Executive Conversations is a HousingWire web series that profiles powerful people in the financial industry, highlighting the operations and the people that make this sector tick. In the latest installment, we sit down with CEO of SoftWorks AI Ari Gross to discuss how his background in machine perception and computer vision research shaped the development of the company’s mortgage automation software.
Q. The mortgage industry has been trying to completely automate the loan process for years, with uneven results. What factors are making that ambition more urgent and more possible today than ever before?
A. The biggest reason is simply the cost of doing business. According to the Mortgage Bankers Association, it costs nearly $9,000 for originators to process a loan. In addition, mortgage banks are reporting a net loss of $200 for each loan that was originated in the fourth quarter of last year.
Plus, there’s increased competition in a time-sensitive environment to turn mortgages around in near real-time. Many of the first-time home buyers entering the market are millennials. Most are accustomed to purchasing everything on Amazon, which is not only fast, but delivers a superb customer experience. Lenders are beginning to realize that these buyers aren’t interested in waiting weeks or months to find out if they’ve been approved for a loan.
As a result, we’re seeing companies across the mortgage lifecycle look for reliable solutions to automate and accelerate all aspects of mortgage processing. Whether you’re in origination, mortgage insurance, or servicing, you’re feeling the pressure to bring automation into your workflow. Companies that are slow to adopt advanced technology will find it increasingly hard to compete in the mortgage industry.
Q. What is the biggest obstacle to achieving a completely automated mortgage process?
A. The biggest challenge we’re seeing across the industry is that organizations haven’t found an automation solution that is sufficiently reliable. As a result, human validation is still required at every step of the mortgage loan process, meaning someone must go back and fix mistakes that the software made. This essentially eliminates much of the ROI that was supposed to be achieved by implementing an automation solution.
Q. How do SoftWorks AI solutions overcome that obstacle?
A. There are really two ways we can overcome that obstacle. First, our team has deep expertise in the mortgage industry. As a result, Trapeze for Mortgage Automation is essentially an expert system that has already been extensively trained and highly optimized on recognizing hundreds of mortgage documents and extracting thousands of data fields.
In addition, our background in OCR-based recognition and machine intelligence allows us to extract data with a high degree of accuracy. As part of the cognitive automation process, the system assigns a precise probability to every machine operation, which lends itself to a very reliable auto-validation protocol. This gives our mortgage automation system a detailed understanding of what the solution does and doesn’t know. From this deeper understanding of mortgage data, we can build an increased level of touchless automation. This, in turn, enables our solution to drive higher performance across all facets of the mortgage workflow.
Q. Your company’s software, Trapeze for Mortgage Automation, has its roots in a Computer Vision & AI research lab. How did that experience help shape its development?
A. I started my research over 20 years ago in a research lab in the areas of computer vision and machine perception. Originally, the goal of this research was to allow machines to recognize their environments and help the visually impaired better navigate their surroundings using 3D OCR. This eventually led our team towards adapting machine learning methods to develop new document understanding and data compression technology. The document understanding and compression technology became part of the ISO JBIG2 standard, and we subsequently licensed this technology to industry leaders including Adobe and Panasonic.
The advanced methods we developed using computer vision, AI, and document understanding enabled our software to achieve a level of precision significantly beyond what normal OCR and RPA technology can achieve. It allows us to automate reasonably complex problems in tax, accounting, and finance that are typically considered knowledge work. These tasks include reliable text extraction, classification, data extraction, footnotes, stacking and routing. We have successfully applied our technology to several verticals in fintech including the mortgage industry.
Q. What kind of results are your clients seeing by leveraging Trapeze?
A. Our mortgage automation solutions have shown over 50% cost reductions in the mortgage underwriting space. This has allowed underwriter productivity in the Mortgage Insurance (MI) industry to increase by over 100% with respect to loans processed daily per underwriter. In addition, many of our origination clients have realized a reduction of more than 80% in their loan application processing time.
