Introducing Cenlar’s Cutting-Edge Data Effort
Introducing Cenlar’s Cutting-Edge Data Effort
You can’t manage what you don’t measure.
With this philosophy at the forefront, Cenlar has sharpened its focus on data in 2023. We’ve hired Michael Biddle in the newly created role of Chief Data Scientist, and charged him with leading Cenlar’s just-launched information factory, an effort that will focus on leveraging the company’s large amounts of data into actionable information for our own operations and for our clients.
It’s work on the cutting edge — combining math and statistics, specialized programming, advanced analytics, artificial intelligence and machine learning. Luckily, Biddle is the right person for the job, with more than two decades of experience in building and sustaining quantitative analysis and research, most of it focused on using mortgage performance and servicing operations knowledge to uncover actionable insights hidden within vast data sets. The insights gleaned from Biddle’s work can guide decision making and strategic planning, as well as improve operational performance.
Biddle recently sat down to share more about the field of data science, as well as his vision for how data analytics will allow Cenlar and our clients to act faster and make better, more informed decisions.
 | Question: What is data science? Michael Biddle: Data science is just the new term for what I’ve been doing forever. It involves the accumulation and cleansing of data. But the primary function of a data scientist is the exploration of data to gather insights and offer guidance. Q: How do you view your role at Cenlar? MB: A big part of my role is getting a handle on the data within the organization — procure the data, gather, cleanse and bring it all together into one place in the form of a data warehouse. We also need to institute processes for converting that data into information through effective reporting. Ultimately, we’re trying to achieve an information democracy, where everyone that needs data to make a decision has access to it. That’s a big part of this. I’m also working on the institution of artificial intelligence solutions to help with guidance of collections calls and really maximizing the performance of the portfolio through deep learning methodologies and artificial intelligence. The primary initiative with implementation of data science and data modeling at Cenlar is to detect patterns that there is changing performance. One of the great benefits of using the model is we recognize this early, and we can be in contact with the right homeowners when they need us to be in contact with them. Not only do we get to them quickly, but we also head off any catastrophic issues. It betters the homeowner’s experience. |
Q: How does deep learning work in servicing?
MB: Deep learning is essentially a pattern-recognition mechanism. It will identify characteristics that define the disposition of the homeowner, whether they’ll make their payment or not. Once they’re a couple days late, we then assess how much we should call them based upon the likelihood they’ll make their payment by the end of the month. The last version of my model had 54 million patterns that would determine the probability they would make that payment. It’s based upon things like the terms of the loan, the homeowner’s geographic location and the external stimuli that are affecting conditions in that area. So, if there’s a severe depression localized in their area, they’re going to have a different disposition than someone in an area that’s thriving. We take that into account.
The collections operation has a limited capability of making calls, so this will direct them to the people who need the call and those who will react to it. We can really maximize the operation.
In loss mitigation, we simulate different actions a homeowner can take — modification, forbearance, etc. There’s going to be a pattern that some people are going to react well to the terms of the modification and some folks will re-default in two months. What we do is take those patterns and match the right action or strategy to the homeowner.
We monitor economic indicators like unemployment in an area, home price appreciation, population migration. The most recent version of the model has 109 inputs that identify the pattern. That’s also taking into account the homeowner, the property, the loan, the economic conditions. An average model has two inputs: pay history and FICO score. Using factorial, that’s four patterns, maybe five. My model has 54 million.
If you go to any other industry outside of mortgages, this kind of deep learning is heavily utilized. Amazon, for example, is using these exact methods to determine what products to present you and get you to buy. Freddie Mac has made announcements that they are adopting AI solutions, so I think this will start to open the gates. We’ll see more adoption marketwide.
Q: You mentioned AI. There seems to be a lot of concern around Artificial Intelligence. Does it deserve that reaction?
MB: Most people, when they hear the term “AI”, think of movies — the Terminator or iRobot, something like that. That’s just science fiction.
When you say “artificial intelligence,” all that means by definition is a computer system programmed to make a decision. That can be as simple as having it pick the item in the middle. Or it can be a deep learning neural network that’s gathering a ton of information.
In our application, there’s always a human element. It may issue guidance that you should offer a loss mitigation option to a homeowner, but it’s still humans who decide to offer it.
Q: What’s going into the launch of Cenlar’s cutting-edge data effort?
MB: We’re going to build out a team. Things like data warehousing, particularly data procurement, is a tough job, and it can take a while. We plan to do this in increments.
The implementation of our modelling suites is going through process now. The command of the data is probably the biggest piece of the effort, and it always is, in data science. We’re going through a project now to analyze and clean the data to make it acceptable for modeling. Then, we’ll make it more accessible for our operation areas.
We’re being aggressive with this. We implemented the testing in April. We’re aiming to have the model integrated by June 1. We believe that 10% of the population will be serviced with the new modeling starting in June, for a 2-3 month period. That way, we can test and compare it to the rest of the population.
Q: How often do you currently sweep Cenlar’s data looking for changes?
MB: Every day. I do a full analysis each day. We’re working on implementation now, but the plan is that every day at 4 a.m., we get yesterday’s data, analyze it and offer guidance to the operation.
Q:Cenlar services hundreds of thousands of loans. That’s a large and diverse set of data. Ultimately, will Cenlar be in the position to use its standing as the nation’s leading subservicer to take data and share insights about what we know about market conditions?
MB: Oh, 100%. We have a wealth of knowledge hidden within our data. My task is to gather that data and turn it into knowledge so we can make decisions and share that knowledge with our clients. That’s the very idea behind our data effort.