In the US residential mortgage market, Fusion MortgagebotPOS lenders are struggling with rising interest rates, decreasing mortgage activity, and an increasing number of non-bank mortgage originators advertising simple and easy online mortgage applications. Lenders are asking, “How do I get a bigger share of purchase mortgage loans?” One answer is that they need to market to potential borrowers and let them know they offer an even better online mortgage experience than firms like Quicken Loans, the largest retail mortgage originator. If they can get their name out there before the borrower starts looking at homes, they’re less likely to lose that borrower to a Google search for “online mortgage application” or a link to another lender from a Realtor’s website.

But how do they know WHO to target to make the most efficient use of their marketing efforts? Using data from the MortgagebotPOS product, we can provide a solution that helps our clients market to the mortgage leads who are most likely to qualify for a mortgage. The MortgagebotPOS database is a goldmine of mortgage application data, including borrower and geographic demographics. This is just one way to tap into that resource.

What it does

We’ve created an API that allows lenders to send us profile information for their mortgage leads. Based on machine learning results, we can respond with whether that borrower is likely to be approved for a mortgage. If the borrower is likely to be approved, we will run a rate search via the POS Rate Search API using the lead profile information provided in order to get the most accurate rates possible. Then our lenders can alert the leads that they’re eligible to apply for a pre-approval AND share estimated rates for some of their popular mortgage products. They can link directly to their MortgagebotPOS application and, in the future, use APIs to prefill as much borrower information as possible.

How we built it

MortgageConnect is deployed as a .NET Core API on top of an Azure Web App. As new profiles are pushed to MortgageConnect API, the profile data is saved, an event is raised, and a new profile identifier is returned to the end user. Azure Event Grid manages directing this event to a subscribed Azure Storage Queue for processing. Downstream, a background process within the Azure Web App pulls events off of the Azure Storage Queue and begins processing the profile eligibility data.

First, the background process loads the requested profile from the event and calls an Azure Machine Learning Studio Web Service, filling in criteria based off of the profile. Secondly, we call the FFDC MortgagebotPOS Rate Search API using our profile data to fill in the rate search criteria.

Data from both of these calls is saved to the profile, and an event is raised signaling process completion. All of this data is now accessible through the MortgageConnect API.

End-users can subscribe to Azure Event Grid notifications via WebHooks to be notified of completion events for eligibility, to know exactly when the data is ready to be fetched and processed further.

Challenges we ran into

As we formed this idea, we realized how much potential an API like this could have and all the various capacities it could be used in. We had to narrow down exactly what we wanted to tackle for the Hackathon, which meant discussing all of our ideas and determining which ones to cut. We also had to learn how to use Azure Machine Learning Studio, since none of us had experience with the tool.

Accomplishments that we're proud of

We were able to incorporate an existing API, create a new API, learn how to develop a machine learning model and then deploy it. This allows us to monetize existing data within our MortgagebotPOS product to create a new and relevant service in FFDC. In addition, we created prototypes for two use cases to tell the story about how clients could potentially use our new service. We feel that we were able to accomplish a lot during this short time period!

What we learned

We learned a lot about how to develop a machine learning model using Azure Machine Learning Studio. We researched current US mortgage market conditions and learned that the percentage of refinance applications hit a 20-year low last year and is predicted to break an even older record this later this year. The application of purchase applications, however, hit a 9-year high just last month. We determined the most important data points to include in our Machine Learning dataset through discussion of the automated underwriting systems integrated with MortgagebotPOS.

What's next for Mortgage Connect

We have a list of ideas that can expand upon this new service and apply to new use cases. For example, we have discussed how 3rd parties (meaning they are not MortgagebotPOS clients) could use this service:

  • House hunting/Realtor websites
  • Financial applications such as Mint that are not specific to a financial institution
  • Companies that generate and sell mortgage application leads such as Simply Online Leads

In addition, we could flesh out this service and determine what additional data points we could collect, analyze, or provide to our clients.

Slide deck


Built With

Share this project: