From Data to Decision: Understanding Credit Decision Engine Algorithms

Credit Decision Engine is the heartbeat of any bank or financial institution. Data is the blood that flows through with each heartbeat. In today’s banking and lending environment, credit decision engines are expected to intake thousands of attributes at the same time and make decisions within a fraction of a second.

Credit Decision Engine: Identity Data Integration

Whether it’s identity verification data, banking transaction information or credit data, credit decisions of today are made in a multivariate fashion. An experienced bank will use application data, credit information, identity verification data and render a comprehensive decision. 

Identity verification data is probably the first set of third party data a bank will encounter to fulfill their underwriting needs. Not only KYC or Know Your Customer is required by law, it is a good practice in today’s digital banking environment where the end customers may never meet a banker face to face. 

Identity data bureaus such as Sentilink and Socure can serve a variety of datasets through their API. Your decision engine should already have these identity bureaus connected and ready to be used. Most of these identity bureaus have hundreds and if not thousands of identity attributes as well as canned scores that predict the likelihood of identity fraud. 

Some of these identity verification vendors also provide specific scores and attributes to detect first party fraud as well as third party fraud or identity theft. All of these dimensions of identity verification should be passed into the decision engine and its variables and scores made available for risk management professionals to write underwriting rules.

Credit Decision Engine: Banking Data Integration

Many of today’s lenders are looking at alternative ways to underwrite a client or a small business. With the advent of banking transactions data providers such as Plaid, lenders are using account balances, bank account transaction, payments information to make an informed decision to make a quick credit decision.

Credit decision engines should be able to work with user consent based banking transaction data assets such as Plaid or Flinks. These datasets sometimes contain 30 to 120 days of transaction history. Often, these transactions are categorized and ready for risk management experts to write rules.

Here are some underwriting examples using bank data. If the average bank balance in the past 3 months is greater than $2,000, this client might have enough funds to carry a $10,000 loan amortized over 5 years with a monthly payment of $200 at 7.5% APR. In other words, she can spare $200 to service a $10,000 loan if she so desires to take out that loan for a medical procedure or a home improvement project. In these cases, the lender may not care too much about her credit score or credit history. Her income is sufficient to satisfy the lender’s underwriting criteria and approve this person for the loan.

Credit decision engine is expected to not only ingest banking transaction data, it is also expected to enhance the raw data and produce a variety of derived variables or summarized variables for the risk management professional to write rules.

Credit Decision Engine: Credit Behavior Data Integration

In the United States, there are 3 major credit bureaus such as TransUnion, Equifax and Experian. These credit bureaus serve an important role in the overall US economy. These gigantic databases contain behavioral information of average Americans on their ability to repay their debt. These credit bureaus contain all of our credit card balances, mortgage repayment history and student loan debt information.

Positive and negative behaviors are all reported to these credit bureaus and they work with credit scoring companies such as Fair Isaac or credit score consortiums such as Vantage to calculate a credit score for just about all working adults on the probability of defaulting on a credit card in the next 90 days.

These credit bureaus also have specific credit scores predicting the likelihood of defaulting on different credit products under a variety of different situations.

Decision engines used by banks and financial institutions are expected to connect to all major credit bureaus and make thousands of credit attributes available to risk managers to write credit underwriting rules. For example if someone is applying for a credit card and their FICO v9 score is less than 500, she might either get declined or offer a starter credit card with a low balance of $500. These types of credit underwriting rules and product offering rules are all expected to be handled by a modern credit decision engine.

Credit Decision Engine: Custom Credit Scores

Just about every sophisticated lender has built their own credit risk score based on the client's true behavior with respect to the lender’s financial product and their servicing tactics.

These custom credit scores are often written in SAS, R, Python and they are often written in a statistical language that requires translation into other forms of computer language. A robust credit decision engine should be able to take python code directly from a risk management group and implement the python code right into the credit decision engine and calculate the custom credit score on the fly and use them to execute additional credit underwriting rules.

These custom credit scores must also produce statistically sound adverse action codes. These codes signify the most important attributes that led to a credit decision. If the score is used and the client is declined, it is required by law that the lender through its credit decision engine to produce NOAA or Notice of Adverse Action code and letters to fully inform consumers on exactly why they have been denied credit. 

It is the credit decision engine’s job to calculate the custom credit score as well as work with risk management professionals to produce adverse action codes per Fair Credit Reporting Act, in the United States.

Credit Decision Engine: Core Banking, Loan Management and Credit Card Management Systems

After a bank account, credit card or an installment loan is approved, the journey doesn’t end there. A decision engine is also expected to complete a feedback loop from the downstream servicing platforms such a bank core or a loan management system or a credit card management platform.

Sometimes, banks and financial institutions want to offer a different product or increase the customer’s line of credit or make a modification to their existing loan. The post originations payment behavior is expected to be fed back into the decision engine.  Combined with the original originations data, a credit decision engine could be making a subsequent decision, say automatically approving a line increase for a client that exhibited good behavior and her credit card limit might be increased from $500 to $1,500.

Credit Decision Engine: Building and testing

Finally, you can take a look at LendAPI.com’s free decision engine offering. We are offering a fully integrated (from application to bank core) system where the credit decision engine is at it's core. We welcome any feedback at info@lendapi.com and please goto www.lendapi.com for more.