• postmaster167

Credit decision based on AI algorithms



Traditionally loan applications were processed by humans, with a lot of paper job and often with multilevel decision making process. Such process was administrative cost demanded. And these costs multiplied in parallel with multiplication of customers.

Development of artificial intelligence (AI) has open new ways for lending process and financial institutions took advantage of this opportunity.

Primary benefits of the application of AI systems in the financial industry are the following:

· automation of routine processes;

· increased speed of service;

· reduction of the costs for solving standard tasks;

· enhanced accuracy of processing of large amounts of data;

· improved quality of client support system.

Actually, AI and machine learning can be used by financial institutions across the all customer life cycle.

On the stage of customer acquisition AI can support marketing channel mapping, personalization of offers, retargeting, KYC procedures.

In credit decisioning AI helps to qualify credit, assess credit limit and optimize pricing, to prevent frauds.

On monitoring and collection stage AI can generate early-warning signals, define probability of default, provide value at risk customer segmentation.


Making decision with AI

As world becomes increasingly digital, financial institutions looks to computers to help make credit decisions based on different source of data. Beyond just analyzing data, some digital lenders employ artificial intelligence and machine-learning (ML) algorithms to make faster, more efficient credit decisions.

Modern AI/ML models can qualify new customers for credit services, determine loan limits and pricing and reduce the risk of fraud.

Qualification of new customers can be done for a particular type of loan. On early stage of AI/ML models development customers were analyzed basing on rule-based or logistic-regression models which worked with credit bureau reports. Such approach failed to serve potential clients without formal credit history. That’s why recent year financial institutions pay attention to complex models for analyzing structured and unstructured data, loading into them hundreds of data types from social media, mobile phones, social networks etc.

Different customers can be able to repay different loan amount. That’s why credit decision closely related to a credit limit. It pushed lenders to use AI/ML models also to automate the process for determining the maximum amount that customer can borrow. For such models data scientists take data from conventional data sources like bank statements, tax payments, utilities invoices. As a result, models can quickly assess a client’s income and capacity to make regular loan repayments and generate a highly precise prediction of loan repayment.

Client’s risk level in lending traditionally correlates with interest rate level. Next logical step was to use AI/ML models also for pricing purposes. And market leaders in this approach have received a considerable advantage against traditionalists, especially in competition for borrowers with a strong risk score. They have been able to offer competitive rates while keeping their risk cost low and optimize the balance of total asset volume, risk, and interest income within a lending portfolio.

Automated loan decisioning not only opens new sales opportunities for financial institutions but also opens new opportunities for fraud. On-line lenders face with following typical types of fraud: identity fraud, employee fraud, partner fraud, customer fraud and payment fraud. To mitigate fraud risk, lenders make their models more and more complex using, for example, face recognition technologies, behavioral models etc.

By eliminating human prejudice, incorporating alternative data, and training the models in real-time, AI modelers say they're able to make more accurate predictions of a consumers' creditworthiness.


AI can help, but there are trade-offs

For lenders who are going to use AI-based lending, there are two options to build a model: scrub historical data or design the model from scratch.

First option is easier for lenders with long history of a market activity because they can use own data if they store them properly. But this option has downsides. For example, if lender’s decision makers in the past had bias to any category of borrowers. In this case AI model will be trained on the data there such category of borrowers were marked as unwanted.

The second option, focused on the model itself rather than the data, is costly.

For AI models it is also important to keep balance between risk and profitability. It means that there is no sense to create risk free model except for ability of a model to identify with 100% probability as performed as non-performed borrowers.

Compassway has own expertise in development of automated credit decisionig systems. We can propose our client as ready-to-use AI/ML model for lending process or develop unique model taking into account client’s possibility to gather customer’s data on the local market or stored client’s datasets.

Risk model can be delivered by us as a part of credit system or separately. In this case our specialists help you to embed risk model in credit process.

4 views0 comments