AI’s Second Job: Helping Mortgage Lenders Grow

John Paasonen is co-founder and CEO of Maxwell.

I was sitting with the CEO of the mortgage division of a large bank recently. His team had just spent the better part of a day with a large consulting firm talking about AI. So I asked him: “What was your biggest takeaway?”

“Cost,” he said. “It was all about driving down costs.”

Then he paused.

“That’s great. We’ve been working on our costs for years. But I sat here wondering: how can AI actually help me grow my mortgage business?”

I haven’t stopped thinking about that question.

For the last four years, mortgage leaders have spent enormous energy on the cost side of the business, and understandably so. Volumes fell. Margins compressed. Excess capacity had to come out. MBA’s own Q2 2026 data show production expenses still running at $10,936 per loan, well above the historical average of $7,945.

Every mortgage executive I know has spent serious time figuring out how to take another dollar out of the cost to originate a loan.

AI is already helping. It automates work that shouldn’t require a human and can make loan officers, processors, underwriters and closers considerably more productive.

But there’s a limit to what you can accomplish through efficiency alone. Eventually, a business has to grow.

That’s the second job for AI that now deserves more attention. Can it help a lender become more effective at finding, winning and retaining business?

I think there are at least three places to start.

Start With the Opportunities Already in Front of You

Most lenders already have hard-earned sources of demand: real estate agent relationships, referral partners, past customers, bank customers, direct-to-consumer marketing and the leads their loan officers generate every day.

Before spending more money creating demand, I’d look hard at how much more growth you can get from the opportunities already coming in.

Think about where a good opportunity gets lost. An agent sends over a buyer, but the LO doesn’t connect quickly enough. A borrower gets 70% through an application and stalls. Someone explores products and pricing, then quietly starts shopping somewhere else.

Small improvements are worth real money.

Take a lender with 10,000 qualified opportunities that converts 10% into funded loans. Improve conversion by one percentage point and you’ve created 100 additional loans. At MBA’s Q2 average first-mortgage balance of roughly $386,000, that represents nearly $39 million in additional funded volume.

This is where I think AI gets genuinely useful.

Most lenders don’t have an opportunity problem as much as an attention problem. Which borrower needs a call today? Who stalled halfway through an application? Which agent relationship has quietly gone cold? Which borrower is looking at a product that may not actually fit what they are trying to accomplish?

Give a good loan officer that context at the right moment and let them do what a good loan officer does.

Before spending another dollar finding the next thousand leads, it’s worth asking what technology could help you do with the thousand opportunities you already have.

Make Good Producers Better

Mortgage CEOs already know one way to grow: recruit more loan officers.

There’s another lever sitting right in front of us. Help the producers you already have win more business.

Our industry spends enormous amounts trying to make sellers incrementally better: coaches, conferences, scripts, playbooks, training programs and sales managers. We do it because even modest improvements in producer effectiveness create tremendous value.

And selling a mortgage isn’t getting simpler.

HomeLight recently found that a third of prospective borrowers are asking about down-payment assistance or lower-down-payment programs, while 45% of lenders say they are often or very frequently asked about zero-down mortgages. Basic misconceptions remain common, from believing you need 20% down to waiting for rates to return to 3% or 4%.

The best loan officers navigate that complexity exceptionally well. They ask better questions. They understand what the borrower is really trying to solve. They explain tradeoffs and help someone see a path forward.

AI can make them better at doing that.

We’re already beginning to see the economics. Rocket has reported that loan officers using its AI-powered tools are handling nearly 40% more clients than a year earlier. loanDepot recently reported an 18% increase in funded loan units per loan officer.

It’s still early, and I wouldn’t attribute every productivity gain at either company to AI. But those are the kinds of numbers I’m interested in.

There’s a recruiting implication, too. One of our lender clients tells top-producing teams they will close more loans and ultimately earn more at their company than somewhere else. Their average LO is closing 6.8 loans per month in 2026.

That pitch gets a lot more compelling if your technology can actually help a good producer become a great one.

