Playing the Long Game with Big Data and Artificial Intelligence
An Interview with Michael Bradley
We talked with Michael Bradley about the impact of big data and artificial intelligence on loan production and its implications for loan officers, processors and underwriters. Michael, in his role as senior vice president of modeling and analytics strategy for Single-Family at Freddie Mac, closely studies the evolution of this technology and its deployment in the mortgage industry.
Q: Can You Define Big Data Within the Context of Mortgage Lending?
A: There’s no commonly accepted definition of big data among technologists and within financial services. Most experts characterize it by the “four Vs”—massive in volume; extremely varied in format; intense in velocity; and challenging in rating its veracity.
Examples of big data in mortgage lending include detailed credit information, information applicants give us permission to access, e.g., bank statements, and text and photos from MLS and appraisal reports.
Q: What is Data Engineering and Data Science all About?
A: Data engineers are the experts that capture, store, read, and make searchable these vast and varied data sets. Machine learning, a subset of artificial intelligence, allows data scientists to convert these data to produce insights that create value, new products, and services.
Freddie Mac is harnessing big data and artificial intelligence to drive significant value for our clients, their borrowers and the industry. There’s no shortage of data at Freddie Mac and we’re putting it all to work to build solutions that help us mitigate risk and that address customer pain points or real business opportunities.
In short, we’re Reimagining the Mortgage Experience® by bringing big data and machine learning to bear at almost every aspect of the mortgage process. Big data and machine learning factor into nearly every model Freddie Mac builds today.
Q: Can You Explain the Link between Big Data and Machine Learning?
A: Look at big data as the premium, high-grade gasoline and machine learning as the engine that runs on it. A regular car takes 87 to fuel a standard engine. Big data is premium, higher-grade fuel powering a high-compression engine in a car that outperforms and better handles the road compared to regular ones.
Machine learning allows us to make sense of these data. It allows us to discover patterns and correlations in huge data that offer us novel and invaluable insights. But don’t get confused, the real revolution taking place comes from the sheer volume of data itself, not from enablers such as machine learning and cloud computing.
Q: Why Do Lenders Care About Big Data and Machine Learning?
A: Big Data is the game changer in financial services today, especially in the mortgage business. To understand the topography of this new tech landscape, you need to realize that the implications of big data permeate the entire lending process. That means knowing its impact at each stage of the mortgage cycle.
Q: Why the Focus on Big Data in Terms of Origination and Underwriting Risk?
A: In this business, we’re not talking about lenders buying and storing consumers’ social-media posts or click-and-buy patterns. Most players are sourcing official financial data from third-party suppliers—e.g., credit bureaus, banks, employers, and government agencies. It flows through direct feeds into a lender’s systems and is continually refreshed to reflect the most recent status of a borrower’s financial condition.
Soon, however, big data might extend beyond standard borrower records to cash-management metrics like cell-phone payments, payments for utilities and checking account activity. This kind of information can give lenders a more comprehensive picture of a borrower’s ability to repay a loan.
Borrower information pulled from primary sources and supplied to lenders by third-party vendors speeds up loan production time to such a degree that it materially decreases a lender’s cost and the time to produce a loan, improves the borrower’s experience and decreases the likelihood of fraud. Borrowers don’t have to pull the information together, and lenders’ employees don’t have to extract it from documents and plug it into their systems—a manually intensive, error-prone process.
Q: How Will Big Data and Machine Learning Change the Way Lenders Generate Loans?
A: Technology advancements have been transforming mortgage lending jobs for the last two decades, starting with automated underwriting in the ‘90s and progressing to workflow automation tools following the financial crisis. Now we’re in the next big stage of innovation, and its effects on the lending process will dwarf that of past tech advancements.
What’s the Main Takeaway for Loan Officers?
Big data technology will save loan officers an inordinate amount of time and give them more control over the integrity of information on applicants.
With potential borrower information fed directly into a lender’s origination systems, loan officers are freed up to do what machines can’t: win the confidence of customers and cultivate relationships with new ones. On an operational level, they’ll oversee the process rather than actively move customers’ loans through the pipeline. They can also produce loans faster allowing them to close more loans.
How Do Processors Gain an Edge?
Like loan officers, processors are no longer bogged down with so much “stare and compare” work that’s easy to stumble on and best done electronically. They’ll step in when a model’s prediction doesn’t pass the smell test, based on experience and professional judgement.
What are the Biggest Benefits for Underwriters?
With big data and machine-learning analytics, underwriters are better positioned to understand and evaluate risk, and they don’t have to parse loan files to double check the data or crunch the numbers themselves—both of which are a major drag on productivity.
However, for underwriters, it’s not just about turning around more applications on much shorter timeline. Machine learning models can empower underwriters to approve mortgages for customers who haven’t borrowed enough to be judged on a generic credit score or those that have income from the gig economy—an increasingly bigger segment of the population.
Q: How Does Big Data and Machine Learning Take Automation to a New Level?
A: These technologies digitize the lending ecosystem versus automating only some steps of the process. Machine learning algorithms can electronically map borrower information with a system’s data fields and does the necessary calculations.
Machine learning can also determine if a loan can prudently bypass certain steps. For example, Freddie Mac’s Automated Collateral Evaluation (ACE) can determine if a borrower can save time and money with an appraisal waiver on mortgages that are eligible for sale to Freddie Mac. This way, the lender doesn’t have to worry about ordering an appraisal or assessing the appraisal report.
Q: How Will Lenders Address Compliance Challenges Tied to this Technology?
A: All data has biases that can lead to unintended outcomes. Lenders always compliance-test all asset and income verification solutions and underwriting models to ensure results adhere to fair lending regulations. Lenders will have to guard against privacy violations when leveraging a vastly larger, ever-changing amount of borrower data. The mortgage industry is inherently cautious, as it must be, and that’ll serve it well as lenders scale the learning curve of these new technologies.
Q: What’s Your Sense of Mortgage Technology’s Longer-Term Impact on the Industry’s Workforce Prospects?
A: Most people today shop online—even if just to compare prices—whether they need a mortgage or a dishwasher. To meet consumer expectations, lenders must sharply cut loan turnaround times and find ways to prudently expand the credit box with a more comprehensive picture of a borrower’s risk profile. Machines—more specifically big data-driven analytics—are better than people at making predictions and do it a fraction of the time.
Better predictions can help people make better business decisions. Richer data, stronger models and more powerful computers drive progress in machine learning-driven prediction. In economic terms, data becomes more valuable as prediction becomes cheaper.
New AI technologies make prediction cheap! As the history of economics tells us, we are not only going to start using a lot more prediction, but also to see it emerge in surprising new places. The drop in the cost of prediction will impact value of other things, increasing the value of complements (data, judgement, and action) and diminishing value of substitutes (human prediction). Adam Smith’s economic thinking on the division of labor teaches us to allocate roles based on relative strengths (comparative advantage). In generating predictions, the division of labor is between humans and machines.
The major benefit of prediction machines is that they can scale in ways humans cannot. One downside is that they struggle to make predictions in unusual or unprecedented cases. Prediction machines learn when data is plentiful, which happens with more routine or frequent scenarios. However, their value is limited during rare events. In such cases, human judgment is critical.
Going forward, there is room for both people and machines if people focus on tasks where they have a comparative advantage – this includes building borrower relationships and client trust, protecting against fraud and checking for data integrity.
Learn more about achieving greater efficiency and improving borrower experiences with end-to-end solutions that will give your business an edge.