How to Reduce Bias in Consumer Lending

How to Reduce Bias in Consumer Lending Informed

To err is human. “But,” it is said, “to really mess up takes a computer!” Perhaps that quip is the origin behind the suspicion that Artificial Intelligence (AI) algorithms could add to unfairness in consumer lending. Properly used, however, AI can actually reduce bias.

 Bias in consumer lending occurs when a lender makes a decision related to a loan and denies credit or imposes “non-standard” terms for reasons other than the borrower’s creditworthiness. If you get a break in your interest rate on your car loan because the salesperson thinks you “look reliable,” that’s bias. If a lender denies your application for a home loan because you live in “the wrong area,” that’s bias too.

A “good” decision process eliminates all considerations of race, religion, marital status, gender/gender identity, age, or disability. Bias leads to less than optimum decision-making. Reduce bias and risk reduces, and profitability increases.[1]

Bias is not just about personal characteristics though. Lending also has bias toward regular W2’s versus the self-employed and gig economy workforce. On “paper” some occupations look better than others. A lot of that bias arises from what documents are provided with the loan application and how documents are processed. Improving the quality of the documents’ data is key to removing bias. And, Informed’s solutions extract more accurate data than human review!

It’s ironic – we’re all more biased than we realize. Researchers have identified as many as twenty types of bias that affect human decision-making. Seeking to avoid human bias is the reason for using AI to make consumer lending decisions. And yet, the Consumer Financial Protection Bureau (CFPB) seems to be suspicious about whether AI-generated credit scores comply with the Equal Credit Opportunity Act and the Fair Credit Reporting Act.

So, what’s the story? Does AI reduce bias in consumer lending, or are the results no better than are achieved without it? Solutions provided by Informed demonstrably remove some of the organic bias in lending processes.

The key to reducing the effect of bias is to understand the consumer AI process. There are three major elements:

  • the expansion of data available for decision-making
  • the models that detect relationships in data
  • the automation of decision-making based on model predictions of loan profitability. 

“Big Data” makes it possible to collect much more information, of types that weren’t available before. Traditional approval processes depend on the information on loan applications, credit bureau scores, and previously collected information. However, those sources are limited. Many loan applications contain fraudulent data. Credit bureaus build credit scores ignoring significant facts. Now, however, a tsunami of data is available to assess loan applicants – information from payment systems, social networks, web presence, and more. The question is, “what is the “right” data for decision making?” 

The availability of a broader spectrum of data cuts both ways. More data types support better decision-making by allowing correlations between creditworthiness and other factors. On the other hand, there’s a risk that some of those factors — like gender or name (which might hint at ethnicity) — ought not be considered in the lending process.

A bias-free dataset cleansed of all attributes that might allow biased decision-making. Recital 26 of the European General Data Protection Regulation (GDPR)[2] provides a helpful description of data that should be used in the consumer lending process, “data rendered anonymous in such a manner that the data subject is not or no longer identifiable.” Perfect anonymity might not always be possible – but good solutions seek to minimize the risk.

There’s plenty of supplemental data to improve the loan approval process. But the starting point is the data provided by the applicant. Informed’s use of AI, and ML allows us to understand the documents and extract and classify the underlying data into a deal jacket, greatly reducing the need for manual intervention and improving speed and accuracy. This provides a better customer experience and for auto loans, gets the dealer paid faster.

The second element in Consumer Lending AI is Machine Learning (ML). In ML, a model is created using a sample data set (training data). The model can then make usable predictions from a new dataset with similar characteristics to the training data.

The essential question answered is, “what attributes of the subject best predict an accurate outcome?” Or, for Consumer Lending, “how do we know how much we should lend to whom, at what interest rate, and what is the risk of default?”

AI in Consumer Lending can be a “win/win.” It can radically reduce bias and eliminate risk for borrowers and lenders alike. It can improve lending profitability and reduce the risk of loss. It can enhance borrowers’ ability to get appropriate loans and manageable terms and reduce the risk of default.

If you’re intrigued by the opportunities presented by AI in Consumer Lending, Let us know by dropping us a line at We’d love to have a conversation with you!

[1] Will Dobbie, A. L. (2021). Measuring Bias in Consumer Lending (Vols. The Review of Economic Studies, Volume 88, Issue 6). (N. Gennaioli, Ed.)

[2] GDPR Recital 26: “Not Applicable To Anonymous Data.” (2016). Retrieved from

author avatar
We are digital transformers of tedious manual tasks, bringing robotics process automation to financial services.

New: American Banker - Sharing information = best defense against AI fraud