Don’t Let the Customer Relationship Go Quiet

There’s another growth asset most lenders already own: the customers they spent years and a lot of money acquiring.

That might mean a servicing portfolio, a bank’s existing customer base or years of past originations. You’ve already earned some level of trust with these people.

Then, too often, the relationship goes quiet.

No team of humans can continuously watch hundreds of thousands of relationships and know when one becomes relevant again. A homeowner may not need another 30-year mortgage today. Six months from now she might benefit from a HELOC. She might be preparing to move. Rates might change enough to create a refinance opportunity. Her property value or financial situation might have changed.

This matters because the acquisition cost has largely already been paid.

Rocket said existing servicing customers accounted for 57% of its Q2 refinance closed volume. PennyMac reported its government first-lien refinance recapture rate reaching 59%. loanDepot reported a 68% recapture rate across a servicing portfolio of roughly 465,000 customers.

Those numbers should get the attention of any lender with an existing customer base.

At Maxwell, we’re working on a version of this problem ourselves with our partners at Ardley. We’re exploring how a lender can analyze servicing portfolios and historical originations alongside current property and market data to identify customers who may have a new financing need, then get those opportunities back in front of the lender.

The specific implementation will vary by institution. The broader idea is pretty simple: there is economic value sitting inside relationships you already have.

AI gives us a much better way to find it.

Put AI on the P&L

This is where I think mortgage leaders need to get pretty practical.

You don’t need another dashboard telling you how many employees opened an AI tool or how many pilots are underway. Those are fine adoption metrics. They just don’t tell you whether the business got better.

For every meaningful AI investment, I’d want to know three things:

What business number are we trying to move? What is that movement worth? How will we know AI actually helped?

Say an AI tool eliminates 30% of the manual work your processors perform and creates capacity for another 400 loans a month.

Great.

But if you’re still funding the same number of loans with the same headcount and the same overtime expense, the economics haven’t changed yet.

You created capacity. Now you have to decide what you’re going to do with it.

Maybe you take out cost. Maybe you absorb more volume without hiring. Maybe faster response times improve pull-through. Any of those can create value, but somebody has to decide which outcome you’re actually pursuing.

The same issue comes up when you optimize one part of the mortgage process.

AI might cut underwriting touch time by 30%. But if appraisal turn times are still the constraint, making underwriting faster may simply create idle capacity somewhere else. Mortgage has too many dependencies to look at productivity one task at a time.

The business outcome is what matters.

And decide how you’re going to measure that outcome before you roll a capability out everywhere. Mortgage is too volatile for simple before-and-after comparisons. Rates move. Seasonality changes. Product mix changes. Lead quality changes.

Hold back a comparable group. Stagger the rollout. Compare similar borrower cohorts. Do something that lets you answer a basic question: compared with what?

Saving an hour is useful. The financial value comes from what your business is able to do with that hour.

Start Asking a Bigger Question

I keep coming back to that CEO’s question: “How can AI actually help me grow my business?”

AI is going to make mortgage dramatically more efficient. Given where industry economics have been for the past several years, that matters a lot.

But I think we’re still early in understanding what happens when we apply the same technology to growth.

Can we understand sooner why a qualified borrower is sitting on the sidelines? Help someone find a mortgage product they didn’t know existed? Spot emerging demand earlier? Help a great loan officer manage twice as many relationships without making the experience feel less personal?

I don’t think any of us know all the answers yet.

That’s what makes this moment interesting.

For several years, our industry has understandably been asking how to operate with less: fewer people, lower costs, less excess capacity.

We should keep asking those questions.

But AI gives us a reason to start asking a bigger one again: How do we grow?

(Views expressed in this article do not necessarily reflect policies of the Mortgage Bankers Association, nor do they connote an MBA endorsement of a specific company, product or service. MBA NewsLink welcomes submissions from member firms. Inquiries can be sent to Editor Michael Tucker or Editorial Manager Anneliese Mahoney.)