Artificial intelligence is moving beyond experimentation and into the operational core of mortgage servicing. What began as targeted applications for document processing and workflow automation is now expanding into areas such as customer engagement, loss mitigation, compliance monitoring, portfolio analytics, and decision support. Now servicers face two challenges: leveraging AI’s efficiency and scalability while maintaining transparency, fairness, and regulatory compliance.
To better understand how the industry is navigating that balance, MortgagePoint spoke with leaders across servicing, technology, and consulting. Their perspectives shine a light on both the opportunities and the complexities ahead.
John Hubbarth, VP, Servicing Business Innovation, Rocket Mortgage
As AI becomes more embedded in servicing operations, how do you evaluate and mitigate model risk when those tools directly influence borrower outcomes, such as loss mitigation or modification eligibility?
John Hubbarth: AI model drift is something we encourage team members to be aware of when implementing this technology in our business. Programs for oversight and principles that keep humans in the loop are important to ensure you are creating the best outcomes for clients.
Qualification for delinquency details is rule-based and predictable, so it’s easier to control unintended results than in some other areas that require more judgment. When AI operates within clearly defined rules and calculations, it’s easier to maintain consistency and oversight while reducing the risk of unintended outcomes over time. The biggest benefit of AI in this space we see is the scale and automation potential with document extraction, matching clients to the best possible products that will help them succeed, and eliminating barriers for them to achieve the assistance they need. A multi-model approach—agents quality controlling agents and ensuring team member oversight at critical control points—is an example of a framework that can properly address risks.
What challenges do you face in ensuring that rapidly evolving servicing technologies remain aligned with investor and agency expectations around transparency and fair servicing practices?
Hubbarth: Rocket Mortgage is focused on ensuring that new technologies enhance the customer experience without changing the standards of fairness, transparency, or compliance. AI is only as effective as the data, rules, and controls that support it.
That’s why transparency, testing, ongoing monitoring, and human oversight remain critical. Whether a client is interacting with a team member or an AI-powered tool, the same servicing guidelines, eligibility requirements, and regulatory standards still apply.
Where do you see the greatest tension between automating servicing workflows through AI and maintaining a human-centered borrower experience?
Hubbarth: It’s important to ensure clients can work with a human when they want to. The challenge is making sure staffing and capacity models account for that accurately, and your technology facilitates it well. The opportunity we have is making the agentic experience so good that clients love it and want to use it. They can achieve the outcomes they want, whenever they want, and using whatever tool they like.
There are always going to be clients who want team member connection, and we want that to be available to them at the right time. Yes, AI can deliver great efficiency when done correctly, but our north star will always be client satisfaction.
Bhupinder Singh, SVP, Head of Product & Operations, BSI Financial Services
How is BSI Financial Services incorporating AI and automation into its servicing operations today, and which workflows have seen the most measurable efficiency or accuracy gains?
Bhupinder Singh: The honest answer is, across the entire servicing lifecycle. BSI’s AI and automation deployment isn’t concentrated in one function—it follows the loan from boarding through default resolution, with purpose-built tools at each stage delivering measurable impact.
Loan Boarding
Loan Boarding EZ Board™, our OCR-enabled boarding platform, automates document ingestion and validation at the point of onboarding—extracting key data fields, flagging inconsistencies, and dramatically reducing exceptions before a loan enters our servicing system. This is further powered by AI-driven Doc Indexing and Extraction: AI-powered document processing and data extraction streamlines the loan application process by pulling relevant information from various documents, reducing manual effort and improving accuracy—with 50% straight-through processing already achieved. The result is a cleaner data foundation from day one and significantly fewer downstream errors.
Customer Care
Milo, our agentic AI and automation-powered chat and voice platform, achieves a 28% containment rate, deflecting a significant volume of live agent calls with 24/7 availability.
It works alongside MyLoanWeb™, our self-service borrower portal, to give borrowers multiple channels for account access and issue resolution without agent involvement. Call Listening and QA AI and automation now monitors calls for quality, compliance flags, and coaching opportunities—expanding coverage to 100% of calls, with results available in minutes.
Collections & Early Delinquency
AI and automation-driven propensity models identify at-risk borrowers early in the delinquency cycle—segmenting the portfolio by risk profile and timing to optimize Right Party Contact, driving a 50% increase in Right Party Contact rates. This allows us to intervene early and precisely, rather than applying uniform collection strategies across the portfolio.
Default Management & Loss Mitigation
The Loss Mitigation Underwriter Co-pilot automates our LM workflow—targeting 80% of all loans processing straight-through with full compliance decisioning. AI and automation also power our gold standard ops manual for Loss Mitigation—automatically generating SOPs with screenshots from agent workflows, linked directly to our servicing guide and compliance controls, creating full traceability from process to compliance rule.
