United StatesFinance and insurance2024 – ongoingBankUnited
SAVI policy assistant for banking staff
BankUnited's SAVI answered policy questions in under 10 seconds, according to AWS
Problem
Employees struggled to find policy information while serving small-business customers. Searches prolonged calls and produced inconsistent answers.
Approach
BankUnited built SAVI using Claude 2 on Amazon Bedrock and Amazon Kendra search. Employees asked natural-language questions against approximately 400 policy and procedure documents. The application supported round-the-clock self-service.
Results
Takeaway
Frontline staff could resolve procedural questions without waiting for the support team. The reported accuracy rate did not establish that every answer was correct.
Categories
- Industry
- Finance and insurance
- Business function
- Customer serviceInternal work and knowledge management
- Innovation type
- Process innovation
- AI technology
- Generative AI
- Adoption
- Built in-house
- Evidence
- Vendor case study
Sources
Source summaries
AWS described SAVI's use by banking employees to retrieve policy answers. It reported 95% accuracy and responses in under 10 seconds, alongside fewer procedural calls and emails to support staff.
Full details are in the original sources.