AI Banking Support

In short

The simple FAQ bot at Bank CenterCredit poorly understood free-form inquiries and transferred almost every dialogue to an operator.

Outcome

According to the case author, AIBEK handles 55.6% of inquiries, and operator wait time decreased from 10.2 to 3.8 minutes after developing the knowledge base, LLM layer, and protection modules.

How the automation runs

Trigger

Client sends a message to the banking support channel

Automation steps

1. Checks incoming request for personal data and prohibited topics
2. Determines intent and selects appropriate module or knowledge source
3. Assembles response from current Knowledge Hub, considering bank policy and tone
4. Verifies the prepared response for factual accuracy, toxicity, and potential leakage
5. Responds, initiates an authorized action, or transfers dialogue to operator with context log

Human check

The product team manages knowledge and rules, security monitors restrictions, and the operator handles complex and non-standard inquiries.

Outcome

According to the case author, AIBEK handles 55.6% of inquiries, and operator wait time decreased from 10.2 to 3.8 minutes after developing the knowledge base, LLM layer, and protection modules.

Using it

When to use it

When the first line is overwhelmed with repetitive questions, but answers involve personal data, regulatory rules, and regularly changing products.

How to verify

Verify the share of correctly closed dialogues, escalations, wait times, knowledge freshness, and a sample of responses with experts; consider published figures as the case author's statement.

Tools