Editor’s note: This feature originally appeared in the January issue of MReport.
No doubt you’ve heard and read plenty about how artificial intelligence, or AI, is quietly revolutionizing every aspect of our lives, from transportation to how we shop for goods and services. But these emerging technologies are having an equally profound impact on the financial services industry, and more specifically, the mortgage industry.
In fact, no industry stands to gain more from AI—and specifically, from machine learning, a subset of AI—than mortgage lenders. For years, our industry has been handcuffed to manual processes that have prevented lenders from achieving stronger loan quality and a better consumer experience for a reasonable cost. Yet, even as mortgage organizations begin to invest in these innovations, few understand just how truly revolutionary they are to the mortgage production process. Even fewer organizations may grasp how these emerging technologies are fueling a new phenomenon called Capture 2.0 technology—and how this development could forever alter our industry’s fate for the better.
What Machine Learning Is All About
The buzz on AI and machine learning has reached a fever pitch as new solutions emerge almost daily that pledge to streamline how mortgages are originated, underwritten, and sold in the secondary market. Yet, they are also among the least understood terms in our business today, which is preventing many organizations from recognizing their full potential.
To put it simply, AI is generally a catch-all term describing technology that can analyze data and identify patterns in that data to make decisions. AI is about applying knowledge to a specific task or range of tasks to find the best answer. Think of AI as a parent teaching their child how to
make their own decisions based on past experiences, logic, and cognitive reasoning. Machine learning, on the other hand, is a little more specific. It also involves data and pattern recognition, but it also enables systems to learn and improve as new information comes to light.
This is done through a combination of human instruction and self-learning algorithms that distinguish data patterns. With machine-learning technology, organizations can train their systems to analyze large quantities of data and essentially complete tasks on their own. To put it even more simply, AI is about mimicking human abilities and machine learning is about training systems how to learn and complete a task with accuracy.
Even if you currently do not use them, it’s not terribly hard to visualize the impact AI and machine learning tools could have on things such as making credit decisions and meeting the requirements of regulators and investors. Both processes, and indeed many others, involve massive amounts of data that are collected and shared throughout the mortgage process. When used effectively, AI and machine-learning tools can help lenders lay the groundwork in pursuit of the fully digital mortgages. In fact, in 2018, Fannie Mae found both technologies were gaining momentum within our industry, and that 63% of lenders were familiar with AI and machine-learning technology, and 27% of lenders were already deploying them. However, leveraging these tools effectively is where most lenders are falling short.
The Quest for Better Data and Lower Costs
The most important thing to understand about AI is that it is only as effective as the data that goes into it. The way data is currently collected in our industry is not only time consuming but expensive as well. This is where machine learning comes in. The vast majority of lenders rely on manual processes, in combination with some form of optical character recognition (OCR) technology, through which they are able to “grab” data from documents provided by borrowers in either paper or scanned electronic format.
The idea behind OCR technology is that it saves the time and money that lenders would otherwise spend by having their employees read documents and retype what they see into their system of record. Yet when “reading” loan documents, template-based OCR tools count on data being found in approximately the same location on every document—which almost never happens. Complicating matters are the wide variations of data patterns found in most loan documents.
As a result, these types of tools work best when identifying information on structured documents, leveraging templates or keyword search to find and extract information. For structured documents, such as the URLA or closing disclosures, this works well enough. This is not the case, however, if image quality is low or the document type has a high degree of variation. As well, many other types of unstructured loan documents represent a challenge and can make OCR an imperfect “data-picking” tool. Even in the best of cases, when reading borrower documents, OCR is only able to capture a portion of data, which is why lenders continue to spend heavily on human “checkers” to capture what OCR missed and validate information pulled from borrower documents. To be sure, OCR tools are valuable, but cannot get the job done alone.
Where Capture 2.0 Comes into Play
When combined with OCR technologies, machine learning tools enable lenders to accurately identify and classify more loan documents and extract more data, more accurately, from them. That’s because machine-learning tools can be “trained” to recognize patterns on both structured and unstructured documents. This blend of tools is generally referred to as Capture 2.0 technology, also referred to as “intelligent capture.” A great example is gift letters. With no standard format, OCR alone is likely to fail. Yet every gift letter has the same information—somebody is giving someone else money. By using machine learning, trained algorithms, and leveraging a large enough document sample size, Capture 2.0 technologies can classify gift letter documents and pull data from them with as much as 95% accuracy. For the vast majority of mortgage participants, the difference maker for Capture 2.0 will be partnering with a Regtech provider that has tested, scalable solutions that can be delivered in a cloud environment to ensure performance.
Such providers should also have access to a document training set that is large enough for machine-learning tools to maximize the document types that can be classified and the data that can be effectively extracted. Grasping the Potential With Capture 2.0 technology, lenders are not only better able to capture and interpret greater amounts of data and route it to automated business processes, but also to evaluate loan quality and identify defects throughout the origination process, inclusive of due diligence between buyers and sellers of loan assets and servicing rights.
These processes are being streamlined because of the accuracy with which machinelearning tools can identify and classify documents and extract data with higher quality. Using more verified, validated data to power loan file reviews enables audit staff to perform additional audits in less time and refocus their efforts on managing exceptions. In fact, I have personally witnessed loan auditors evaluating as many as 18 compliance reviews a day, compared to the industry average of about five or six. When leveraged appropriately, Capture 2.0 technology can also help lenders leverage larger sets of “purified” loan file data to analyze their origination practices and get a much better view of their lending patterns.
They can “slice and dice” large, high-quality datasets to gain new insights on their operations, implement process improvement plans and even root out possible issues related to their lending practices. Ultimately, this can help lenders find and fix any problems before they gain the interest of regulators. Capture 2.0 technology can also help lenders deal with what is perhaps their biggest challenge of all—controlling costs. As recently as Q1 2019, total loan production expenses were over $9,000, according to the Mortgage Bankers Association, which is the highest amount ever. For most lenders, staffing is a huge cost factor. By using Capture 2.0 technology, lenders can improve staff efficiency and streamline processes to stem the ever-increasing expense of manufacturing loans—while also creating a better borrower experience. The bottom line is that improving loan quality and achieving efficient loan production should not be a mystery that takes unnecessary time and effort to solve.
Today’s machine-learning tools are already helping lenders compensate for the limitations of OCR technologies and reduce human intervention in many aspects of the mortgage production chain. At the same time, today’s tools are improving lenders’ ability to meet constantly changing regulatory and investment requirements. The potential of Capture 2.0 technology doesn’t stop there. In the end, it will enable the automation of all kinds of decisions throughout the mortgage lifecycle, including the point-of-sale stage and servicing. It will help borrowers as they are applying for loans, and help lenders streamline underwriting decisions and even find out which borrowers might be ready to refinance.
Indeed, it may be a while before machine learning and Capture 2.0 technology becomes ubiquitous in our industry—but it is starting to happen. Today lenders are implementing these tools to slash labor-intensive tasks and reposition staff on more valuable work in which their expertise is truly needed. That’s just too strong of a value proposition in the current lending environment to be ignored. In other words, it’s no longer a matter of if Capture 2.0 technology develops into an industry norm, but when