Perhaps the most ubiquitous form of generative AI on the market, ChatGPT is an obvious place to start when adopting AI within risk management.

The use of ChatGPT is a key priority for many risk leaders in our network, but few have started using it in their day-to-day work. There's a lot of trial and error, and many risk leaders struggle to find time to really run with it on top of their existing workload.
To support them, we arranged for a risk modelling expert to present at a virtual member meeting, showcasing how some risk leaders are successfully using ChatGPT to enhance their risk function, and where members could start using the technology in their risk processes.

Here are some high level tips shared at the virtual meeting.
Using a chat interface to support risk management
Model selection
Framing of questions
Discussion sequences
Develop an application programming interface (API)
Developing an API can help you to complete a multitude of tasks in a fraction of the time they would normally take.
To run an API, you can set up an Excel sheet with a prompt and content to analyse, before inputting this into the model. This document should also include some code on how the model should run, which may require technical expertise to set up.
Key features of an API include:
Automation
APIs allow you to develop repeatable tools (or "agents") to undertake certain tasks on your behalf. These have the power to produce multiple answers or "trees of decisions".
Recursion
APIs can help you develop chained processes, which take the outputs of one task, split them up and feed them into other tasks.
Structured outputs
Through basic coding, it is possible to define exact structures of frameworks for ChatGPT responses; for example, splitting outputs reliably and organising them in forms.
Function calling
Advanced models can leverage mathematical calculations and algorithms, which are "called" from external inputs (as opposed to the model working this out itself, which it is not designed for).
The most important key principle when using this technology is to never trust the final output completely — GenAI can do 95% of the work, but you have to put in the other 5% and perform due diligence on any outputs provided.
What's next?
These are just a few tips shared during a 90 minute virtual meeting with an AI and modelling risk expert. Members were able to see how it worked on screen, ask questions, and discuss specific applications to their business. Enquire about membership here.
Share this
Related posts you may be interested in

How to apply AI to risk management

AI and risk management - 5 actions your peers are taking now
