An Impassioned Plea for AI-Ready Mortgage Policy Data

By Tela G. Mathias

The single hardest problem in mortgage is compliance. Being compliant. Staying compliant. Proving compliance. Understanding compliance change. The goal of our regtech startup, Phoenix Burst, is to simplify mortgage compliance. Imagine a world where compliance just happens. What if you could push a button and provide regulators and housing agencies with evidence of process compliance?

It all starts with AI-ready policy data provided by housing regulators and agencies.

At the heart of compliance is change. Something changes and we have to figure out what to do about it. At PhoenixTeam, we envision regulatory change fulfillment as a wheel, with change at the center and evidence of compliance as the outermost ring. The bullseye of the wheel is some kind of a change notice. Today, we can find out about a change in a variety of ways – a notice can be published on the federal register, an email notification from an attorney regarding a state change, a marked-up handbook chapter, a webpage update. Maybe a combination. Some of us subscribe to AllRegs. Throw in a Thompson Reuters and a Lexis Nexus. Maybe hire an attorney firm to decompose the Freddie Mac servicing guide. You get the point.

I liken this whole process to one of “trolling the internet for changes”.

I say this lovingly. I’m certain that regulators and housing agencies are truly doing their absolute best to communicate what they need. I just think we can do better. I know we can do better.

Let’s take the Federal Register application programming interface (shout out to whomever stood that up – we love it, and we use it almost every day!). We are able to consume regulatory language through this interface, compare current to prior versions of the regulations, determine the changes, and create plain English change statements in minutes. And get this – we can do it with almost 100% accuracy.

We can do that same process with a “track changes” portable document format (PDF) document as well. But it’s so much harder, and our accuracy rate is only 79%. And when I say much harder, I mean really a lot harder. The document snippet above was created by Phoenix Burst using generative artificial intelligence. No human was involved. As we are still testing this capability, we did perform 100 percent substantive human evaluation of all results to ensure accuracy. These human evals will ultimately move to system evals as we productionize these capabilities.

We intend to take the following actions to help move the industry towards AI-ready mortgage policy data:

1. Meet with the Mortgage Industry Standards Maintenance Organization (MISMO) and start the conversation about a standard interface for regulatory and policy data.
2. Meet individually with each regulator and federal housing agency to collaborate on needs and start to formulate a standard.
3. Publish a white paper for the industry on how agencies and regulators can work with industry to simplify compliance, starting with AI-ready mortgage policy data.
4. Host a generative artificial intelligence (genAI) summit in April of this year, in Washington DC, centered around regulators and agencies and how the industry can work together to take advantage of these new technologies (responsibly!).

We are looking for ideas and input from everyone, so please reach out if you would like to join us. Stay tuned for more on how genAI will upend compliance in mortgage. Nothing changes if nothing changes, so let's change it!