The mortgage retention crisis

Repeat borrowers were supposed to be the easy part of the business.

Every loan officer who’s been around long enough has heard the math. A past customer costs a fraction of what a new one does to acquire. They’re pre-qualified by relationship. They already trust you. They should call back when rates drop, when they need to pull equity, or when the next house comes along.

Except they don’t. For most companies, 70% or more of their clients will use a different lender on their next transaction.

If you’ve been watching the data on mortgage borrower retention over the past two years, you’ve seen the same uncomfortable pattern: top IMBs lost more than $130 billion in repeat-loan opportunities in 2025, according to RETR’s Loan Loss Report. That’s not a marketing problem, a CRM hygiene problem, or even a brand problem, exactly. Somebody else is getting to your past borrowers first, they have better information than you do (and most lenders are simply leaving the door open).

That somebody, in most cases, is whoever now holds the servicing.

The thing nobody talks about when they sell MSRs

Selling mortgage servicing rights makes a lot of sense on a spreadsheet. You get the cash, you simplify operations, you free up capital. Especially useful after a stretch like 2022-2024, when origination volume cratered and a lot of shops needed liquidity badly. That trade has happened thousands of times and nobody’s going to argue with the math in isolation.

The part that doesn’t show up in the MSR sale agreement is what you’re actually transferring. Not just the right to collect a payment. You’re handing over the monthly customer touchpoint, the payoff visibility, the escrow file, the home-equity context, and the entire set of signals that tell you when a borrower is about to do something.

Large servicers figured this out a while ago. Read any of the recent investor decks from the public servicers and you’ll see it stated more or less plainly: servicing portfolios aren’t fee-income assets anymore. They’re customer acquisition engines. Mr. Cooper’s 10-K, Rocket’s investor materials around the Cooper acquisition, the Rithm earnings supplements. They all describe the same flywheel in slightly different language.

The numbers, and what they mean

ICE’s December 2025 Mortgage Monitor pegged overall refinance retention at 28% in Q3 2025. The split by lender type, though, is where it gets interesting:

• Banks: 13% refi retention
• Industry average: 28%
• Nonbank servicers: 35%
• FHA/VA loans: 36%
• GSE loans: 25%
• Portfolio-held: 23%
• Privately securitized: 6%

Note that nonbank servicers have by-far the highest retention of any lending institution type - and the numbers have widened even more significantly with time.

The averages still hide what’s actually happening at the top of the table. Public disclosures from the scaled servicers show recapture performance that bears almost no resemblance to the industry mean:

If you’re a bank holding 13% retention and you’re reading 71% on the loanDepot row, that’s not a small gap to close with better email cadence. That’s a structural advantage they have and you don’t.

Why scaled servicers keep winning

Whoever owns the servicing has a few things you don’t.

They have the payment relationship. Twelve transactions a year, every year, in a category where most consumers think about their lender exactly never. When the borrower needs to remember who they got their mortgage from, the servicer is the answer their statements have been training them to give.

They have the payoff signal. A payoff request is probably the cleanest indicator in mortgage that a borrower is about to do something: sell, refinance, restructure, anything. The originator who sold the servicing finds out about that payoff after the fact, if at all. The servicer sees it in real time, often with enough lead time to make a competing offer before the borrower has fully committed.

They have home-equity context. Property value, UPB, escrow movements, tax bills, insurance increases, tappable equity. The servicer can run all of that against the portfolio every month and identify the borrowers who are sitting on tappable equity or whose payments just spiked because of an insurance shock.

They have scale. A centralized call center, predictive models, automated outreach, next-best-offer workflows, portfolio-wide rate triggers. Once you’re doing this at three million loans, the unit economics make adding the next million almost free. A retail lender with a 20-person retention team can’t run that machine.

And they have timing. They get to act when a borrower becomes refi-eligible, not when the borrower has already submitted an application with somebody else.

This is why Rocket’s Mr. Cooper acquisition was structured the way it was. The combined book will sit at roughly $2.1 trillion across nearly 10 million clients, about one in six U.S. mortgages. Rocket’s investor materials tied the deal economics specifically to applying Rocket’s recapture engine to Cooper’s base and pushing recapture from ~50% to 65%. That 15-point swing across a multi-trillion-dollar book is most of the deal thesis.

