Partnering with Google Cloud to empower borrowers with bias risk detection

In its first 100 days, the Biden Administration has made uncovering and eliminating bias across federal agencies a top priority. Executive order 13985 directs all departments to evaluate bias and return recommendations to reduce its impact. Bias is often hidden in the data that the federal government uses to make decisions and deliver critical services across the nation. While agencies begin the work of recognizing and reporting potential bias, Data Society, alongside Google Cloud developers and Brllnt designers, is already supporting the CFPB (Consumer Financial Protection Bureau) in that effort. The team developed “Loans Like Me,” a user-friendly CFBP tool to empower regulators and consumers with clear data. Loans Like Me enables the CFPB to improve fair lending practices and expand credit to those negatively affected by data bias.

User-Friendly Data

“Loans Like Me” is an AI-driven web app that democratizes HMDA (Home Mortgage Disclosure Act) data, enabling consumers to compare quoted interest rates and mortgage details against similar loans and detect the probability (low, medium, or high) of bias. Directly from the app, the user can report possible bias to the CFPB and request assistance to find a more fair rate. The app’s dashboard also informs the consumer of factors that could influence interest rates, including key details from comparable loans and historical bias in the data (e.g. institution’s past lending patterns, or geographic trends, affecting individuals that objectively ‘look like them’). With easy access to this information, consumers have a firm foundation to advocate on their own behalf and find lenders with less propensity towards possible bias.

For regulators, Loans Like Me overlays incoming real-time data from consumers over a map of historic credit deserts and automatically flags financial institutions within and across geographic regions as automated calculations and comparisons reveal biased/discriminatory practices. With this tool, regulators are able to readily identify emerging or enduring bias, and target remedial actions in short order, rather than awaiting annual submissions to identify issues retroactively.

Innovative AI

Our solution is predicated on the use of four pillars of information and incorporating artificial intelligence to create a more robust, and transparent solution for lending discrimination analysis that has the granularity to facilitate consumers and regulators across the agency. 

Our approach deconstructs a complex problem into its smallest building blocks and then addresses each through means of analysis of the following types of data sources:

 

  • HMDA public data
  • Census public data
  • Social media / news sources  to determine soft indicators/reputation
  • Private Data pipelined from CFPB

 

We built an algorithm that clusters similar consumers based first on financial factors with the greatest impact on potential interest rates and compares them to peers within a subgroup using “important features” to predicts the risk of discrimination. Finally, with reputation, soft score (externally derived) and HMDA LAR data incorporated the application can forecast discrimination risk of an institution and county, state, and national level trends.

The Impact

In the hands of the consumer, this user-friendly platform powers the self-efficacy that brings equity closer to the norm, one deal at a time.  For the CFPB, moving from annual bank submissions to real-time consumer reports means understanding changing interest rates as they’re quoted, informing agile and relevant decisions. Long-term, as an agency, such valuable tools help the CFPB strengthen their position as a key consumer resource for fairness in lending and beyond.

Data Society leads government agencies and global enterprises alike to understand, and then unleash, the unlimited potential of their data-driven workforce.

At Graphs in Government on April 28th, Nisha Iyer, VP of Data Science and Engineering, spoke to the power of graph databases, specifically knowledge graphs, and how we use them on projects like the CFPB tech sprint to highlight areas that are usually very hard to connect – both schematically and ontologically – and pull information from these sources to help a specific population understand relevant and impactful data.