Hakan Yilmaz | Head of Credit and Model Risk Management, Akbank
in non-performing loans while maintaining the same approval rate
in “no-credit history applicants” approval rate while maintaining the same risk appetite
Akbank becomes the first global lending institution fully powered by ML underwriting across all lines of business
Akbank is Turkey’s third largest private bank but leads the country’s financial sector in digital smarts. Akbank invested heavily in technology over the last few years to reach more new customers on the screens where they want to bank. Turkey has one of the youngest populations in the world. Half its residents are under 30 and, like all young people, they want to do everything on their smartphones. Trouble is, millions of them have little to no history in their credit file.
Serving this mobile, digital customer effectively presented a challenge for Akbank even before Turkey’s economy hit a major speed bump. New personal loan applicants were hardly matching the profiles of previous customers and the bank’s credit models were rejecting significant number of new applicants to the bank. Akbank needed to evolve its underwriting.
Executives at the bank decided to invest in machine learning, which applies sophisticated math to massive amounts of data to generate more predictive credit models. Akbank, like a lot of big banks, had a gold mine of information about its customers’ spending habits and finances. Even better, the bank had invested hundreds of millions of dollars to create a unified data center to do more effective underwriting.
Zest worked closely with Akbank’s credit and risk management teams to explore the data on hand and began training Akbank modelers in machine learning skills. The teams looked at 26 times more data than Akbank had used previously and ended up training ML models on triple the number of variables the bank had been using. The result was a handful of machine learning credit models that could replace the bank’s tens of linear regression models across five of the bank’s product lines. Each of Akbank’s traditional models had taken up to 6 months to create, validate and test before going into production. The team’s new goal is to shrink that time to under three months for ML underwriting.
Galip Berker, head of credit and model risk management at Akbank as responsible for credit models says “ ML experience with Zest is an exciting journey and both sides have a lot to learn and benefit from each other.”
“Our partnership with Zest has helped us unlock a huge amount of value from our investment in digital transformation,” says Hakan Yilmaz, head of retail credit and analytics at Akbank. “By switching over to ML for credit underwriting, Akbank can find more good borrowers with no added risk and react faster to changes in the market.”
Each model was tested against actual credit applications and transactions to see how much better they were at re-scoring good and bad borrowers across the risk spectrum. The results were impressive: The ML lending model will allow Akbank to slash non-performing loans by 45% across the overall portfolio while maintaining the same approval rates, including thin- and no-file applicants. If the ZAML model were applied to growth, Akbank could double its approval rate for “no-credit history” borrowers who may or may not be Akbank customers. The bottom line: Using ZAML, Akbank could confidently increase its lending volume by more than $300 million per year.
The Zest-Akbank partnership includes a strong skill-building component in which Zest experts train and certify Akbank’s modeling team in machine learning. By the end of 2019, ZAML will be in all the bank’s retail divisions and Akbank will be the first global lending institution fully powered by ML in underwriting across its line of business.
“Digital banking is core to Akbank’s lead in the market. AI-powered underwriting will help support our profitable growth for years to come,” says Akbank’s Hakan Yilmaz. “Zest machine learning was crucial in maximizing the value of our investment.”