plateforme Rasa NLU for the original Frag Magenta stack and the voice capabilities of Frag Magenta Voice since fall 2023
- plateforme Frag Magenta 1BOT internal rebuild on a multi-agent LLM platform to move past scripted NLU; the underlying models are not disclosed
- infra Canaux chat et voix Telekom integration with the telekom.de chat, the MeinMagenta app, and the phone hotline
How it runs, concretely
For ops teams-
1Map the contact reasons Customer service team + AI
More than 380 reasons are modeled with their journeys, from moving to outages, so the assistant knows what to do with the request.
-
2Understand natural language AI / NLU
The NLU reads the customer's free-form phrasing and links it to a reason, in chat and by voice on the hotline.
-
3Resolve or transfer AI then human advisor
More than a third of requests are resolved immediately; the rest go to an advisor with the context already captured.
-
4Move to the LLM (1BOT) AI competence center
Rebuild on a multi-agent LLM platform to widen coverage and move past purely scripted answers.
The immediate resolution rate versus transfer. If it drops, the chatbot no longer deflects and pushes the load back onto advisors instead of absorbing it.
How your customers perceive this type of use
Sourced studiesLes consommateurs n'acceptent pas les chatbots par defaut : 64% prefereraient que les entreprises n'utilisent pas d'IA dans leur service client (Gartner, 2024) et pres d'un utilisateur sur cinq du service client par IA n'en retire aucun benefice (Qualtrics, 2025). L'acceptation se construit sur trois conditions mesurees par Salesforce : savoir qu'on parle a une IA, pouvoir escalader vers un humain, comprendre la logique de l'agent.
Acceptance conditions
- Etre informe qu'on parle a une IA et non a un humain (pres de 75% le demandent, Salesforce 2024)
- Un chemin d'escalade clair vers un agent humain (45% plus enclins a utiliser l'agent IA, Salesforce 2024)
- Une logique de l'agent clairement expliquee (44% plus enclins, Salesforce 2024)
Red lines
- Rendre l'humain injoignable : c'est la premiere inquietude des consommateurs sur l'IA dans le service client (Gartner 2024) et 50% craignent que l'IA les coupe du contact humain (Qualtrics 2025)
- Remplacer le service client par l'IA sans alternative : 53% envisageraient de partir chez un concurrent (Gartner 2024)
Sources: Salesforce 2024 · Gartner 2024 · Qualtrics 2025
How to replicate
Inference, not sourcedData prerequisites
- A map of contact reasons and their resolution journeys
- A corpus of conversations to train intent understanding
- Access to self-care systems to take action (billing, options, outages)
Org prerequisites
- An AI/customer service team that maintains the journeys
- A clean escalation process to advisors
Possible stack
- Rasa or a market LLM in RAG for understanding
- A voice component for the hotline
- API connection to service back offices
The plan, step by step
- Step 1Prioritize the 10 to 20 highest-volume contact reasons and map their resolution journeys.Deliverable: Catalog of reasons with modeled journeys
- Step 2Build language understanding (NLU or an LLM in RAG) and connect the self-care back-office APIs (billing, options, outages).Deliverable: A bot able to resolve the priority reasons in test
- Step 3Launch in beta on web chat with an escalation to an advisor that passes on the context already captured.Deliverable: Bot in limited production, immediate resolution rate measured
- Step 4Generalize chat, enrich the reasons continuously, and monitor the resolution / transfer ratio.Deliverable: Extended coverage with a deflection dashboard
- Step 5Extend the stack to voice on the hotline with a voice component.Deliverable: Voicebot in production on the highest-volume reasons
First step: Prioritize the 10 to 20 highest-volume contact reasons, model their journeys, and measure the immediate resolution rate against transfer before expanding.
The same pattern, proven in other industries
ADT
genAI customer agent
Air India
genAI customer agent (automated customer service)
Bank of America
conversational financial assistant at scale
Sources
- S1 The Digital Service Assistant / Frag Magenta - Deutsche Telekom Primary archive pending
- S2 Chatbots: BT and Deutsche Telekom share insights Established press archive pending
- S3 Artificial intelligence at Deutsche Telekom Primary archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.