How PMSI Is Leading the Way

The mortgage servicing industry is primed to take advantage of AI. The smallest short-term adjustments can dramatically impact an investor’s long-term gains, and AI can be the perfect tool for processing the vast quantities of data that go into perfecting business and operational strategies. But maximizing returns with AI is not as simple as typing a prompt into ChatGPT. Results depend on picking the right application, and on supplying highly specialized algorithms with pristine, well-structured data and countless hours of Machine Learning (ML) training. 

While many mortgage servicers may be hesitant to embrace this approach, PMSI has been training and optimizing systems and AI algorithms for more than a decade. Here’s what sets our approach apart from the rest of the industry to give our clients a distinct advantage.

The Challenges of Generative AI

AI can be broadly grouped into two categories: generative and narrow. When most people think of AI, they’re thinking of generative AI, the technology powering productivity-enhancing platforms like OpenAI’s ChatGPT or Google’s Gemini, as well as the predictive text on your phone’s keyboard and the self-driving functions of autonomous cars.

Anyone who has experimented with generative AI knows it is equally capable of amazing things and startling errors. Many organizations are looking to quantify how useful generative AI really is. Stanford University has conducted a number of tests, published annually in their AI Index Report, to track the effectiveness of AI predictions against human benchmarks. They have come to the consensus that generative AI will usually score higher than an average human but below expert level in more complex fields.  

Results below a human expert’s are unacceptable in an industry like mortgage servicing, where relying on anything less than expert knowledge can lead to costly mistakes. These tests have also shown that the more specific the field, the wider the gap between a human expert and generative AI, and mortgage servicing is a highly specialized field of knowledge. 

So how can PMSI use AI to deliver better results for our clients? Because we’re not using generative AI. We’re using narrow AI.

How PMSI Uses Narrow AI

Narrow AI is a category of AI systems purpose-built for a specific task. This is generally done through research and testing of leading open-source, special-purpose ML algorithms designed to excel at one desired use case.  These systems must be carefully calibrated and trained with large amounts of the user’s highest-quality data. Using bad data in an ML training scenario is like teaching a new employee the wrong way to complete a task.

At PMSI, we have leveraged over a decade of high-quality data augmented by expert systems to train narrow AI systems, specifically focused on the mortgage servicing industry. We’ve trained our platform on over seven million loans and have thousands of lines of code powering algorithms that give expert-level results. 

Our unique segment of the mortgage industry requires very high precision to meet complex investor reporting and accounting requirements. To manage servicer, MSR owner, and investor cash flows, PMSI must ensure all information is accurate with absolute certainty. Even a small discrepancy can have drastic consequences. 

Over the past five years, we’ve observed on average that 1.5% of the loans in a portfolio will have either a reporting exception, a loan attribute exception, or a cash flow variance between the servicer and the investor.  This can range from as little as 0.5% to as much as 6.0% any given month.  While 1.5% of a portfolio is not a big number, multiplied across several million loan observations, 1.5% equates to 45,000 items that require research each month, exceeding over a half billion dollars of servicer and investor cash flow variances to research every year.  

This intensive data and cash flow reconciliation reveal nearly $60 million in test of cash and shortage surplus variances each month that most servicers do not notice.  It is therefore essential that any AI tool we use is trained for absolute precision.

Training for Precision

Generative AI can’t hit the extremely demanding precision benchmarks required to assure client accounts are accurate, and investors can’t accept anything less than perfect accuracy. But narrow AI can meet these expectations — if it’s trained properly. 

ML algorithms that serve specific purposes must be trained like human workers. Let’s say you have a human being doing something for the first time. You teach them how to perform the task, then you let them practice on their own. Eventually, you come back and see how they’re doing, and make corrections or provide further instruction as needed. 

AI is no different. If you don’t spend the time and resources necessary to train your AI, or you don’t have the data repositories and technical infrastructure to do so, you run the risk of getting lulled into complacency, leading to inaccurate results. At PMSI, we have made a long-term commitment to building expert systems, AI, and ML programs, so our systems provide results far beyond what others in our space can offer.

Preparing for the Future

Our AI initiative offers promising results, because PMSI has been committed to developing this expert systems program for over a decade. We have built a pristine data environment with millions of individual records. To oversee this uniquely ideal data pool, PMSI employs extremely qualified individuals with advanced degrees who analyze, train, and constantly improve our narrow AI systems, achieving the best possible results for servicers and investors.

These efforts have put PMSI in an industry leading position. We encourage everyone in our field to start investing in AI as soon as possible. Advances in generative AI are happening at an exponential rate, multimodal models already excel at image classification, with text comprehension meeting or exceeding human benchmarks. 

However, generative AI trails in complex tasks like advanced mathematics whereas the research demands of cash flow analytics can be augmented through Narrow AI. Our prediction is that many forms of AI and ML will be inextricably woven into our industry in just a few short years. Any business leader not at least evaluating AI today risks being left behind. 

To learn more about PMSI and the ways our proprietary technology is helping the mortgage industry, get in touch with us today!