Three forces at once
If you run a retail mortgage lender, three forces are reshaping your business at the same time. Rates dipped below 6% early this year and gave lenders their biggest refinance quarter since 2022 (Mortgage News Daily). That window has closed. Freddie Mac's 30-year average hit 7.03% on September 24, its highest reading of 2026 (Freddie Mac), with the 10-year Treasury above 5% (Mortgage Daily).
Affordability was stretched before the spike. Harvard's Joint Center for Housing Studies put the payment on a median-priced home at $3,100 a month in late 2025, requiring more than $120,000 of income, up from $66,000 in 2020 (JCHS). That leaves a smaller purchase market full of borrowers who shop hard on price.
The third force matters most over the long run: AI is cutting the time and cost to originate a loan, and the cuts are accelerating. Freddie Mac says lenders that fully use automation in Loan Product Advisor originate loans $1,500 cheaper, about 14%, with production cycles five days shorter (Freddie Mac). Rocket says loans closed per team member are up 75% from two years ago (Yahoo Finance), and its days to close ran under half the industry's 45-day average in March (Rocket Q1 call).
Those gains are early. Personnel makes up roughly two-thirds of production cost (Ocrolus), and software handles more of that work every quarter: the document chase, income calculation, condition clearing, borrower follow-up. The industry's average cost per loan still sits near record highs, so the savings are landing at the few lenders who rebuilt their workflows. They will turn that cost gap into lower prices, and a retail shop still paying for a full human touch at every step will compete against lenders whose cost per loan keeps falling.
Any one of these would change how you operate. Together they force a choice. I spend my days working with lenders on retention and recapture, and I see four paths that hold up:
- 1Retain and buy servicing.
- 2Build an external acquisition flywheel you own.
- 3Acquire smaller retail lenders to reach scale.
- 4Roll up into a larger lender.
Each path answers the same question: where will your next thousand loans come from, and what will each one cost you? A mid-size lender that can't answer it will spend the next two years funding a cost structure its volume can't support.
1. Retain servicing, and buy more of it
Servicing is the one asset in mortgage that gains value when rates rise and hands you leads when they fall. For a retail lender whose production revenue swings with every move in the 10-year, that makes it the natural hedge.
The hedge runs through prepayment speeds. When rates climb, borrowers stop refinancing, loans stay on the books longer, and MSR values rise while production shrinks (MCT). With the 10-year above 5%, a lender holding servicing has a cushion that a release-only shop lacks.
The second half of the case is recapture, and Rocket has shown the industry what it's worth. After closing Mr. Cooper, Rocket grew purchase share from 5.5% to 6.2% and refinance share from 12.2% to 14.3% in the first half of 2026, with legacy Mr. Cooper recapture rates hitting a record in Q2 (HousingWire). The telling detail: when Rocket sold $53 billion of MSRs that quarter, it kept subservicing and recapture rights on nearly 80% of the loans. Rocket sold the cash flow and kept the customer.
The accountants are catching up. In March 2026, FASB's Emerging Issues Task Force voted 9 to 1 that an MSR's fair value should include the effects of recapture (Deloitte). Buyers already pay for recapture in MSR prices. Soon the balance sheet will show it.
Your servicing is worth what your recapture rate says
Most servicers still recapture poorly. ICE found servicers kept 32% of refinancing borrowers in Q1 2026, down from 35% the prior quarter, and that counts as one of the best readings in a decade (Mortgage News Daily). Two of every three borrowers still walk. Purchase moves are harder: a borrower can list the house, get pre-approved elsewhere, and close before anyone on the servicing side notices.
So holding servicing without a recapture engine means holding a lead list that shrinks every month. Buying servicing only pencils if you recapture better than the rate baked into the price you paid.
