BNP: Integrating AI Into ESG Scenario Analysis

The following Q&A occurred during the Everyday AI Conference in Paris.

Watch Video

What Are the Steps in a Stress Testing Use Case?

When we do a stress testing use case, we have basically four steps. The first one, and this is a critical one, is the design of the relevant scenario. Depending on the type of analysis we want to conduct, we have to find a relevant scenario which is both sufficiently severe, but still plausible, which is a key concern. 

When we work, for example, on climate analysis, we will have to mix several risk drivers and do that at a sufficiently detailed level. After that, we can go to the second step, which is gathering the data that will then be used in our models that we run, applying the scenario with design.

Then, the last phase is to provide a restitution of the analysis with various levels of the technicality of our outcomes, so that they can be fine-tuned to the various users we have. Still, when we think about climate scenarios we have a lot of different stakeholders and so, we will have to adapt the sophistication level of the comments regarding the outcome to the type of user we get. 

How Do You Use Dataiku?

Our organization has been leveraging Dataiku now for six years, so it’s been a long story with Dataiku. The beginning of the story was around data management. We have a lot of data sources and data preparation work, with several teams that interact in different locations, with different complementary competencies. Dataiku has been a wonderful platform to process work efficiently and manage a direct collaboration between these various teams. 

Going forward, we extended the use of Dataiku to the modeling part. The ability to run our models in our processes goes beyond the data preparation to the results of our analysis. We also try, at least for the technical users, to leverage Dataiku to get an automatic detailed analysis of our results. 

What Main Benefits Does Dataiku Bring You?

When we started to use Dataiku, the first benefit was the speed — the speed of delivery and the speed to convince people. I was quite amazed by that. My teams were mostly SAS users and the timing to be up and running in the solution was incredibly fast.

The second benefit is the ability to have a detailed process operated by teams that are not in the same location or belonging to the same culture, and that efficiently collaborate on the platform in a smooth way. For us, that is a significant benefit of Dataiku.

Going forward we expect to leverage other features in Dataiku, notably the ability to run production environments and design environments in parallel. We will also go for the ability to model in Dataiku, using all the AI suites which we have started working on and have great expectations for.

What Does Everyday AI Mean to You? 

I believe that the challenge is to embark on AI in our daily processes and operating framework. Today, it’s true that we have some use cases of AI, but very few are in production. We are more in a proof of concept (POC) phase of some AI use cases. For me, Everyday AI is the next step. It’s the phase during which more POCs will be pushed to production, and we will be using AI safely. This also includes knowing when to use AI and when to keep using more traditional approaches.

Everyday AI is the ability to have a sound, free, and without burden use of AI in our processes.