The speed and complexity of mortgage bars make it difficult to keep discrepancies and resolve, which often results in substantial cash flow varies. For example, investor accounting represents around 40% of the GSE scorecard statistics, but only about 2% of the cost structure of an administrator. That is an important mismatch.
Independent mortgage banks (IMBs) dominate the origin sector, accounting for more than 80% of all new mortgages. These loans are often sold to entities such as Fannie Mae and managed by large companies. The IMB or the bank will collect a portfolio for a few months and then sell their mortgage services to an MSR Reit, and the cash flows will change hands again. In order to make things more difficult, these loans are sold to the GSEs and GNMA with different types of transfers that determine how cash flows are treated for planned or actual main and interest handling.
All these clouds add the visibility of actual data by adding layers and layers of sub -service functions. The excess maintenance strip can be further carved between the service costs, guarantee costs and the excess or interest only. And we are not even ready.
Now, Overlay Broer behavior – some loans are delinquent, some have been paid in advance, some have limitations, etc. – and these must all be normalized on the basis of the transfer type, so that the investor is assured of its cash flows and the GSE’s catering accounts are well financed.
So how can the mortgage industry cut through this complexity and get a clear picture of things? Fortunately, there is a simple answer, a new tool that makes it possible to quickly process huge parts of data and see a sharp image – it is called Narrow ai.
The transparency dilemma
Think of cruising in a boat in quiet waters. It is a sunny day, but you look ahead and see an iceberg floating on the horizon. It is an unexpectedly beautiful sight to see … But an expert captain sends wide because she knows that only 10% of the mass can be seen, while 90% – the most risky part – is below the waterline. Accounting of investors is no different.
Our segment of the mortgage industry is unique because it requires a very high precision due to the complexity of accounting requirements. We must ensure that the cash flows that we manage with absolute certainty are accurate. Even a small discrepancy can have drastic results.
In the past five years we have observed an average of 1.5% of the loans in a portfolio with an exception every month. This includes reporting exceptions, exceptions of loan attribute or cash flow varies between the investor and the manager. These exceptions varied from only 0.5% to more than 6% on a certain month, usually driven by human errors when touching a loan during transitions, transactions and other terminal events.
Although an exceptional percentage of 1.5% may not seem like a large number if it is used every month on a few million loans, this corresponds to 45,000 items to investigate. And don’t forget, that’s just an average. This in turn represents almost three -quarters of a billion dollar in cash flow varies to conduct research over the course of a year.
That is very real money, but very few organizations can even quantify it because they have no transparency in the problem. They look in Excel -spreadsheets, often Freeform -notes to describe their research with few extra details about the cause and the effect why they have labeled it as such. If they no longer have time, they succumb to the temptation to copy and paste – or even worse to connect.
Introduction of narrow AI
When people think of AI, they immediately think of chatgpt. Chatgpt is a multimodal generative AI tool that does many things well, but is in very little expert. When you do this generative AI tools benchmarkt against specific disciplines such as companies, science and art, they still follow human possibilities with a gap of 30% or more. Although they close this gap quickly, a considerable amount of time is still expected before AI of this type can be trusted with very accurate tasks such as those with which we are confronted.
Smalle AI is a category AI systems that is specially built for a specific task mathematics, decision-making analysis, etc. consider them as specialists, while Chatgpt is a generalist. Smalle AI is made by thorough research and testing of leading open source, special-Purph ML algorithms that are designed to excel in your use case. But they must be carefully calibrated and trained with your data.
Machine learning of this nature requires huge amounts of data and extreme care to ensure that data is of the highest possible quality. The use of poor data in a training scenario for machine learning is essentially the same as teaching a new employee the wrong way to complete a task.
Our world is one of demanding precision. A good profile for a person who is skilled on investor accounting is someone with a strong predisposition for problem solving, excellent math skills, a high tendency to learn and someone who naturally gives every small detail.
The machines that we want to do are not otherwise wanted to hang out with the Math Nerd machine, not the well-rounded popular.
AI in action: Success stories
For more than ten years we have built systems from a perspective of analytical use. The challenge is how you can maintain high quality, robust data for long periods, so that we can continuously utilize the power of that underlying data. At PMSI we have achieved this at large costs in time, efforts and financing to create narrow AI that is able to significantly reduce the time needed to go through all this data. Our researchers – assisted by our expert narrow AI algorithms – are able to perform more checks and balances than a standard operation.
That’s why we catch more.
Some examples of our performance using these methods are:
- Reducing $ 6 million in data differences and cash varies in portfolio versions that would otherwise have been accepted by the customer because we could identify and quantify the 1/10 of 1%.
- Return more than $ 38 million in unnecessary P&I claims for one customer on one GSE portfolio.
- Customers from the worst to the first elevation in their GSE scor cards, reaching ongoing FNMA star ranking and FHLMC Sharp rangers.
- Solving 24,000 polar reports -Extrazements in 3.5 hours to prevent GNMA fines.
- Research into more than $ 60 million in cash variants every month, whereby our algorithms automate 75% to 85% of the research.
The future of AI in mortgage control
There are no two ways around – integrating AI into your workflow will make your mortgage control more efficient. Nowadays there are areas that AI is equal to or surpasses human possibilities – image recognition and text comprehension, to name just a few. You should concentrate on integrating these options in your workflow today, such as engaging AI for document classification during loan transfers and loan boarding.
Within call centers, chat lines and legal guidelines, AI can be used for data commitment and document parsing – you only have to use your imagination to find places where an AI processes can speed up by discharging your staff.
For narrow AI applications as described above, it is honestly too late to start building your own solutions. By the time you have collected the necessary data for training and have ensured that it meets the strictest quality standards, the innovators in the room will be called in to models of the next generation that you are starting now, surpasses a lot. But it is certainly time to achieve these innovators and to investigate how their technologies can be used to help your company get ahead.
Simply put, AI transforms the mortgage control industry just as quickly and radically as it transforms so much differently around us. It is time to board and find ways to implement AI in your own workflows.
Daniel Thompson is the CEO of PMSI.
This column does not necessarily reflect the opinion of the editorial department of Housingwire and the owners.
To contact the editor who is responsible for this piece: [email protected].
