Risk Management and Technology Services

AI/ML Credit Risk Scoring

LendAPI's credit risk engine hosts credit risk rules, custom credit scores, product and pricing startegies for banks and financial services organizations

LendAPI's credit risk engine hosts credit risk rules, custom credit scores, product and pricing startegies for banks and financial services organizations

LendAPI's credit risk engine hosts credit risk rules, custom credit scores, product and pricing startegies for banks and financial services organizations

LendAPI AI/ML Credit Scoring Services

LendAPI AI/ML Credit Risk Scoring Services

What is AI/ML Credit Risk Scoring?

In every bank or financial institution, credit risk management is paramount to the success of their lending business. Credit risk scoring is one of the most commonly used tools to approve an applicant for a line of credit.

Credit risk scoring is also an important tool in the overall credit risk management practice. These credit risk scores are used to measure the likelihood of a person defaulting on their credit obligations. 

Most common types of credit risk scores measure the likelihood of a person defaulting on their credit card revolving payments in the following 90 days. There are specialized credit scoring for various types of credit offerings such as installment loans, point of sale purchase, auto loans, mortgages and BNPL or Buy Now Pay Later types of credit offerings.

In the past decade or so, ML or Machine Learning and A.I. Artificial Intelligence modeling techniques have emerged and are deemed to be regulatory compliant. These models almost always outperform traditional modeling techniques. 

However, there are some hurdles with Machine Learn and Artificial Intelligence based models that we must consider and address. Before we get there, let’s go through some the necessary steps to get you started with building a custom risk model.

Who provides credit risk scores?

Commonly speaking, credit bureaus provide credit risk scores. They work in conjunction with the likes of Fair, Isaac and Company to create an universal score that measure’s the client’s likelihood of default. FICO scores as it's commonly known are widely used by private and government agencies to measure the default risk of a potential borrower.

Major credit bureaus have recently gotten together and created their consortium score known as Vantage Score which competes directly with FICO score to measure even more consumers that are new to credit with less tradelines reports to the credit bureaus.

Lately, large credit bureaus are combining credit attributes as well as banking activities to create VantageScore Plus which can increase the credibility of a potential borrower. However, it requires the borrower to log into their bank account and allow VantageScore to download all of their banking transactions to be included in the scoring process.

How much does credit risk scores cost?

Credit risk scores cost anywhere from $0.50 to $15 or more depending on the type of financing activity. If the borrower is considering a mortgage, the mortgage companies might request your credit score from all major credit bureaus. Sometimes they will pick the best score from the bunch and use that to provide their approval decision. In this case, the mortgage company might be paying $10 to $15 for credit services. However, given the size of the financing product such as a mortgage, $10 to $15 worth of credit report is not cost prohibitive.

If you are a non-bank lender and providing small dollar credit to consumers such as a point of sale finance product or a personal loan, the cost of credit score or credit pull might be on the cheaper side to make it affordable for the lenders to issue small dollar finance products and keep their overall cost of operations low.

Can I create my own credit score?

It is almost always a good idea for a bank or a lender to create their own credit score. The reason is that the generalized credit scores provided by credit bureaus works off of a generalized population and its behavior.

Also, these credit scores are updated infrequently and don't consider the ebb and flow of your business and therefore less accurate than if you create your own credit risk score. 

When you create your own credit risk score, you are training the credit risk model to seek out defaults that’s aligned with your product offering, your customer servicing behavior and your collections and recovery strategies. In other words, having your own credit risk score is more accurate and more calibrated to your own business.

Who can build these credit scores?

LendAPI works with a variety of credit risk management consulting firms that can build credit risk models. Check out our credit and risk modeling marketplace for a group of firms that can help you build a credit risk score and credit risk strategy. https://www.lendapi.com/marketplace/risk-modelling

The cost of building your own credit risk score could vary from $50k to $250k or more depending on the complexity of your business and the number of model segments to consider. A model segment could be a distinct marketing channel or a specific product offering amongst various products you offer.

These risk modeling consulting firms might have statisticians with specific skill sets to help you collect data, define modeling targets (defaults, conversions, charge offs) and select variables that will eventually go into your own custom credit risk models.

What types of information do I need to build these credit risk models?

First, you must have a good amount of applications or loans you’ve already booked. And secondly these loans must have been aged enough to have payment level performance.

For example, if you are running a point of sale financing company and you have a few hundreds loans and the average life of the loan is about 12 months on a 36 months term and if you have a good amount of defaults (say, 10%), then you might have enough data to build a credit risk model.

Another thing to look for is that the original credit report pulled at the time of application should be preserved. The credit risk modeling teams will need that information to train your custom model by looking at the future defaults.

One more point to watch out for is the level of granularity on your payment records. Some of the old antiquated loan management systems only record the last status of the loan and do not remember each install payment’s behavior. This is a bad practice and could create a noisy target when training your custom risk model.

