As mortgage industry technology continues to evolve, opportunities for greater efficiency and cost-per-loan reductions are continually increasing. Today, the utilization of robotics and automation – along with decisioning logic – has helped to advance the mortgage process from the sluggish, error-prone efforts of a decade ago to a far more productive environment with a focus on data integrity and customer experience.
However, newer technology innovations have now reached a point where it is possible to do much more than simply automate tasks.
With machine learning, mortgage professionals can now achieve bigger goals and deliver greater value to their customers. This advanced technology can handle much of the labor-intensive work that experienced, well-trained underwriters and processors have been responsible for in the past. The result is that these professionals can focus their expertise on managing processing exceptions and problem solving, rather than spending a large percentage of their time buried in “stare and compare” activities.
While the terms “machine learning” and “artificial intelligence” are often used interchangeably, there is a distinction between the two. Using the broadest of definitions, artificial intelligence is a subset of computer science that looks to replicate human reasoning through learning, problem solving and pattern recognition. Machine learning is a specific application of AI.
For simplicity’s sake, let’s consider an industry-specific example. AI-powered machine learning enables technology to ‘remember’ standardized forms, learn from them and then anticipate the type of information that should be in each field of the form. It can also leverage visual recognition to image and index a wide variety of documents that are typically reviewed by processors and underwriters, such as tax returns, W-2s, property titles and appraisals.
When machine learning is used, processors and underwriters can focus more time on higher-value activities to keep the mortgage process on track, and less time on document-level work. For example, by freeing up mortgage professionals time from tasks such as comparing and validating data on standardized documents, they are able to spend more time making sure the mortgage remains on track and the homebuyer’s experience is as trouble-free as possible.
Machine learning can also learn a task and combine it with other tasks to complete a specific process. For example, a system can learn to look at two paystubs, determine that the customer gets paid every two weeks, and then do the math to confirm the annual compensation on the application is correct. In addition, it can look at information under review, evaluate results, and employ interactive communication bots to advise the processor or underwriter of an issue that may need attention.
Neural networks are a machine learning approach that mimics the neurons of the human brain. They excel at performing such tasks as image recognition (as in the use case described above), piloting driverless cars and speech recognition, to list just a few. Neural networks form the basis of much of today’s applications of AI, and it is a widely-held view that their use will bring incredible change across virtually every aspect of the industry. In fact, many believe this technology is poised to transform almost everything we do, and is the basis of much of today’s applications of AI.
While the concepts behind neural networks have been around for some years, the necessary computational power and vast amounts of data required to build functional neural networks at scale has only recently become available. Neural networks are becoming increasingly more accessible and easier to adopt.
Tools such as Google's TensorFlow, Microsoft's Cognitive Toolkit and MXNet from Amazon are providing the necessary frameworks for developers to bring the promise of neural networks to a variety of industry verticals. Some of the most innovative and leading edge companies in the mortgage vertical are now bringing this technology to bear on the very specific needs of the industry.
At a high level, machine learning is the process by which AI deepens its knowledge by completing a task, processing information or accessing functionality – literally many thousands of times – via neural networks.
Building on the example from above, in evaluating one paystub, AI may look at hundreds of thousands of available samples to determine exactly where specific information on a paystub is most likely to be located. The system learns to anticipate where the required information should be and how to find it. Moving forward, it looks for the information it needs in the place where it expects it to be found. When it encounters less traditional layouts of information, it uses this prior-gained understanding to find what it is looking for (or, to ascertain that the information is indeed missing). This then becomes part of an expanding knowledge base. The more an AI scans, the more it learns, and the more accurate it becomes.
Shaving Time Off the Mortgage Process
For mortgage applicants and lenders, one of their most pressing concerns is often how fast a mortgage can be approved and closed. By expanding the use of machine learning and AI, the origination process can certainly be expedited, with processors and underwriters dedicating more of their time to addressing exceptions identified by the technology. In this way, days can be shaved off the mortgage timeline which drives down the cost and enables staff to handle larger volumes. This will be a welcomed outcome for all parties involved.
In addition, machine learning/AI can help avoid last-minute delays by prompting processors and underwriters to take early action to keep the origination process moving forward. For example, if the system identifies a large bank deposit in a borrower’s account, it can prompt the mortgage professional handling the loan to request clarification documentation from the customer. By leveraging machine learning/AI, an issue that could otherwise hold up the loan is resolved long before closing day.
The Impact on Jobs
When the subject of automation, machine learning and/or the deployment of artificial intelligence in the workplace is discussed, it inevitably raises the question of the impact on jobs. Do these technologies and advancements mean that jobs will be lost? Or, will they simply be shifted to higher value activities?
There is no perfect answer to this question since the utilization of AI is different from company to company. One thing that seems certain is that the future will require new skill sets that fit into this shifting technological paradigm.
However, as it applies to the mortgage industry today, these technologies can certainly enable mortgage professionals to spend less time on remedial work, increasing capacity and decreasing costs. Advanced technology can handle the validation of much of the mortgage-related documentation, while processors and underwriters focus on ensuring that any issues that may arise are quickly resolved, essentially turning task executors into knowledge workers
In addition, freeing up the valuable time of mortgage professionals, like underwriters, may allow mortgage companies to be more aggressive in promoting growth. With experienced talent already on board and the availability of more time to focus on moving loans through the pipeline, new opportunities are created for employers and employees alike. Not only can lenders scale and decrease their cost per loan by keeping the staffing levels flat, but it also supports other investments, such as product development, marketing and needed infrastructure.
The Integration of Humans and AI
Moore’s Law states that overall computational power would double every two years, and in large part, that theory has borne out. Not only has computational power expanded exponentially over the years, but it has been made available to almost everyone in the industrialized world.
Couple that with advances in machine learning, and we’re doing things that were inconceivable just five or so years ago. Smart homes controlled by our voices; driverless cars; lights-out processing of previously hands-on, labor-intensive tasks involved in loan origination.
In many ways, the mortgage industry continues blazing new trails as this technology advances. As a whole, we’re still learning how best to integrate the invaluable skills and talents of humans with the tremendous opportunities presented by AI.
Of course, humans still decide what problems they want to solve for and how they can leverage technology in that effort. For those of us in the mortgage industry, we must still play an active role in decisioning the loan, determining what kind of data to consider, what is and is not relevant to the equation, and what the mitigating factors might be. As technology generates results, humans decide how to use them, if at all.
At the end of the day, the mortgage industry’s future largely depends on how well we adopt and adapt as the power of technology accelerates and expands the possibilities. The mind reels at how far we can go.