TürkiyeFinance and insurance2024 – ongoingGaranti BBVA
Ugi virtual assistant rebuilt on large language models
Garanti BBVA rebuilt its Ugi banking assistant on an LLM, and the bank says it now understands 90% of requests
Problem
Ugi, launched in 2016, ran on rules and answered predefined questions. The bank wanted an assistant that could follow a conversation and handle more banking tasks in the mobile app.
Approach
The bank moved Ugi to an LLM-based setup, a shift it says took about two years. In May 2024 a Garanti BBVA executive told the Turkish outlet Ekonomist that the bank was building its own LLM models and testing Ugi with internal users. The new assistant is linked to product catalogs, customer profiles and transaction data and works by text or voice. Cards, payments and transfers were among the tasks added.
Results
Takeaway
Ugi gained conversational and banking functions as the bank moved it from rules to an LLM over about two years. The bank did not explain how it measured the 90% understanding rate.
Categories
- Size
- Large enterprise
- Industry
- Finance and insurance
- Business function
- Customer service
- Innovation type
- Product and service innovationProcess innovation
- AI technology
- Generative AIConversational and voice AI
- Adoption
- Built in-house
- Evidence
- Company's own claim
Sources
Source summaries
BBVA says Ugi moved from rules to an LLM infrastructure over about two years. Ten years after Ugi's launch, it reports 6.4 million average monthly interactions, 1.6 million active users, 300+ end-to-end transactions and 90% of requests understood.
A Garanti BBVA executive told Ekonomist that the bank develops its own LLM models, made Ugi smarter and more personalised, and has Ugi in a pilot with internal users. The executive also mentioned an LLM co-pilot for software engineers but gave no Garanti-specific figures.
Full details are in the original sources.