We recommend that you work with a modern LMS or Bank Core and have every payment’s behavior history memorialized. This will help you greatly not only with building your own credit risk models, but your creditors and regulators will thank you as well. If you are interested in looking at some of our LMS/BankCore partners, please visit this link: https://www.lendapi.com/marketplace/lms-and-bank-core

How do I use a custom risk model?

If you are already a LendAPI customer, you are in luck. We have a state of the art decision engine that connects to a variety of third party data providers as well as a credit risk engine that can run rules and credit risk models.

We’ve already implemented a few custom credit risk models that you can use today to buy leads or score applicants. These scores are currently built on specific sub-prime credit bureau attributes. If you are interested in talking to us about these ready made scores, please email us at info@lendapi.com

If you aren’t not a LendAPI customer, that’s alright as well. Our decision engine is a stand alone tool in which we can install your custom risk model and make it available through an API for your loan originations systems to call and retrieve the scores and attributes.

Some of the risk management teams code in SAS, R or Python, we can take these custom risk model scoring languages and translate them into native programming languages and begin quality assurance testing.

LendAPI will also provide the necessary computing infrastructure to make sure the performance of the decision engine and the custom risk model’s API response time is within tolerance.

What are the most common compliance and technical hurdles of a machine learning risk models implementation?

In the financial services industry, your ability to explain how your model works is not only necessary for your own business but by laws and various financial regulations. 

Various banking regulators such as OCC or Office of Currency and Comptroller have a variety of publications regulating how banks and financial services work when it comes to creating and using credit risk models.

Why are regulators so keen on regulating credit risk models? The regulators want to make sure that your credit risk model doesn’t not discriminate against protected classes such as race, gender, nationality, age etc.

Historically, there were practices such as “red lining” that banks used to intentionally discriminate against African Americans and denied them access to credit. With ECOA, Equal Credit Opportunity Act and FCRA, Fair Credit Reporting Act, the United States led the way to curb these activities by the banks and forced them to issue credit across the board and prohibit them from discriminating and denying credit based on the aforementioned protected classes of people.

One of the most important characteristics of a matured credit risk model is the ability to present accurate reasons for denial. These are commonly known as Adverse Action reasons. 

All major credit scores from credit bureaus provide these adverse action reasons or adverse action codes. 

Our readers might have received letters from the bank when they were denied credit and on these letters which is also known as Adverse Action Letters might say something like “We are unable to offer you credit for the following reasons: 1. Number of Inquiries in the last 90 days. 2. Most recent trade opened in the last 2 years or 3. Utilization of your existing credit card etc”

These reasons provided by the credit bureaus as part of the credit scoring process represents the most important attributes that influence the individual score. 

Traditionally, credit risk modeling techniques such as linear regression techniques work off of variables’ monotonicity, meaning that the variables have a linear relationship to the modeling target such as defaults or charge offs. For instance, a variable could be the number of inquiries in the last six months. As the value of this variable increases, the risk also increases based on analysis. This increase in the number of inquiries in the last six months and its correlation to the heightened risk is of a linear nature and monotonic (ever increasing or decreasing). Therefore this variable can be one of the candidates for an adverse action reason.

These linear relationships are incredibly difficult to explain in non-linear modeling techniques such as some of the more popular machine learning modeling techniques like gradient descent or gradient boosting methods.

However, in recent years, teams of researchers have developed methods to make sure that hundreds of variables that enter the machine learning modeling processes can be used and made available as adverse action reasons. For more information, please email info@lendapi.clom. In essence, make sure your machine learning models built by your team or a consulting firm can use these adverse action codes by various aforementioned regulations such as FCRA, EOCA, etc.

LendAPI welcomes all Credit Risk Modeling organizations

LendAPI is working with some of the best credit risk modeling organizations to help our clients to create credit risk strategies, custom credit risk scores as well as pricing strategies. 

We also welcome more firms that’s interested in joining our marketplace and help the larger ecosystem to produce the best conversion rate and at lower default rates.

Our decision engine can run complex credit underwriting rules, product strategies as well as deliver custom credit risk scores through our scoring engine. 

We would love to host your rules, strategies and custom models. Please write to us at info@lendapi.com and explore these services.

LendAPI Decision Engine main offerings

Implement your custom credit risk score inside of LendAPI’s decision rule builder
  • Implement your custom credit risk score inside of LendAPI’s decision rule builder

Implement product and pricing strategy and linking it to decision trees
  • Implement product and pricing strategy and linking it to decision trees

  • Access decision trees, credit risk models and product offerings via LendAPI’s Decision Engine API

About LendAPI

LendAPI is a plug and play digital onboarding platform with a fully featured product designer, rules builder and a workflow orchestration platform that helps banks to launch their products within hours. For demos please visit www.lendapi.com and our youtube channel. Follow us on Linkedin and X.