Risk management is the next frontier for AI adoption

Artificial intelligence and machine learning is beginning to blur the boundary between risk management and fraud, but data availability remains a hurdle, says risk provider Provenir’s latest AI survey.

AI investments in finance continue to gather steam in 2022, as more firms win the management buy in to automate processes with machine learning, implement near-autonomous trading algorithms, and deploy predictive analytics on the end-user side. Specifically, this is fast becoming the year that automation and AI will finally make its way into risk decisioning, among other areas of the front and middle office.

Beyond fundamental automation of back-office processes and administration, machine learning is more and more making its way into the front and middle office through predictive analytics, thanks to the incorporation of alternative data.

Barriers to entry

The availability and quality of the data remains a problem in the industry, as 48% of survey respondents indicated that they plan to make further investments here in 2022, compared with 35% currently have the capability or find it to be vitally important.

As with other applications of AI and automation, however, data quality frequently presents a hurdle to getting a return on the initial investment.

“In the survey, we saw that only 21percent of respondents begin to see returns from their AI projects within the first 120 days using AI in risk decisioning,” Carol Hamilton, SVP of global solutions, comments for Bob’s Guide.

“This indicates that people are seeing the value when they want to embark on these projects, but they’re not actually being able to operationalize it.”

Key factors for success

Hamilton believes that three main factors need to be present for AI to deliver a return on investment.
These include access to relevant data, availability of a distributed AI development platform to build best in class models, and a deployment platform to implement, monitor and track models in production.

“Organisations that fail to understand the importance of this combination of factors will continue to struggle with the success of AI initiatives,” she adds.

According to the survey, less than 5% of respondents saw a return on their AI investment within 60 days, with most (59%) spending between three and four months improving processes and technology before they saw increased operational efficiencies.

As the industry matures, decision makers are focusing more on validating the processes and results they’ve generated through AI. Nowhere is this topic more relevant than in risk management, where automation has only recently started making a dent.

“It’s been really fascinating to see that less than 20 percent of people have found the accuracy they want in their credit risk models. Access to reliable data, as well as the complexity and cost of writing good models and operationalizing them play a huge part in this lack of confidence.” Hamilton says.

The push to get better data and improve credit risk models is apparent in the survey responses, however, as 74% of respondents plan to invest in a real time credit risk and decisioning platform, versus just 24% who do not.

The survey also indicates that the vast majority of financial firms have dedicated investment to more than one area of AI development.

71% plan to invest in a digital wallet solution, 63% plan to invest in automation and machine learning for products that enable inclusion, and 55% have set aside budget for embedded finance technologies.

AI and alt data blur the boundary between fraud prevention

Fraud prevention is the strongest factor motivating firms to automate, according to the Provenir, survey, with 76% citing it as the most important driver for investment. This is followed closely by a need to automate decisions across the credit lifecycle (58%), pushing for cost savings and efficiency (57%), more competitive pricing (51%) and more accurate risk profiles (48%).

According to Hamilton, this marks a shift in the way risk management teams have historically been structured. Automation and AI are allowing for a more comprehensive view of risk, whereby fraud prevention is increasingly considered a part of the risk function.

“Fraud prevention as a risk driver is a bit of a move from my perspective. I’ve spent nearly 20 years in fraud and risk, and they were historically so separate,” Hamilton explains.

Fintechs lead the way

It is clear, according to Hamilton, that even large and, historically siloed banking institutions are increasing the interaction between the risk and fraud prevention functions. However, this is ultimately an area where fast-moving fintechs have an edge.

“With Tier 1 banks it has largely remained so. But even across the big banks where you have very separate departments, there’s increased collaboration, increased awareness that actually you’re leveraging the same sort of data, just perhaps in slightly different ways. Meanwhile, for the fintechs we work with, fraud is really blurring into credit risk.”

According to Hamilton, this blurring of fraud into risk management has been a key driver for the incorporation of more alternative data sources and vice versa.

“We’re almost blurring the lines and that’s reflected in these answers. Previously people might have seen it as a hurdle to use specific credit worthiness data, she states.

“Different types of data can be used now. This includes what people might think as fraud data, but it can be used in the same sort of way to make that initial risk decision. This is especially true for fintechs, where risk is risk and brings together these quite separate worlds.”

• Read more under:
• BankingTech
• Compliance Management