Eimantas Palionis. Artificial intelligence will decide whether you will get a loan – why is this good news? | Business

From the approximately 180 variables that are provided to DI, it selects, based on historical data, a combination of up to 10 key criteria that best reflect the ability of customers to repay the loan over a certain period of time.

In order to ensure that the selected criteria meet regulatory requirements and are not discriminatory (for example, they are not judged solely on the basis of age or belonging to one or another gender), they are reviewed and, if necessary, adjusted by experts from financial institutions.

AI training is a never-ending process

The main criterion by which a customer’s solvency is assessed is an indicator of how he has managed to fulfill his financial obligations in the past. Failure to repay loans on time in the past is a strong warning sign that this may happen again in the future. Other important criteria are the individual’s income, length of service, the sector in which the client works, pre-existing liabilities and the ratio of income to liabilities.

In order to “teach” a machine learning model to accurately and correctly determine the probability of insolvency, the data provided must be logical and correct, without any exceptions or missing data that could mislead the model. If DI made decisions today by evaluating the macroeconomic situation of 5 years ago, it would be a failure, because in five years the macroeconomic situation in Lithuania has changed very strongly.

Models must be continuously maintained, monitored and adjusted to make decisions based on today, even though historical information was used to develop the model. This is a long and complex job that requires advanced mathematics, macroeconomics and programming knowledge.

Advantages of AI vs Human

First, speed. A person can take a dozen minutes or an hour to assess the solvency of one customer. Artificial intelligence can efficiently and accurately evaluate huge amounts of data in a matter of seconds, so the customer receives an answer in essentially real time.

Secondly, if everything is tested and it is confirmed that the model works correctly, we can be sure that it will not go wrong. A person is a person – when assessing solvency, they may have some subjective opinions about the client’s personality or lifestyle, or personal connections with the assessee. Models have no settings. Their assessment is based only on data and is therefore objective, fair and eliminates the risk of discrimination.

Can AI be tricked?

Financial institutions use machine learning not only to predict the client’s ability to repay loans and to calculate the maximum amount that can be borrowed. The main goal is to assess a person’s financial situation and reduce their potential financial burden. Yes, assessing the client’s solvency protects the financial institution from a possible loss, but the most important thing is that the client himself is protected from a loan that he would not be able to repay.

Attempts to provide false information are considered fraud and/or forgery, which incurs legal liability. Therefore, in order to improve your credit rating, it is recommended to borrow responsibly and not to make reckless financial decisions that would prevent you from fulfilling your obligations on time.