Garanti BBVA
end-to-end genAI customer assistant
Ugi, the assistant in Garanti BBVA's mobile app migrated to an LLM infrastructure, assisted more than 5 million customers and held more than 60 million conversations in one year, understands 90% of requests, and covers more than 300 end-to-end banking transactions.
Key points
- Ugi, the assistant in the Garanti BBVA mobile app, moves from a rules engine to an LLM infrastructure.
- Over one year: more than 5 million customers assisted and more than 60 million conversations.
- Understands 90% of requests and covers more than 300 end-to-end banking transactions.
- Level B evidence, two consistent official releases; live status confirmed (the service's tenth anniversary, May 2026).
Objective
Handle most customer requests in self-service within the app (card requests, card payments, transfers, card passwords, information) and absorb the daily volume away from costly channels, by pushing a conversational experience that anticipates needs.
The deployment
Ugi is the assistant built into the mobile app of Garanti BBVA, BBVA's Turkish subsidiary. Launched in 2016 as a rules-based chatbot, it has been migrated to a large language model infrastructure, which lets it track the context of a conversation and anticipate the customer's needs. The customer talks to Ugi in natural language to carry out operations: a credit or debit card request, a card payment, a card-password transaction, a transfer, or an information request. In the twelve months before the April 2025 release, Ugi assisted more than 5 million customers and held more than 60 million conversations. The May 2026 release (the service's tenth anniversary) states that it understands 90% of requests, covers more than 300 end-to-end operations, has 1.6 million active users, and handles on average more than 6.4 million interactions per month.
Results Proof B
Figures self-reported by Garanti BBVA in two consistent official releases over more than a year (the genAI upgrade of April 2025; the service's tenth anniversary in May 2026), redistributed by a press wire. Primary sources from the subject brand, with no third-party vendor bias, but not tied to a financial disclosure: hence a B rather than an A.
How it works
Inferred typical approachThe internal detail is not public. Here is a proven approach that leads to the same result, to adapt to your stack.
The stack in detail
- outil Ugi (assistant genAI base LLM) The mobile app assistant, initially rules-based, migrated to an LLM infrastructure that enables context tracking and anticipation of needs. The sources do not name a specific third-party model.
- plateforme App mobile Garanti BBVA Ugi's channel, integrated with accounts and banking operations in real time.
How it runs, concretely
For ops teams-
1Customer request customer
The customer states their request in natural language in the app (card request, payment, transfer, card password, information).
-
2Understanding and context tracking AI
The LLM engine interprets the intent and follows the thread of the conversation; it understands 90% of requests.
-
3Executing the operation AI
Ugi carries out the banking operation end to end among more than 300 covered transactions, drawing on account data.
-
4Anticipation and notifications AI
Outside of requests, Ugi anticipates needs and pushes instant personalized notifications.
-
5Continuous learning and supervision data team
The in-house team uses transactions and user feedback to widen the scope handled and fix poorly understood cases.
The intent the customer expresses in natural language, plus the account and transaction signals that feed the anticipation of needs. Without up-to-date account data and an intent history for continuous learning, understanding and proactivity degrade.
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
- Real-time account, card, and transaction data
- A history of customer intents for training and continuous learning
- A reference of banking operations eligible for self-service
Org prerequisites
- A dedicated digital and data team over time
- A banking security and compliance framework for the operations executed
- Governance of proactive notifications (frequency, relevance, consent)
Possible stack
- A large language model with context tracking
- An orchestration layer linking the understood intent to the execution of operations
- Deep integration with the core banking system and the mobile app
The plan, step by step
- Step 1Map the most frequent operations and requests in the app from support logs and customer journeys.Deliverable: A reference of intents and operations prioritized by volume.
- Step 2Migrate the understanding layer from a rules engine to an LLM able to track context, in a test environment.Deliverable: A read-only genAI assistant, tested on real data.
- Step 3Connect end-to-end transactional execution for a first batch of operations, with security and compliance sign-off.Deliverable: An assistant able to execute a subset of operations in beta.
- Step 4Widen the scope to sensitive operations (transfers, cards, passwords) and add anticipation and proactive notifications.Deliverable: A generalized assistant covering a broad catalog of operations.
- Step 5Set up continuous learning on transactions and feedback to improve the understanding rate.Deliverable: A continuous improvement loop in production.
First step: Start from an existing rules-based assistant or conversational FAQ, map the most requested operations, then migrate understanding to an LLM before opening up transactional execution.
The same pattern, proven in other industries
ADT
genAI customer agent
Air India
genAI customer agent (automated customer service)
AT&T
Governed enterprise genAI assistant: an LLM connected to internal knowledge, open to tens of thousands of employees and called at scale by automated systems
Sources
- S1 Garanti BBVA enhances its smart assistant Ugi using generative artificial intelligence Primary archive pending
- S2 Garanti BBVA's smart assistant Ugi turns 10: AI transforming banking Primary archive pending
- S3 Garanti BBVA Advances Digital Banking with Generative AI Upgrade to Smart Assistant Ugi Secondary 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.