Our journey to an AI Future

By Arnab Mitra, Programme Manager, BIAN

A few months ago, I wrote about how BIAN is experimenting with the use of AI to determine how we can get the most out of the technology.

Since then, we have been focused on generating sample data for the BIAN Service Domain APIs. This will help us create a sandbox with quality test data that developers can use to mock up innovative solutions using these APIs.

Following interest from our members, we have also started to pilot an AI-based API mapping app. Anyone embarking on a journey to map their APIs to BIAN’s standards face a huge challenge in identifying which BIAN APIs and endpoints their APIs map to, which can take thousands of hours of human effort. Our aim is to try and automate this by 50 to 60%, with humans coming in only to verify and approve the recommendations of the AI mapping app.

We’re currently in the process of finetuning the app, following feedback from our members. This has been primarily around improving the quality of the data we train our large language models (LLMs) with. This quality improvement also includes how we train our LLM with the data we have.

Improving our offering

As we receive feedback from user testing, we are able to test our AI capabilities with real data, both in terms of the BIAN Chat and the API Mapping App. This has enabled us to identify and improve our mapping methodologies and the way our AI engine links the various BIAN artefacts.

This has helped us overcome challenges around the quality of input data, while improving the instructions we’re able to give to the LLM to find results.

Following these positive results, we’re looking forward to having sample test data across the entire landscape. I remember thinking that it would take at least a year for us to do this manually when looking into doing this a few years ago. With AI once the process is in place, it will take us less than an hour.

What does the future look like?

We’re also looking forward to how we can generate use cases in banking based on all the BIAN data, such as using BIAN service domains to identify a bank’s existing customers that may be likely to leave. This would allow the bank to create a campaign to offer them better rates on term deposit products for example, to increase customer retention. We can then instruct the LLM to be very factual or get creative. When it gets creative, we often see it return results that do not make sense. So, the challenge would be to find that balance between something factual and creative, which will benefit our members and improve the industry.

How is your businesses’ AI journey going? We’d love to hear from you!