Recapture comes down to knowing a borrower is back in the market before a competitor does. The signals exist: a credit pull by another lender, a listing, a marriage or new baby, high-interest debt crossing a threshold, home-search activity. Congress just made those signals more valuable to servicers. The Homebuyers Privacy Protection Act took effect March 5, 2026 and banned mortgage trigger leads except for lenders with an existing relationship, such as the borrower's current servicer, or with consumer opt-in (NAMB). Conquest lenders lost that feed. Your servicing book kept it.
The flywheel, and where the margin comes from
The biggest payoff from recapture is who originates the next loan. When your servicing book generates the lead, you don't need a loan officer earning 75 to 150 basis points to close it. A salaried retention team on a small per-loan bonus can do the work: consumer-direct shops pay around 40 basis points, and bank and credit union LOs often earn a salary plus 5 to 10 (STRATMOR). LO commission is the largest variable cost in retail lending. On a $350,000 loan, dropping from 100 basis points to 10 saves about $3,150, more than three times the industry's average production profit of $973 per loan (MBA).
Stack the savings and a recaptured loan becomes your highest-margin product: no acquisition cost, a low-commission originator, and an AI-driven cost to close that keeps shrinking. It does force a comp decision, because retail LOs will argue their past clients belong to them. Retain servicing on the new loan and you reset the clock on that relationship, so the portfolio feeds itself. Over five to ten years, that loop becomes the cheapest origination channel you own.
The cash flow tradeoff
The drawback of holding servicing is timing. Sell servicing released and the buyer pays you a premium at funding. Retain it and you trade that cash today for a stream of fees, about 25 basis points a year on a conventional loan, plus the recapture loans that stream produces later. You get less cash now and more over time. For many retail lenders, that upfront premium is what's keeping them alive right now, so the switch has to be financed.
That's why this path fits two kinds of lenders: those with enough volume to carry a thinner near-term cash position, and those cutting costs hard enough to make money on a loan without the premium. Everyone else risks running out of liquidity before the flywheel turns.
Get the cash flow right and the business builds on itself. Each retained loan adds fee income and a future recapture lead, and each recaptured loan adds another. The book keeps paying in years when production doesn't, which makes it the closest thing mortgage has to a future-proof model. That's the bet behind the biggest deals of the past year. Rocket paid $14.2 billion for Mr. Cooper (NMP), CrossCountry bought Two Harbors and its servicing platform, and servicers and lenders across the industry are adding MSRs while others sell.
Beyond cash flow, you need agency approvals, a subservicer or servicing platform, and hedging discipline, because MSR values drop when rates fall. That drop is exactly when your recapture engine has to earn it back. If you can't fund a full servicing strategy yet, retain a slice of production and build recapture muscle on that book first.
2. Build an acquisition engine you control
Retail has always sourced loans through two pipes: the real estate agent and the loan officer's personal network. Both pipes are narrowing, and a lender that depends on them needs a repeatable way to buy customers at a known cost.
Why the old pipes are narrowing
AI undercuts the loan officer from the cost side. When software handles the application, the document chase, the income math, and status updates, a borrower no longer needs a commissioned LO shepherding every step. Rocket's LOs already serve nearly 40% more clients than a year ago using AI pipeline tools (HousingWire). A lender with that productivity can price below a retail shop paying full comp on every file, and rate shoppers follow the price. Borrowers are doing more of the legwork themselves too: 45% of prospective buyers used AI tools in their home search in Q2 2026, up from 37% a year earlier (Veterans United).
The agent referral is under pressure from the other side. Brokerages and portals are wiring mortgage into the transaction. Redfin agents now send 47% of their buyers to Rocket, closing in on a 50% target, and Rocket Pro offers Compass-specific pricing that has produced more than $2 billion in locks (HousingWire). If your LO's best agent works for a brokerage with its own lender partner, that relationship now has a competitor inside the building.
The paid channels, and what they cost
The replacement is demand you generate or buy: Google search, Meta, TikTok, Reddit, lead marketplaces, and your own content ranking in AI answers. Every one of those runs on an auction, and the auctions got pricier this year. When the trigger lead ban removed the cheapest data source in March, lenders moved that budget into bidding. Cotality estimated internet lead costs on some platforms rose about 45% year over year (HousingWire).