Portfolio Analytics & Governance
Asset360™ delivers role-based, life-of-loan visibility across the portfolio through real-time dashboards—giving operations, risk, and executive teams a single view of performance without navigating multiple systems. Real-time Default Dashboards drive queue management and daily operations.
Libretto™, our automated compliance oversight platform with 1,400+ rules, governs the entire portfolio—running in real time against loan data, flagging exceptions, and ensuring regulatory alignment without waiting for periodic manual review. A full AI and automation governance framework—including policy, model inventory, and auditability infrastructure—ensures every tool operates within a structure designed for compliance and scale.
Taken together, these tools represent a coherent AI and automation strategy—each deployment tied to a specific operational problem, measurable outcomes, and a governance framework designed to scale responsibly.
What are the biggest technical or organizational challenges in modernizing servicing infrastructure, and how are you balancing legacy system constraints with the need for real-time, data-driven decisioning?
Singh: Legacy infrastructure remains the defining constraint for most servicers, and BSI has approached that challenge methodically. Rather than betting on wholesale platform replacement, we’ve invested in proprietary tools that layer modern capabilities on top of existing systems.
EZ Board™ removes friction at onboarding through OCR-enabled automation; Asset360™ delivers real-time portfolio visibility without requiring a core system replacement. AI and automation accelerated our data modernization meaningfully—reverse-engineering legacy pipelines compressed an 18–24 month program by approximately six months.
However, if I’m being honest, the harder challenge isn’t technical—it’s human. The fear of job displacement is real, and it’s present in every organization navigating AI and automation at scale. We’ve found that ignoring it or minimizing it only deepens resistance.
Our approach has been to address it directly: we’ve been transparent with our teams about where AI and automation are being deployed, what it’s designed to do, and—critically—what it is not designed to do. In every case, our goal has been to eliminate the repetitive, low-value tasks that drain people’s time and energy and redirect that capacity toward higher-judgment work that machines can’t replicate—borrower relationships, complex problem-solving, quality oversight.
We’ve backed that commitment with investment. The full organization has been trained on AI and automation use cases, Microsoft Copilot has been deployed to all staff, and we’ve stood up a leadership cohort specifically focused on AI and automation fluency. We’ve also built an AI and automation governance framework—with model inventory and auditability—so that employees and regulators alike can see how these tools work and where humans remain in the loop. The organizations that will navigate this well are those that bring their people along rather than leaving them to wonder what’s coming next.
How is technology reshaping the borrower experience in mortgage servicing?
Singh: MyLoanWeb™ established our foundation—a self-service portal giving borrowers direct access to their loan data, payment tools, and account management, reducing call volume while improving satisfaction. Milo takes that a step further—think of it as a personal concierge for every borrower, available 24/7 via chat and voice, handling inquiries, guiding borrowers through options, and resolving issues with the speed and consistency no
human team could match at scale. With a 28% containment rate, Milo is already deflecting a significant share of live agent contacts while delivering a more immediate, personalized experience.
The more important shift, though, is making the experience proactive: Portfolio GuardianSM identifies at-risk borrowers before they reach a delinquency trigger, enabling outreach with the right options at the right time. Libretto™ ensures every borrower-facing interaction is governed by 1,400+ compliance rules, delivering consistent treatment regardless of loan type. And Call QA AI and automation expanding to 100% of calls—with results in minutes—continuously raises the quality of every live interaction. The goal is a borrower experience that feels personalized and proactive at scale.
Servicing data often sits across multiple systems and vendors. How is BSI approaching data unification and analytics to create a more complete, real-time view of portfolio performance?
Singh: Asset360™ is our most direct answer—role-based dashboards providing life-of-loan visibility across the portfolio from a single interface. Underneath it, our Databricks platform consolidates Loan and Customer 360 data across core servicing and all lending source systems into unified data marts, with AI and automation shaving ~six months off the build timeline. Real-time Default Dashboards drive queue management and daily operations, while a Lending Ops SLA and Efficiency Dashboard tracks pipeline performance live. Libretto™ adds a compliance dimension—its 1,400+ rules run continuously against portfolio data, flagging exceptions in real time rather than through periodic manual review. We’re continuing to expand the Databricks consolidation and extend our BI reporting layer for broader AI and automation consumption across the business.
Looking ahead, what emerging technologies do you believe will most transform servicing products and operations over the next 3–5 years?
Singh: Three areas stand out—and we’re already building toward all three. First, agentic AI and automation will reshape how servicing operations are structured. Milo is our proof point today; we’re expanding it to Lending and Non-Agency loans and building agentic capabilities for compliance and operational risk identification. MyLoanWeb™ will evolve from self-service toward intelligent borrower guidance.
Second, AI and automation-assisted compliance will move from differentiator to baseline expectation. Libretto™ already automates oversight across 1,400+ rules; the next frontier is real-time AI and automation monitoring so gaps are caught and remediated in the moment.
