Bradesco
Bradesco reports its generative AI chat resolves 82% of first-level customer requests
Bradesco, a large Brazilian bank, began using its own generative AI platform, Bridge, in April 2024, alongside its BIA virtual assistant. A bank director says the AI chat resolves 82% of first-level customer requests, and more than 10 million customers use it.
Key figures
Copied exactly from the sources. Copy any figure with its citation.
82%
Resolution rate at the first level of customer service with generative AI, as stated by a Bradesco director in October 20251
More than 10 million
Customers who use the AI chat, as stated by a Bradesco director1
40%
Productivity gain in technology work, as stated by a Bradesco director1
90%
Customer chatbot conversations resolved without human intervention, against a 50% target (Bain case page, undated)2
What it took
The problem
Bradesco has used AI for almost a decade. Its virtual assistant, BIA, started in 2016 on IBM Watson, a traditional AI system. Bain says that when the bank decided to improve the assistant, the work became a way to spread generative AI across the whole organisation.
What they built
In April 2024 Bradesco began using Bridge, its own generative AI platform. Generative AI writes answers from text rather than following fixed scripts. BIA helps customers and helps internal staff with product and procedure questions. Bain also describes a chatbot for call-centre staff. A director also describes virtual developer teams and an AI tool that reviews collections calls and advises agents.
How they did it
Bain describes a low-risk first step: upgrading the internal chatbot for branch managers, which took 12 weeks. The customer chatbot followed in eight weeks. Bain also helped build the platform so it is not tied to one language model, and trained technical, business and customer experience teams. Bradesco added guardrails (rules that limit what the assistant answers) because customers talk to BIA about everything, including topics far from banking.
Results
A Bradesco director said at a Bloomberg Linea event in October 2025 that the AI chat resolves 82% of first-level requests, that more than 10 million customers use it and that the bank runs about 450 AI use cases. Bain's page reports 90% of chatbot conversations resolved without a human against a 50% target, 95% accurate answers and a 40% cut in handling time at the call centre in a pilot. The director also cites a 40% productivity gain in technology work.
All reported figures
| Measure | Figure | Citation |
|---|---|---|
| Resolution rate at the first level of customer service with generative AI, as stated by a Bradesco director in October 2025 | 82%1 | |
| Customers who use the AI chat, as stated by a Bradesco director | More than 10 million1 | |
| Productivity gain in technology work, as stated by a Bradesco director | 40%1 | |
| Customer chatbot conversations resolved without human intervention, against a 50% target (Bain case page, undated) | 90%2 | |
| Accuracy of customer chatbot answers (Bain case page, undated) | 95%2 | |
| Reduction in average handling time with the call centre chatbot, during the pilot (Bain case page) | 40%2 |
Lessons for your project
Start with a low-risk internal use. Bradesco first upgraded a chatbot for branch managers, then used that success to move into customer-facing tools.
Build the platform to work with more than one language model. Bain says Bradesco's platform is model-agnostic, so the bank is not tied to a single model.
Put limits around the assistant's scope. Bradesco built guardrails because customers talk to the assistant about topics outside banking, and money transactions add risk.
Train business, technical and customer experience teams together, not only the engineers.
Caveats
- The 82% and 90% figures describe different scopes and sources and must not be combined. The 82% is first-level customer service per a Bradesco director; the 90% is the share of chatbot conversations resolved without human help, per Bain.
- All figures are company statements. One is reported by Bloomberg Linea from a conference panel; the others come from the page of Bain & Company, the bank's consulting partner. No audit or independent measurement is cited.
- The Bain case page has no date. The 40% handling-time cut was measured in a pilot.
- Neither source names the language model. Bradesco's earlier BIA ran on IBM Watson, a different technology; the generative AI platform is described as model-agnostic.
- Go-live is taken from the April 2024 start of the Bridge platform per Bloomberg Linea; the sources do not give separate launch dates for each tool.
Sources
- IA generativa resolve 82% dos atendimentos iniciais no Bradesco, diz diretora
Bloomberg Linea · 2025-10-28 · News · bloomberglinea.com.br
- Uma grande aposta em IA generativa coloca o Bradesco na frente
Bain & Company · Undated · Vendor case study · bain.com
Summarized in our own words from the sources above. Spotted an error? See corrections.