How Onity Group is making use of AI

July 8, 2025 5:00 AM

Before most of the public understood terms like "artificial intelligence" and "data science" Onity Group's Jack Cavanagh was exploring how the fundamentals behind their development might apply to mortgage origination and servicing.

The growth from machine learning to the large language models and generative AI we know today can be seen as a continuum, and he has used his knowledge to help companies like Morgan Stanley and Caliber Home Loans develop strategies throughout his career. Today, he continues that work at Onity, where he serves as head of financial planning and data science.

With artificial intelligence improving at a rapid clip and becoming increasingly essential to business, Cavanagh recently spoke with National Mortgage News to share thoughts ranging from recent development in data science and its effect on mortgage workflows to consumer AI hesitation.

The interview has been edited for length and clarity.

How data science is employed at Onity

Cavanagh: There's a lot of things that can be done now at much more scale than could be done before.It could be anything from matching leads for loan officers, automating routine processes, enhancement of customer leads on the origination side, personalization of communications to borrowers.

Machine learning, natural language processing — those are two branches that really could touch on nearly every aspect of our business.

On the servicing side, it's which borrowers to contact, when to contact them, how to contact them. It helps you understand more about risk — fraud detection or prepayment risk, delinquency risk — being able to predict those.

I think another branch is across robotics — robotic process automation. We've automated at Onity 150-plus processes, saving tens of thousands of hours in a month.

[In] secondary markets — traditionally, the way loans or assets are valued is really more on a grid-based, cohort-based grids. By applying machine learning to it, you're able to really identify more hidden value. We're able to make smarter choices, smarter trades and then understand where there's opportunity in there.

A lot of it is really just being able to do it more at scale, and continuing to see where there's further opportunity with some of the chatbots and voicebots.

How AI in mortgage is poised to develop

Cavanagh: If you're going through all the different opportunities or all the potential implications that you could use it within the business, my guess is it could be a couple different paths. It could be some of the advancements that's gone on to automate underwriting.

That's your machine learning or deep learning. There is really kind of a rules engine that's running through 1,200 pages of rules, but at the same time, that's where I think there's continued opportunity.

Secondarily, the talk, especially the last two and a half years, is related to some of the large language models. Those aren't prime-time ready. We can still see even when we go Google something now, and you can see the AI results. There's all the warnings at the bottom, so we know it's not ready to be borrower facing, but there's certainly lots of opportunity for efficiency behind the scenes of trying to start to streamline some things there.

Addressing misconceptions about AI and public hesitancy

Cavanagh: I think there is a misconception about how AI or machine learning is being used within financial services, or specifically within mortgage. People think it's a black box driving the decisions.

That's not the way it works. It's all rules based on tried-and-true criteria: credit scores you get, income or loan-to-value ratios.

People don't really understand what AI is and what it can do. People think it's new. We run into this when we're trying to work with a different business leader or introduce it to a new segment of a business.

People's thinking falls into two camps. One is it's magic, and it can predict anything. I can remember having conversations where we would be focused on what happened in the past.

I remember a very strong business leader asking why. Understanding what the drivers were and what the causation was or what the strong correlation is — that's really going to help you be able to predict what's going to happen in the future.

I think the second camp that people fall into is, "We don't need it. We're doing fine. You think you're going to solve all of our problems." That's not the way it works.

It's in the middle. It's definitely going to improve the experience for the borrowers. It's not going to solve everything overnight, but there's definitely incremental improvements to be made.

How companies can leverage data to their benefit

Cavanagh: We've been in business for a long time, within servicing and more recently, ramping up the origination business. As we've been doing that, we have that history, and can use it to really understand what is driving, for example, payment patterns for our borrowers.

That keeps expanding too in terms of what's available about borrower data, additional information about economic insights. It's taking that new data and you can find additional insights that you weren't really able to just a few years ago. There's different opportunities with new algorithms coming.

It's going to be not just customer facing. It's also providing competitive advantage. It's helping to give sharper forecasts in terms of financial forecasts.

If you want to understand what your volume is going to be, if you want to understand what the prepayments are going to be, what the delinquency is going to look like, it really kind of fits together. To be able to translate the predictions into the financial impact, I think it fits together in that angle to connect those two pieces.

The challenges and opportunity data with AI growth

Cavanagh: I think it's less the growth that's giving them challenges. I think it goes back more to educating the partners that they're working with.

Data scientists, they're part of a team. That's the way I think of data science. It's everything from the business operators, subject matter experts, technology people; it's really all coming together as a team. It's making sure that everyone's on the same page, and then everyone can start to work together to be able to identify and approach solutions.

That's been my experience of managing these teams. That's really the first kind of hill that you've got to get over.

I think some of the advancements are exciting for the data science community. It's definitely been exciting as algorithms have been expanding and improving, and as more data has become more accessible,

Understanding your infrastructure and building that data engineering is really kind of that first step of what you need to do to make sure that you have the infrastructure and the data available. That could be your own internal data as well as the external data.

One of the things that we put in place are variable libraries that you can build for the data scientists. You have data engineering folks that are focused on making sure the environment is ready, reliable and the inputs are already validated.

The future effect of AI technology on consumers

Cavanagh: I think a lot of it is just continuing to improve the customer experience. Especially within the mortgage business, the ideal is a streamlined experience that everyone's looking for.

More and more, especially over the last 25 years, everybody's focused on wanting to solve it themselves. That's easier and preferred to be able to self-service and get the information that you need. I think there's enormous opportunity for that.

If you think about it, you're not really looking to call your mortgage servicer anymore than your mortgage servicer is looking to make you call to get a question answered. We want to be able to streamline the process. I think the opportunity is to streamline the processes so it's easier to have those interactions back and forth.

Even in terms of the underwriting, reducing the paperwork and reducing the processes and the wait time is just something that has long been sought after. I think that we're really close to making significant progress.

On how the technology environment has changed

Cavanagh: I've been in mortgage and analytics and finance for a couple decades. I've definitely seen a change in terms of how it's evolved, and for the most part, it's the application that I've seen with the more advanced machine learning and natural language processing.

There's less of a debate, I would say, about whether we're going to need it. I think there's more consensus around the fact that the environment is changing.

Specifically, over the last 10 to 15 years, is really where I've seen more of an introduction within the mortgage business.

Over the last 10 years, I've seen varying levels of interest within different competitors in the mortgage business.

If you go back, machine learning has been used within financial services for 30-plus years. That could be linear regression, logistic regression; those are the earliest algorithms that were applied and used within financial services.

It goes back to the way I think of that continuum. When I started, there was analytics. Can we understand and be able to predict loans that are going to pay off, loans that are going to go delinquent? It's really just further refinements of solving those same problems, or making those same kinds of predictions that's evolved.