Call QA expanding to 100% coverage is a direct expression of this—those economics only work with AI and automation.
Third, predictive and prescriptive analytics will move servicers from reactive to anticipatory. Portfolio GuardianSM already demonstrates this at scale. The next evolution is closing the loop—predictions automatically triggering the right workflow through Asset360™ without human routing. Servicers who build this integrated intelligence infrastructure now will be operating at a fundamentally different cost and quality curve than those who wait.
Andrew Weiss, Partner, Managing Director, Technology Consulting Practice, BlackFin Group
Both Fannie Mae and Freddie Mac have recently issued AI oversight and governance requirements for lenders. From your perspective, how significant is this moment for the mortgage industry’s broader adoption of AI?
Andrew Weiss: Well, “significant” is the interesting word in your question, right? Fannie and Freddie have always been focused on governance to some degree. It’s their job, and it’s a valuable thing. I think the question is, to what degree are their requirements for governance something that can easily be done by the lenders, and will they enforce it in any particular way? It was suggested; it wasn’t mandated.
They were good suggestions in general. They were the kind of suggestions that I think larger lenders probably have already been thinking about and probably have some structures in place that they can fit that into. And the smaller lenders will struggle to keep up with that. This has been a pattern common to many of the governance things that Fannie and Freddie have put forward. It was generally good guidance, but it’s a little bit of “make it work safely, and we’ll decide later what to do about it.” I don’t mean to be dismissive of what it was, because I think it was generally positive, but was it very actionable? I don’t know. It’s still early days.
A recent Blackfin article describes these agency AI policies as arriving “like lightning bolts.” Why do you think the GSEs moved now, and what risks were they trying to get ahead of?
Weiss: I think now is the time to move because it’s just so pervasive in our industry, in every industry. They could no longer just sit on the sidelines. For many lenders, particularly mid-sized or smaller lenders, it’s going to be provocative, because while you can’t cross the street corner without seeing someone telling you something about AI, I don’t know that governance of AI was something that a lot of people in the smaller to mid-sized companies were thinking seriously about. So, that’s where the “lightning bolt” notion comes from.
We were trying to be a little bit more provocative in the sense that we think people need to be thinking about this and using the agencies as a way to help do that. Mortgage industry professionals are trying to get ahead of the risk of the unknown.
You can go back to 2010, for some of us have been in this industry long enough, where the Agencies were caught a little bit by surprise in terms of some of the risks, and they’re trying to make sure that they protect themselves against that unknown now. Clearly, there are some things, particularly around decisioning, where AI can be quite risky. The fact that it can’t necessarily explain its answers yet is a generalized risk. I think they’re trying to get ahead of things like that.
Many lenders are experimenting with AI tools across origination, servicing, compliance, and customer engagement. Where do you believe AI is already delivering measurable operational value in mortgage banking today?
Weiss: It’s mostly still preliminary. One of the earliest applications of AI was around document recognition, and it’s doing that well. There’s a lot of documented evidence of efficiency benefits in that space. It’s the simplest, and it’s a very bounded space. Everything else is pretty preliminary.
One of the Blackfin articles’ major themes is governance. What does a strong AI governance framework look like inside a mortgage lender, especially for institutions that may not yet have dedicated AI expertise?
Weiss: There are probably three different parts to that. It’s a challenging role because you need to meld traditional governance and compliance thinking with a real understanding of the technology and what it can do, as well as both potential benefits and risks. The first step is getting people in the room, a collection of people who can talk across those different disciplines.
The other thing that’s underrepresented in terms of the conversation around governance is understanding the data around which AI is being trained. Data stewardship, if you will. That is maybe the most important thing.
There’s been AI in this industry for a long time. When I was at Fannie Mae, I ran a group called Advanced Technologies. We brought Desktop Underwriter to the market back in the mid-‘90s. So, I’ve been thinking about this for a very long time. And the problem with using what we call AI today in decisioning is that AI today is just looking at all sorts of data, inferring past relationships, and then bringing those forward.
So, if you train your model against data that’s filled with, in the most extreme negative sense, a bunch of redlining loans, well, guess what? You’re not going to get good decisions moving forward. Understanding the data set you’re using to train [is critical.] Even in customer service, if you’re using it there, are you taking your best customer service agents and calculating their way of dealing with customers and using that to train your AI? The other issue, which is really hard, is that things are moving so fast. It’s a full-time job just keeping up with it. Every vendor in the world is out there saying it’s the best thing since sliced bread. And I’m not saying there are no benefits, but discerning the difference between hype and reality becomes a full-time job.
The article raises questions about ownership between technology teams and compliance teams. In your experience, who should ultimately “own” AI governance inside a lender organization?
