Thanks to a self-learning system based on machine learning algorithms to obtain consumer credit in the Bank can be three times faster
One of the largest banks in the country – b & n found a way three times to reduce the maximum time for making decisions about granting consumer loans. It is planned that in 2018, this procedure will take more three days, and only one. Moreover, for most standard applications, customers will be able to know the answer of Bank in just a few minutes. This saving time Bank sought to achieve by changing the process of decision-making on self-learning model based on machine learning algorithms, the press service of the credit institution.
The Bank noted that the high speed of decision-making will be combined with quality and effective assessment of solvency. Potential borrowers will be evaluated using the latest risk management tools. With this level of credit losses in the Bank’s portfolio will remain at a record low and the Bank optimizes operational costs.
Industry experts highly appreciate the prospects of the use of artificial intelligence in banking.
“The work on these algorithms can lead to significant reduction of the time of the decision on the loan, because this algorithm scans the defined parameters of the customer and on their basis makes the analysis of how high the level of creditworthiness of the potential borrower, – said a senior analyst “Freedomfinance” Bogdan Zvarich. – The algorithm can make a decision on the current request for the loan, based on the statistics for similar customers and seeing customers with the same parameters return the loans. As for the banks, it is actually a translation scoring model for a learning algorithm that analyze current customer base, the database of debtors, how they behave in the light of payments on its obligations.”
In fact, the Bank noted that the developed model for assessing the solvency of customers will allow us to explore the widest possible pool of information from external and internal sources. “The number of variables in the model are much greater than in classical methods of scoring, and the model smoooches, – says head of scoring models and sources of data of the Bank Dmitry Gerasimov. – Preliminary results show that we can increase the Gini coefficient (a universal scoring metric to evaluate the quality) to 1.5 times compared with the figure for the classical methods. In the end the Bank will get unprecedented improvement in the quality of the segmentation of customers according to level of risk, while we will retain the ability to conduct manual verification point in cases where it is really necessary”.
It should be noted that PJSC “Bank” one of the first on the market introducing machine learning technology (artificial intelligence technique) – so, this applies, for example, work with overdue debt in the retail business. The translation of the whole cycle of development and implementation of models in machine learning will enable the Bank to 70% of cases avoiding calls to clients in the early stages of arrears without losing efficiency.