How pervasive is the use of artificial intelligence in the field of financial risk management? What are the key challenges in AI implementation over the next two to three years? These issues were examined in early 2019 via the webinar, Operationalizing AI and Risk in Banking, sponsored by the Global Association of Risk Professionals (GARP).

“We found exceptionally high rates of AI usage among survey respondents,” said Katherine Taylor, Senior Data Scientist at the software company SAS. On January 24, 2019, Taylor was the first of two panellists who presented a “sneak peek” at a joint survey by SAS and GARP.

Currently, AI is being used in processes such as credit scoring and model validation. More than half of survey respondents use AI in process automation. Within the next three years, AI will be applied to such areas as regulatory reporting and loan provisioning.

Of those who use AI, over 80 percent are seeing benefits from adopting AI, “especially for efficiencies and cost reduction in process automation,” Taylor said.

“We do expect to see advantages for analytics,” she said, with improved model accuracy and faster model execution. New forms of data can be accessed, such as text, image, and audio.

The Challenges of AI

The survey results were consistent with some customer feedback that SAS has already been receiving.  “The number one challenge is data availability and quality,” Taylor said. “Number two challenge is the lack of understanding among stakeholders.”

“AI models have a reputation for being hard to explain,” because they do not lend themselves to human intuition. She called this the “explain-ability issue.” There are some common techniques to explain the models, such as referencing an easier, less accurate model.


Tips for Using AI

Taylor had seven “quick tips” for implementing AI (slide 14 of the presentation deck, available from the GARP website). Of these, the most notable was her recommendation to “build data science skills in existing risk talent.” She said the average data scientist has a one-year turnover rate. When management decides to use “existing talent,” this not only provides continuity, it ensures a clearer understanding of the business issues.

Taylor gave an example use case, in which AI could quickly re-price a portfolio on demand. It’s a good case to show, she said, because “it’s a specific problem and there are large amounts of data.”

The more widely AI is used, the more applications people will discover, and the more helpful the “tips” on AI will become. ♠️


Click here to read about the second panellist, Mahdi Amri.

Click here to view the recorded webinar Operationalizing AI and Risk in Banking.

Credit for thumbnail picture of robot hand on keyboard: