Innovative technology could be a key determinant of whether servicers have competitive expense controls that are crucial in the current market, according to executives at mortgage fintech Livegage.
"Mortgage servicing is hard. There's a lot of focus on the cost and the margins. It's not an easy business to be in," said CEO Anupam Sarwaikar.
Because of this, "the people that will actually survive in the mortgage servicing business will utilize technology as much as they can,"
These may be timely messages for servicers given the trend toward
In the conversation that follows, Blair and Sarwaikar share some of the ways servicing technology can be used to manage costs.
Their remarks have been edited for length and clarity.
Where artificial intelligence can play a role in mortgage servicing
Sarwaikar: I realized that someone needed to do something to reduce the cost of servicing and bring in efficiency. That was the genesis of the founding of Livegage. The name Livegage came up because we wanted to bring some life to the mortgage industry.
We started by building our own artificial intelligence engine before ChatGPT became a household name and started using natural language processing. We started creating our own word vectors with mortgage lingo so a model could interpret guidelines and create servicing rules. These rules can run on a portfolio and predict potential breaches. You can also add operational rules.
We also started by working with investor accounting and did three-way reconciliation between banks, servicing systems and cash movement. We calculated servicing fees so that servicers were not leaving money on the table. We designed technology for asset managers and investment companies.
Mike joined us because we had a need for a senior executive who has built a servicing organization before and has helped it grow.
Automating 'where the tube hits the side of the river'
Blair: About 75% of what we do in servicing can be automated, but people don't have time to do it. If you want to change the tires on the car, but you're going at 90 miles an hour, it's a challenge. As Anupam said, it's a tough business. You're counting pennies, especially if you're a subservicer and you don't originate loans.
I do think the people that will actually survive in the mortgage servicing business will utilize technology as much as they can. If you can envision a tube on a river flowing through as very inexpensive, where we want technology to help us is where the tube hits the side of the river. We want to make sure we learn from that, so that it doesn't happen again. We're not fixing the loan. We're fixing the exception, so that that exception doesn't occur again, and that's how we help people to be more efficient.
That gives the servicer a leg up in regards to a lot of processes. There are probably 30 areas within servicing operations. In each of those areas, servicers have procedures related to every type of investor.
Training bots to handle certain exceptions
Sarwaikar: There are many letters which need to be sent during the life of the loan. Freddie Mac's guidelines differ a little from Fannie Mae's, Ginnie Mae's or the Federal Housing Administration's. The FHA insurance claim you are filing may depend on whether you did certain things in accordance with a timeline. Private investors may have their own guidelines.
So automation has to handle different guidelines and different product types, creating a very, very complex system for a process for servicers to follow. When they miss things, there are losses and penalties. A system should tell you about those things ahead of time, so that you have time to execute and don't miss the deadlines.
We also have developed a suite of AI bots, or AI agents, which actually can start picking the low hanging fruit to the extent the servicer is comfortable. Once they have enough data to understand the workflow, a repetitive human task can be automated. AI bots can actually start picking those tasks to be executed from the queue and start executing them, reducing the cost of servicing.
Blair: If your system is stopping a letter from going out, you can train a bot to look at the exceptions, fix them, and then move on again. If it's taking you 10 minutes per loan to clear exceptions, and you can get, let's say, 90-95% of them cleared through a bot, that's a tremendous uplift.
The process and boundaries around adding AI bots
Sarwaikar: We start putting all the rules in place. That will decide the predictive queue of what operations need to be executed. Once you have those queues in place, then the process is about automating those queues using AI bots.
It's always a phased approach. The reason it has to be a phased approach is because AI is a new technology. A servicer has to be comfortable with it. There's always a one to two month testing period where you have the AI bot do it, but you still have a manual review, which is called human reinforced learning.
We are very careful that these bots do not operate independently unless they're working in areas which do not have a borrower impact. If you're making a foreclosure decision, you don't want to go down that path. You want a human involved. We all have seen 2008. We never want to go down that path again. It has to be used in an area that doesn't impact the borrowers adversely. It's focused more on the operational nature of business, which is more repetitive and affects the cost center.
Blair: The package for a foreclosure can be assembled through automation but a person is the one making the decision with everything in front of them. Then you're getting efficiency when it comes to the information gathered, but you're never losing human interaction. If a servicer has a business meeting with an investor, automation can help the investor look at their portfolio and see what is in or out of compliance.
Sarwaikar: Modules for areas like direct servicing (loan administration and escrow and payment processing), default, corporate (compliance and regulatory), and financial control (investor accounting) can be bolted on to existing legacy systems.
Technology built to anticipate borrower needs
Sarwaikar: Areas where borrowers can self-serve significantly reduce call center costs. Automation can answer questions related to basic mortgage knowledge. What is escrow? How do I calculate my payment? How does my payment get processed? Once they log in, it tells them whether their payment has been processed, how much of it was principal or interest and what their next due date is.
The website can predict what the borrower is looking for based on behavior and loan activity for the last 45 days. That reduces the number of clicks a borrower needs to get an answer. That reduces cost and increases the recapture for a mortgage servicing rights portfolio. How much of the MSR pool can be recaptured plays a role into the target yield.
So what we have done is we have covered the entire servicing spectrum with modules that execute the functionality which an AI bot will pick from, queue and use for execution. These modules, once they execute a task, interact with the legacy system as a system of record and keeps it up to date.
Building a bridge between investors and servicers
Sarwaikar: Asset managers and investors, people who are investing and buying loans in bulk or flow, people who are buying these MSR pools where they were keeping the first loss pieces, also have technology needs. They were retaining that in their portfolio, while they were securitizing the rest. Some of them have MSR portfolios and they may have either a captive servicer or have multiple entities managing their loans.
They bought and priced their assets for a certain target yield. The actual yield depends on how well the servicer does. How do they manage default and recapture? It's tough for an asset manager or an investor to manage multiple servicers. Servicers are not allowed to see what price the buyers purchase the loan at, so the question becomes how do you get the maximum yield out of a servicer who shouldn't be knowing the trade economics?
Automation can be mapped to multiple servicers, and portfolio data flows daily to a given asset manager or investor. The asset manager or investor looks at their servicing performance, and compares servicer A with servicer B on recapture and default metrics like claim penalties which represent money lost.
Blair: Parameters can be automated to not allow for modifications to be done that aren't allowed by a particular loan type's guidelines, so that you won't have a buyback, etc.
Use in the government shutdown
Blair: We have the ability to then address that or a natural disaster with a proactive approach, and not just a preventive one.
What I mean by that is that we're able to identify through the ZIP codes what areas are affected, and when that call comes in, direct a message to the borrower that will give them the information so they feel comfortable. It could also be routed to a team that can go further with assisting them.
If you work for the government, you don't know when it's going to open again and you don't know if you get paid, you need guidance. We're saying, 'Here are your options."