That puts the whole game in conversion. At $100 a lead and a 1% close rate, you pay $10,000 per funded loan before any production cost. At 3%, you pay about $3,300. The lenders who win here answer inside five minutes, nurture unready borrowers for months, and compensate LOs for working assigned leads at lower basis points than self-sourced loans.
Pick a segment you can win
The best acquisition engines go narrow. Society Mortgage, a Fort Lauderdale lender approved for FHA, VA, and USDA lending (Society Mortgage), runs a profitable model buying leads for lower-FICO FHA purchase borrowers. Many lenders avoid that segment, so the leads cost less. Those borrowers need patient coaching, which suits a centralized team and a long nurture cycle. Society built its underwriting, staffing, and follow-up around one borrower and one product.
AI makes this path more viable each quarter. As fulfillment cost falls, acquisition becomes the biggest cost you still control, and every dollar AI saves in processing is a dollar you can bid for a customer. The lender with the lowest cost to originate can outbid everyone else in the same Google and Meta auctions and still make money.
Paths 1 and 2 compound
Paid acquisition gets you the first loan at a known cost. Retention makes the second loan close to free. A lender that buys a borrower for $4,000 and recaptures that borrower twice over 15 years has a very different customer lifetime value than one that loses the borrower at payoff. Paths 1 and 2 are the two halves of a direct-to-consumer business, and the best operators run both.
This path requires a marketing analytics function, a sales floor built for speed, and the stomach to lose money for six to twelve months while you find the segment and channel that work.
3. Buy smaller lenders and spread your fixed costs
A retail lender carries a stack of costs that stay flat when volume drops: the LOS and pricing engine, compliance and QC, secondary and hedging, finance, licensing, and the executive team. The cure is more loans running through the same platform. In my view, the break point sits around $3 billion a year in a weak market and $5 billion or more in a strong one.
Cost per loan now decides who makes money. MBA's Q2 2026 report put average production expense at $10,936 per loan, against a $7,945 average since 2008, while the typical lender cleared $973 of profit per loan (Scotsman Guide). Run $1,000 per loan above average and you made nothing. In 2025, MBA's top fifth of lenders earned 115 basis points on production while the bottom fifth lost 64, and the top group spent about $2,500 less per loan (MBA).
Figure 02 shows why volume matters. It assumes $8 million a year of platform costs that don't flex with volume and a $350,000 average loan.
At $1 billion, fixed costs alone run nearly three times the industry's average profit per loan. At $5 billion, they drop to about half of it.
Why the bar moves with the market
Think of the threshold as market share. $3 billion in a $2 trillion origination year and $5 billion in a $3.3 trillion year both equal about 0.15% of national volume. A lender with the LO count and branch footprint to hold that share clears the bar in either environment. Good years also inflate costs, as lenders hire operations staff and bid up LO comp, so the bar rises with the market.
Most lenders sit below it. The average firm in MBA's sample produced $689 million in Q2 2026, its best quarter since 2022 (MBA). Annualize that strong spring quarter and you still get $2.8 billion.
AI raises the stakes because it turns variable cost into fixed cost. Most production cost today is people, paid per file or per hour. As software takes over processing, underwriting prep, and borrower follow-up, more of your cost becomes engineering, data, and licenses that cost about the same whether you spread them over 3,000 loans or 30,000. A scaled lender gets more savings per loan from every dollar it spends on AI, so the gap between it and smaller shops will keep widening.
The buyers are already shopping
The number of nondepository originators fell by about 100 over two years, while their share of total volume rose to 67.5% (HousingWire). NFM Lending reviewed roughly two dozen deals, most of them asset purchases, before buying Homespire Mortgage. Union Home Mortgage took over AmeriTrust's retail branches and origination assets in July (Scotsman Guide). CrossCountry Mortgage closed its purchase of Two Harbors and its RoundPoint servicing platform in August, running Paths 1 and 3 in a single deal.
