Telefonica
Multi-brand customer conversational assistant: a single natural-language AI base handles products, billing, and requests, declined by brand and by language, then enhanced with generative AI
Telefonica has operated Aura since 2018, a conversational assistant declined by brand (Movistar, Vivo, O2) that handles products and billing in natural language across more than 30 channels and claims about 400 million interactions per year.
Key points
- Multi-brand self-care conversational assistant (Movistar, Vivo, O2) in natural language.
- NLP base on Azure Cognitive Services, later enhanced with generative AI.
- About 400 million interactions per year across more than 30 channels.
- Evidence B, confirmed status, active since 2018.
Objective
Give customers a single natural-language entry point to manage their subscription and bills, brand by brand, and absorb the volume of contacts without routing it to advisors.
The deployment
Aura is Telefonica's conversational assistant, launched in 2018 under the group's brands (Movistar in Spain and Latin America, Vivo in Brazil, O2 in the United Kingdom and Germany). The customer speaks or writes in natural language to check their bill, manage their services, or resolve a question, through the app, TV, the Movistar Home device, the web, or the call center. The base was first trained on Azure Cognitive Services to fit the accents and phrasing of each country, then Telefonica added generative AI for personalized responses and content creation. The group claims about 400 million interactions per year and a presence on more than 30 channels, on a base of more than 350 million customers. Seven years after its launch, Aura remains the group's reference conversational channel.
Results Proof B
Case documented by Telefonica (official transformation handbook: 400 million interactions/year, 30+ channels) and by a Microsoft customer story at launch (markets, Azure technology). Concordant official Telefonica and vendor sources; the figures cover volume and coverage, not a retention impact in points.
How it works
Documented architectureThe stack in detail
- plateforme Microsoft Azure Cognitive Services (Language Understanding) Natural language understanding at launch, trained on the accents and phrasing of each market.
- plateforme Socle Aura (in-house) Telefonica's internal conversational base, declined by brand (Movistar, Vivo, O2) and by language, connected to the products and billing systems.
- llm Brique generative ajoutee a Aura Generative AI added for personalized responses and content creation; provider and model not named publicly.
- infra Integration multi-canaux More than 30 channels connected to the same assistant: apps (Mi Movistar, Meu Vivo, MyO2), TV, Movistar Home, web, call center, Facebook Messenger, Google Assistant.
How it runs, concretely
For ops teams-
1Train a base per brand central AI / data team
The same conversational engine is trained on the accents, phrasing, and offers of each country rather than duplicating an assistant from scratch.
-
2Connect all channels local digital teams
App, TV, home device, web, call center, and third-party channels point to the same assistant, for a consistent answer whatever the entry point.
-
3Handle products and billing AI / Aura
The assistant executes self-care actions (check a bill, activate an option) in addition to answering, to avoid handoff to an advisor.
-
4Enhance with generative AI AI / data team
Addition of personalized responses and content creation where NLP scripted fixed answers.
The quality of intent understanding per brand. If the model misreads a local phrasing, it routes the customer to an advisor and the deflection benefit collapses.
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
- Corpus of intents and conversations per brand/language
- Access to the billing and service management systems to act, not just answer
- GDPR governance over customer data
Org prerequisites
- Central AI/product team plus local digital relays
- Ability to integrate the assistant with all existing channels
Possible stack
- Azure AI, Google Dialogflow/Vertex AI, or a market LLM in RAG
- Multi-channel orchestration layer
- API connection to the billing/services back-offices
The plan, step by step
- Step 1Frame a first market and two or three high-volume self-care intents (check the bill, change an option).Deliverable: Corpus of intents, target journeys, and deflection objectives.
- Step 2Train the NLU on the local language and phrasing, connect the billing and service management APIs so the assistant acts, not just answers.Deliverable: Assistant able to execute the self-care actions in pre-production.
- Step 3Launch on a first channel (the app), measure deflection and resolution rate, adjust the misunderstood intents.Deliverable: Quantified review of the pilot channel with resolution rate per intent.
- Step 4Open the additional channels (web, call center) and industrialize retraining as the catalog evolves.Deliverable: Multi-channel base in production and intent update process.
- Step 5Replicate brand by brand and country by country on the same base rather than duplicating assistants.Deliverable: Localization playbook (language, offers, channels) per market.
First step: Frame a first market and two or three high-volume self-care intents (bill, option change), measure deflection and resolution rate before opening other channels and countries.
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 Aura, Telefonica's Artificial Intelligence - Transformation Handbooks Primary archive pending
- S2 Aura, Telefonica's AI, learns the language of people to transform customer engagement Interested party archive pending
- S3 Telefonica launches Aura and leads the integration of AI in its networks and customer care 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.