Summary
Is AI just hype, or does it have a real business impact? Can it understand banking products? We built an AI chat app to answer those questions and the test showed a 76% accuracy in answers and provided valuable insights for better AI integration.
Challenge
What was the need
Slovenská sporiteľna decided to verify the ability of Open AI model to effectively understand banking products and process knowledge.
Solution
How we solved it
We delivered a Proof-of-concept chat app to verify two hypotheses:
- how exact and well answers we can get by using as-is process and product documentation,
- how much customer support time can be saved.
We delivered a standalone chat application connected to Microsoft Azure GPT models – with embedded knowledge in areas of mortgages, consumer loans, current accounts, deposit and corporate products.
Finally, we verified the quality in Pilot test with branch employees.
Tools & means
- React FE
- Python BE app
- Usage of Langchain
- Chroma vector BD
- Azure open AI LLM
Outcomes
What has changed
We have created complex process and product documentation, set up environment on premise and implemented establishment of evaluation metric to measure quality of the solution.
The test showed a 76% accuracy in answers and provided valuable insights for improving product and process documentation to better support AI integration.