How to buy well
- Buy producers and branches rather than balance sheets. An asset purchase brings over the LOs and pipeline without the leases and legacy repurchase exposure.
- Move everyone onto one LOS, one pricing engine, and one compliance and secondary team within a year. The savings come from shutting down duplicate platforms; run two of everything and you bought revenue while keeping both cost bases.
- Pay to keep the top producers, and screen hard for culture. NFM's CEO called cultural alignment the deciding factor.
- Put the acquired past-client database into your retention program on day one. It is often the most underpriced asset in the deal.
The risks are LO attrition after close, integration costs that run over budget, and paying for volume that walks out the door.
4. Roll up into a larger lender
If you can't reach scale, you can join it. The math from Path 3 works from the other side of the table: a bigger platform absorbs your fixed costs, and your LOs keep producing on cheaper rails.
Guild showed this can be a growth move. Guild, a pure distributed retail lender that originated about $5 billion in Q1 2025, sold to Bayview Asset Management, then the second-largest servicer in the country with a $770 billion portfolio (HousingWire). The deal closed 45 days after talks began. Bayview gained an origination engine to keep its servicing customers, and Guild kept its brand, its management team, and access to far more capital. Guild's CEO cited recapture rates of about 60% across its last three refinance surges. Bayview paid for that recapture capability as much as for the volume.
Who should consider it
- Lenders whose cost per loan runs above the MBA average, with no credible plan to reach $3 billion.
- Lenders that release all servicing and can't fund a book of their own.
- Founders facing succession, or companies whose warehouse lenders are tightening terms. Advisors told HousingWire that some smaller operators are selling out of necessity after years of pressure from capital providers.
The pressure has reached larger shops too. NFM's Rick Roque pointed to misallocated capital, missing AI strategies, leadership succession, and the fading protective value of a regional footprint as reasons lenders of every size now take deal calls.
Sell from strength
IMBs have now posted five straight quarters of production profit (MBA). That streak is under threat. Rocket guided its Q3 revenue below Q2, calling it a quarter smaller than the second, a pattern the industry hasn't seen since 2022 (HousingWire). Your trailing twelve months look as good as they will for a while, and a buyer will price your next four quarters.
The pool of buyers is wider than it was. Servicers want retail origination to protect their books. Private equity wants platforms. Proposed changes to bank capital rules could pull banks back into servicing (Federal Reserve), and a bank that wants servicing also needs originators.
What raises your price
Buyers pay for productive LOs with durable referral relationships, clean QC and compliance history, and above all a customer base that comes back. A lender that can prove its past clients return, with recapture numbers by year, is selling a repeat-customer franchise. A lender that can't is selling a group of LOs who may leave after the earnout.
Picking none is still a choice
The lender most at risk runs $1 billion to $2 billion a year, releases its servicing, sources through agents and LO networks, and carries a cost base built for a bigger market and a pre-AI workflow. That lender made money this spring. At 7% rates, it will spend the next several quarters waiting for a refinance wave while the companies running Paths 1 through 4 recruit its best LOs.
The paths combine. Rocket runs servicing, direct-to-consumer acquisition, and acquisitions at national scale. CrossCountry paired retail origination with a servicing platform in a single deal. A mid-size lender can pick a primary path and a supporting one, such as retaining servicing while building one narrow paid channel.
One asset shows up in all four. Your past-client base feeds recapture in Path 1 and lowers your blended acquisition cost in Path 2. It gives acquired databases somewhere to land in Path 3, and it raises what a buyer pays in Path 4. Most lenders let that base go quiet the day the loan funds.
Here is a test for your leadership team this quarter: what does your next loan cost to acquire, and who owns the customer after it closes? If the answers are "we don't know" and "whoever calls them next," pick a path.
Andrew leads Milo, the mortgage retention platform that monitors past borrowers across 2,000+ sites and alerts lenders the moment they’re back in the market.