Brazilian bank Bradesco deploys IBM Watson chatbot, cutting customer wait times from 10 minutes to seconds
Trained with IBM Watson in Portuguese on more than 10,000 customer questions, Bradesco's AI chatbot reduced customer waiting times from an average of 10 minutes to near-instant responses, handling 283,000 questions per month with 95% accuracy after 10 million interactions.
Background
Bradesco's contact center was receiving complaints about excessive customer waiting times. As a major retail bank offering dozens of products, a large share of customer inquiries were repetitive — questions about balances, transactions, account procedures — that did not require specialist agent expertise. These routine inquiries were consuming agent capacity and creating wait-time backlogs.
What Was Implemented
- Partnered with IBM Watson to develop a Portuguese-language AI chatbot
- Trained on more than 10,000 distinct customer questions over approximately 10 months
- Deployed to handle incoming customer inquiries for Bradesco's retail banking product line
- Continuously improved through interaction data, expanding from initial deployment to coverage of 62 products
Results
Customer waiting times reduced from an average of 10 minutes to near-instantaneous (seconds) . After 10 million interactions, the chatbot handles 283,000 questions per month across 62 products at 95% accuracy (96% at initial deployment). Human agents redirected to higher-complexity interactions.
Lessons
- Training a domain-specific NLP model in a non-English language (Portuguese) with thousands of question variants takes meaningful time (~10 months) but creates durable domain expertise
- Continuous retraining on accumulated interactions (10M+) enables both expanded product coverage and sustained accuracy over time
- Routing routine FAQ-type inquiries to AI reduces waiting times for all customers — including those with complex needs — by freeing agent capacity