Weiss: The problem is, if you just give it to IT, while they’re probably going to be most creative about the potential risks and benefits, they don’t have the compliance DNA that it takes to implement safely. If you just give it to compliance, you’re going to lose a lot of the possibilities and the nimbleness.
You need to figure out how to get a partnership between the two. It’s a three-legged stool, because you need the business in there as well. You need production. So, if you have compliance, production, and IT with equal votes at the table, probably anything that comes out of that is going to be both well-managed and useful to the organization.
How are lenders balancing the pressure to innovate quickly with the need to satisfy regulatory expectations around explainability, auditability, and risk management?
Weiss: That’s going to be different depending on the size of your organization. For small organizations, that balance is going to be about the interpersonal relationships and the executive management’s willingness to take risks, and manage that risk well. It’s a mindset. With larger organizations, it’s going to be more about that organizational dynamic. Who owns this? Whose butt is going to be in the sling if I say yes?
Third-party vendor oversight is becoming increasingly important as lenders adopt outside AI solutions. What should mortgage companies be asking their technology vendors right now that they may not have asked a year ago?
Weiss: Right now, they have to ask their tech vendors about their governance and oversight. They need to ask about not just promises but anything demonstrable. They need to create a partnership, as opposed to a vendor-customer relationship.
That’s something that has been elusive for many but is certainly achievable. We all can understand that this is a very rapidly developing field. As a customer, you have to be willing to take some risks, too. That’s something that a lot of lenders, as customers, don’t want to do. They want all the risk to be on the vendor side. You have to recognize that it’s both a shared risk and a shared responsibility, so you need to be able to think that all the way through. This is an area where, in some ways, smaller customers have a real advantage.
Which mortgage workflows do you think are most likely to see meaningful transformation over the next two to three years?
Weiss: Certainly, in anything that’s customer service-related, you can see a lot of benefits. Not everyone is going to want to deal with a robot, but the truth is that if you talk to any customer service operation, it’ll find that some high percentage of the questions are always the same. They can be answered very efficiently in that way. Now, the trick is to build the AI so it knows when it’s no longer in its bounds. That has not always been done well, but I think that there’s a huge potential for benefit there. There are both benefits and risks on the upfront prospecting and selling side of the equation. This is where the risks might be almost the highest, because, depending on how you’ve trained the agents and LOs to interact with both customers and technology there can be great benefits that result in better pull through or there can be significant compliance risks. You don’t want them effectively selling something that’s outside of the bounds of what should be sold.
And, unfortunately, it is quite possible for those kinds of scenarios with some of the AI, particularly unless you’ve been careful about the data you’ve trained it on. It’s sort of the mirror in some sense to the customer service example, where there’s a lot of interaction, a lot of the questions are the same, a lot of the opportunities are there, and there could be real efficiencies for LOs to reach more people with better information than they can just sitting on the phone all the time. Those are real possibilities. Inside the workflow, there’s a lot of grunt work and, if you’ll pardon my expression, a lot of dumb stuff it takes to manufacture a mortgage. It shouldn’t be as hard as we have made it, as an industry. From my vantage point, these are constrained problems with constrained answers, and you can know if it’s doing it right or not. It’s not magic.
So, some of the decisioning tasks are what I would leave for last. There is a place in this world for rules, as opposed to machine learning.
This is where the definitions around “what is an AI?” (or not) start to be important. Anything where someone wants to be able to know why I got to that answer, that’s something that we have to be very careful about how we use a machine learning model to get to. It doesn’t have to be a compliance-oriented thing.
The blog warns that inadequate AI controls could eventually contribute to repurchase risk or liability exposure. Do you think lenders fully appreciate that possibility yet, or is the industry still underestimating the compliance implications?
Weiss: I think it’s starting. The Fannie and Freddie guidelines, that’s where they’re positioning themselves to be able to push back loans that they felt used too much AI in a way they don’t approve of, without saying exactly what that is.
We’re at the beginning of that conversation. It’s going to take a lot more specificity. We’re not there yet, but we’re going to have to be there at the pace that AI technology and its application is changing.
Looking ahead, do you expect AI in mortgage lending to evolve more as a productivity tool that assists human decision-makers, or are we moving toward a future where AI takes on more autonomous operational responsibilities?
Weiss: In our industry, if you’re a betting man, you bet on the former. It’s just a boost to efficiency. There are real opportunities in certain areas of our industry, particularly in servicing. It’s just a “crank them out” kind of business. Our industry, historically, has been very slow to adopt those levels of change, but they’re out there if people want to do it.
The question is, who’s going to step forward, take those risks, and see if the rest of the industry is going to follow? It’s challenging because in this market, there are very few lenders who have some combination of the capital lying around to invest and the willingness to take those kinds of risks. And you’re going to need both in the near term to take those leaps.