Servicing isn’t about servicing fees anymore. It’s about lowering CAC and extending LTV.

The refi cycle isn’t going to be polite about this

ICE reported that first-lien refinances totaled $242 billion in Q1 2026, more than double the year prior, the strongest quarterly print since early 2022. Refis made up nearly 44% of all originations. Rate-and-term refis were 60% of refi activity.

For perspective on borrower behavior: ICE estimated that around 95% of September and October 2025 rate-and-term refis involved 2023–2025 vintage loans, with the average borrower cutting roughly 0.92 percentage points off their rate and saving about $200/month. By Q1 2026 the average rate-and-term refinancer was reducing their payment by $257 through a 97 basis point drop.

The point isn’t that refis are back. It’s that borrowers have gotten faster. They’re more rate-sensitive than they were five years ago, they have more tools to detect payment savings, and they’re more willing to make a move when the math says move. A modest rate decline can flip a million borrowers from inert to actively shopping in a matter of weeks. A home-equity need can turn a “locked-in low-rate forever” borrower into a HELOC or second-lien opportunity overnight.

The lender that detects those signals first gets the first conversation. That part isn’t complicated.

So what does it mean for the rest of us?

If you’re an IMB, the data is already telling you what’s happening. Repeat-borrower leakage isn’t a future risk. It’s costing the top of the industry billions a year right now, and a stronger refi cycle is going to widen that gap before it narrows it. LOs are tough to coach and even harder to manage. The only path to better retention lies in software that can keep you top of mind, uncover opportunities, and drive loyalty at scale.

If you’re a bank, the issue is execution. You have deposits, branches, broader relationships, and most of the time you have trust. But the ICE data shows you retained 13% of your refi borrowers in Q3 2025 against 35% at the nonbanks. The relationships exist; the mortgage-specific machinery to act on them, in most cases, doesn’t. This too drives towards tech as the solution.

If you’re a credit union, your edge is depth of relationship. Your blind spot is that depth doesn’t automatically translate to detection. Member loyalty doesn’t tell you when a member started searching for their next home or when their home’s value crossed the threshold that makes a cash-out attractive.

If you’re a broker, you don’t own the servicing in the first place. If the wholesaler, aggregator, or servicer ends up with better borrower intelligence than you have, the repeat opportunity may never make it back to you unless that partner intentionally routes it.

The retention playbook is different than it was

Here’s the old version: rates drop, you fire up an email campaign to your past book, hope your loan officers remember to call their VIPs, and try to close as many of the inbounds as you can before they shop you.

That doesn’t work anymore. By the time rates have moved enough to trigger your campaign, the servicer-owned lender has already called the borrower, the lead-gen platform has already pinged them through a comparison tool, and a credit trigger buyer is in their inbox.

What works now is continuous-touch and perpetual-insight. The retention playbook starts the day after closing, not the day rates drop. It means monitoring every past customer’s property value, equity position, rate incentive, and engagement signals on an ongoing basis. It means alerting the loan officer when a borrower starts showing intent, not when the borrower has already filled out a form somewhere else. In some cases, the solution may also involve automating communication to limit missed opportunities resulting from LO complacency. 

Here's the part that surprises a lot of executives when they actually look at the math. The cost of monitoring a past-borrower database continuously is small. The cost of not monitoring it, and watching the loanDepots and Rockets of the world recapture your book, is enormous and recurring.

Too many lenders look at the tech cost and say "We can't afford to take that risk". Smart lenders see that the only risk is standing still.

Where Milo fits

This is the problem Milo was built for.

Lenders need a way to stay in front of every past borrower with personalized, white-labeled home value reports, track activity across their book, identify high-intent behavior, and alert loan officers when a past customer is showing signs of being back in the market. As servicing portfolios become recapture engines for the companies that hold them, originators need their own borrower-intelligence layer that keeps them relevant before the borrower raises their hand somewhere else.

The scaled servicers are already proving the model works. Data, timing, and proactive engagement drive recapture. The only question for the rest of the market is whether your next repeat borrower comes back to you, or shows up in somebody else’s portfolio-sourced production report.