How Lenders Are Buying, Scrubbing & Chasing Leads With AI

Last summer, I chatted w/ a big mortgage broker who was scooping up 5,000 trigger leads a day. Every day. That’s about $100K a month, or $1.2M a year, just to get the names of people who’d applied for a mortgage somewhere else. 🤯 Then Congress came along & largely killed ☠️ the business.

Lately I’ve been wondering: What happened to all the lenders buying leads by the truckload? Where are they finding borrowers now? And what's actually working?

I spent the past week talking to some of the savviest people in mortgage lead gen: Praveen Chandramohan at Cotality, Samir Sen at FlairLabs, Rick Doyle at LeadMailbox, Jonathan Haddad at Bevri.ai & Bijoy John at Sage Home Loans.

A few things surprised me. For starters, leads have actually gotten considerably cheaper in some corners of the market. And smaller shops now have access to lead sources that were once the exclusive playground 🛝 of the big boys.

But the really interesting stuff is happening after a lead comes in.

Lenders are increasingly using AI to figure out which borrowers are worth chasing, how to engage them & when to hand them off to a human. The goal isn't to buy more leads; it's to squeeze a hell of a lot more business out of the ones they've already got.

In this special edition of The Mortgage Scoop, sponsored by Plaid, we're digging into the new economics of mortgage lead gen & how AI is changing the game.

Lenders looking for net-new customers are largely shopping in two places, according to Cotality’s Praveen Chandramohan.

The first is the credit bureaus' prescreen & prequal lists. Give them your credit criteria & they'll return a list of potential borrowers. "People are buying those lead lists in the hundreds of thousands," he said. Sometimes for pennies per lead.

The second is property-based targeting. Doing FHA loans in Texas, Alabama & Oklahoma? Identify matching homeowners, then add equity, other properties, servicers & estimated rates. MSR owners are using similar borrower & property based targeting to retain & recapture business from their servicing portfolio.

That's where AI comes in. Chandramohan said companies are developing MCP tools to enrich property & homeowner records. Lenders are building profiles of who might need a loan & when.

Location triggers haven't disappeared, but they're mostly used to monitor existing customers for recapture rather than find new ones, he said.

Then there are aggregators like LendingTree, Bankrate, NerdWallet & Free Rate Update. Rick Doyle, who has routed aggregator leads into CRMs since ‘03, said prices have "come down dramatically." LendingTree once required something like a $10,000 minimum spend. Now, he said, it's referring one-to-10-user shops to LeadMailbox.

Why? Lead sellers have fewer giant buyers, so they're courting shops they used to ignore. The little guys are finally getting invited to the party.

Here's something absurd: Lenders routinely spend a fortune buying leads, then straight-up don't call enough.

Sen, whose voice AI agents work w/ large lead buyers like West Capital Lending, estimates that inconsistent follow-up costs lenders 15% to 20% of the active conversations they could be having. He advocates eight attempts per lead & 100% coverage.

"Most loan officers & sales teams are hitting voicemail & busy signals like 90% of the time, & they really hate that," Sen said. "Because of that, actually, they give up."

Doyle pointed out that aggregator leads often get sold four or five times. By the time the third or fourth LO calls, the borrower is already getting bombarded. A couple busy signals, a text, an email & the LO moves on.

So here's how LeadMailbox & FlairLabs handle it: The LO gets the first call. If they mark the lead "unable to contact," an AI agent takes over. "Okay, Flair, go to work," as Doyle put it.

The agent keeps calling & texting, subject to contact rules, until it reaches someone, then transfers them to the LO. Sen said it checks the CRM before every attempt so it doesn't keep pestering a borrower after a human has made contact.

The AI doesn't sell the mortgage; its job is simply to get the borrower on the phone so the LO can do what he/she does best.

Doyle shared some results. At First Choice Lending, roughly 6,700 leads went through Flair from June to September. Thirty percent (2,019) of 6,678 leads responded by phone or text. It resulted in 154 live lead transfers sent, 163 callbacks & appointments, 309 text introductions to an LO & 604 unique leads handed to an LO. First Choice's leadership told Doyle its closing rate on those leads was up close to 50%.

Here's what came up in almost every conversation: Lenders are so busy chasing new borrowers that they're neglecting the ones they've already paid to acquire.

"The cheapest kind of lead is the one that's free," Sen said. He estimates retail lenders lose 70% to 80% of their repeat business when past customers refi or buy again elsewhere.

Kinda crazy, right? You spend hundreds or thousands acquiring someone, close the loan & then watch them disappear? Sen thinks AI dialers can work those past-client books at scale when rates fall, w/o lenders having to hire an army of LOs.

At Sage Home Loans, part of the Bankrate family, Bijoy John is putting that thinking into practice. "First-party data is where all the fun is," he told me.

Sage built its own CRM, screening tools & coding team. John said AI-powered compliance reviews of marketing materials hit roughly 95% accuracy, cutting turnaround from days to hours.

And the outreach is personalized. A customer opens an email or clicks a HELOC link? Sage uses that signal to determine the next call, text or email.

For purchased leads, Sage scores each one in "split seconds," routing it to an internal LO or a third-party dial center. If nobody answers, AI dialing starts after a few days & a texting bot tries to book a call. If the borrower prefers texting, the system adjusts.

"It might seem kind of normal for a lot of companies," John said, "but you'd be surprised how many companies don't do that today." (Me? Not surprised. I talk to lenders every day 😬)

Here's another wrinkle: Sage uses AI to tell lead sellers what they're doing wrong. It sends anonymized, LLM-generated feedback to aggregators, w/o sharing PII, ages or credit scores. When we spoke, John had just told a major HELOC aggregator that roughly 40% of its leads fell outside the age range his lender partners typically fund.

Jonathon Haddad at Bevri.ai, a broker CRM/LOS platform in live beta, had a simple question for lead vendors: Show me your best client's lead-to-funded conversion rate. "Not one lead provider was able to give me that information," he said. "Not one."

So he's building a scorecard. With both the CRM & LOS, Bevri tracks leads through funding, grading vendors & LOs on speed, talk time & conversions.

"I can see when you buy the lead, & I can see when you fund the lead, all in my system w/o you having to give me the data," Haddad said.

Vendors who won't provide API access? "No problem. You just can't be in Bevri." He also wants to reward high-converting LOs w/ free months if they share successful scripts.

Here's where things get juicy🧃: a broker might buy $80 HELOC leads rather than $200 cash-out refi leads, then try to convert those borrowers to cash-out. B/c Bevri sees what funds, it could show sellers how often their HELOC leads turn into something else. Haddad hasn't figured out how to monetize that info yet, but I believe it illustrates where things are headed in lead-gen.

Four takeaways from these conversations:

Thanks for reading, & thanks again to Plaid for sponsoring this special edition of The Mortgage Scoop! Watch the full convo detailing how digital tools & real-time financial data are helping lenders streamline verification, cut costs, & deliver a faster borrower experience.