Akbank Uses FICO Prescriptive Analytics to Grow Credit Card Approvals 45% And Limits By 60%

Leading Turkish retail bank wins FICO Decisions Award for AI, machine learning & optimization using FICO decision optimization technology

Akbank, one of the largest retail banks in Turkey, has improved the way it makes offers to consumers for new credit cards and credit limit increases by leveraging mathematical optimization and action/effect modelling from global analytics leader FICO. The use of FICO prescriptive analytics allows the bank to comply with rigorous national laws around credit offers while growing credit card approvals by 45 percent and credit limits by 60 percent.

Latest Aithority Insights: Why Contextual Targeting Deserves Another Look with Artificial Intelligence (AI)

For its achievements, Akbank won a 2022 FICO Decisions Award for AI, machine learning & optimization.

“Growing market share and revenues in the credit card portfolio is dependent on the level of credit limits approved to customers,” said Serhan Pak, SVP of retail lending & advanced credit analytics at Akbank. “However, credit loss is also affected by the assigned limits. Therefore, limit assignment is an area that we believed that would benefit from optimization.”

The Turkish market regulator stipulates that the total limit of a consumer’s credit cards cannot exceed twice their monthly income for the first year and four times income after the first year. This rule meant that Akbank had to include existing credit cards from other banks into its optimization model.

Akbank’s profit model also had to consider the numerous options available to customers in the market, such as pay with instalment features, postponing payments or drawing cash advances with instalment repayment options. This meant that from a profit perspective, even though a credit card is a single product, it can behave like four different ones.

“The profit model of the product turned out to be very complex,” said Pak. “In addition to regulation and product considerations, we had to account for pandemic lockdowns, which impacted the application numbers, customer profiles and channel mix. Provisions were also made for the impact of rising inflation on income calculations and historic revenues.”

FICO Decision Optimizer was used to design strategies and set limits for Initial Credit Line (ICL), and Credit Line Increase (CLI) operations at Akbank. The project aimed to help the bank to understand and model the likely customer reactions to various offers and what the trade-offs would be when incremental changes were made to different business goals.

Applying action/effect modelling was a new methodology for Akbank. It led to a lift in decision quality by integrating customers’ reactions into the decision models, allowing for improved predictions.

“A key challenge was the legal caps enforced by the regulator,” said Pak. “There was a high risk of change in the Covid era that we needed to plan for. To overcome this, the team came up with the solution that combines both uncapped and capped flow in the same project, which has not been done before in Decision Optimizer.”

The legal caps on a consumer’s income to limit ratio are applied as a consideration, but not as a direct cap on the optimized limit, during the optimization process. This delivers greater flexibility and speed as the optimization results do not directly depend on these caps but takes into account their impact for when the strategy is deployed in the real world. Due to this process changes to regulation can be implemented and managed quickly and their impact assessed. This double structure allows Akbank to be more flexible to market changes compared to competitors that do not use optimized uncapped decision flows.

Top Artificial Intelligence Insights: Determining the Potential of Your AI Algorithm Starts with Measurement

The most important advantage of optimization was the ability to see trade-offs between different business goals. A framework was developed in Python to examine and visualize the results in a very detailed way, which allowed Akbank to compare different strategies. Using techniques such as machine learning, action/effect modelling, optimization and extensive use of tools such as Python, Akbank has created a strategy management approach that is more analytical, flexible and integrated.

“This was a conceptually hard problem,” said Graham Rand, operational researcher and editor of Impact and one of the FICO Decisions Awards judges. “The nature of what Akbank targeted and the fact that they had to account for competitor banks in Turkey, as well as optimizing for the capped and uncapped scenarios, demonstrates their decisioning sophistication.”

The credit card optimization project has been very successful for Akbank. The optimized strategies allowed the bank to comply with complex regulation while growing credit card approvals by 45 percent and approved limits by 60 percent. This was achieved while keeping credit losses the same. Akbank expects to realize a 129 percent increase in profit from its credit card portfolio using the solution.

“In today’s world, consumers have more options and the stakes are high when it comes to capturing greater market share,” said Nikhil Behl, chief marketing officer at FICO. “Akbank has demonstrated how a lender can build on its analytic capabilities with optimization to increase lending profit.”

AI and ML News: Why SMBs Shouldn’t Be Afraid of Artificial Intelligence (AI)

[To share your insights with us, please write to sghosh@martechseries.com]