Better’s $3,000 Mortgage Claim, Examined

A note to subscribers: Welcome to a special edition of The Mortgage Scoop! Today, mortgage exec & ops expert Karen Postiglioni goes hunting for the math behind Better’s claim that AI has helped bring the cost to produce a mortgage from ~$12K to $3K. Plus, ICE officially punts SDK charges & we revisit the loan-level details behind Lisa Cook’s scrutinized Hotlanta condo.. We’ll be back to our usual routine next week.

If you’ve been reading The Scoop, you already know what you’re getting: real reporting, deep sourcing, & stories nobody else in mortgage media is touching. We’ve exposed shady lender tactics, examined ICE’s hate-love relationship w/ mortgage, broken dozens of tech stories, dug into UWM’s correspondent play, Rocket’s retail strategy & much more… Insiders get the full Monday/Wednesday/Friday edition—scoops, analysis, sourcing, & context you will not find anywhere else—plus early access to new features.

If you rely on The Scoop to stay sharp, informed, & ahead, this is your chance to support independent, scoop-driven mortgage journalism. Insiders pay $275 a year ($22 a month). We also do enterprise packages.

By Karen Postiglioni

One number being thrown around in the Better/Vishal Garg debate jumped out at me that I want to slow down on before it hardens into fact: Better cut its cost to produce a loan from roughly $12,000 to under $3,000.

I went looking for where that claim actually came from, thanks to the editor on this piece, James Kleimann, pointing me toward the original source. It’s not quite what it’s become. But that version is now being used to help build a case that Garg should regain control of Better.

The original claim appears to be from an April ‘26 Fortune profile of Garg. His exact words, per Fortune: Betsy, Better’s AI voice agent, cuts the cost to process a loan from the industry average of nearly $12,000, according to Freddie Mac, to $3,000.

Read that closely. The $12,000 is the industry average, sourced to Freddie Mac. The $3,000 is attached specifically to Betsy, not presented as Better’s company-wide cost per loan.

By August, in HousingWire’s coverage of the board fight, the claim had changed shape: “Better has also reduced its cost to produce a loan from about $12,000 to less than $3,000.” Company-wide, before & after. That’s not what Garg said in April. It’s what his April comments turned into after a few months & a news cycle. 

That’s the version that showed up in a widely-read LinkedIn post about whether Garg should return as CEO. It is an extraordinary number. And if that’s true, it’s the single most important piece of information in the entire Better story right now.

Which is exactly why I wanted to understand it. Because if you really found a way to produce a mortgage for $3,000 against an industry benchmark of roughly $12,000, wouldn’t you be shouting it from the rooftops & telling everyone ad nauseam?

Right now, neither Garg nor Better are publishing any specifics behind it.

So where does the $3,000 actually come from?

Freddie Mac runs an actual, ongoing “cost to originate” study. Its most recent published figure for Q2 2025, is about $11,800 per loan for retail lenders, close enough to “nearly $12,000” to hold up. That number is also consistent with a separate, independently run industry benchmark: the MBA’s own data puts average production expense at $11,898 in Q1 2026 & $10,936 per loan in Q2 2026.

What does Betsy’s $3,000 figure actually measure? Is it calculated the same way Freddie Mac or the MBA calculate cost to originate, or is it a narrower internal number being held up against a fully-loaded industry figure?

There’s a second wrinkle. A separate Better/ElevenLabs case study says Betsy reduced Better’s average cost to originate by 41% in 2025, but provides no baseline or resulting dollar figure. That doesn’t necessarily conflict w/ the $3,000 claim because the two comparisons use potentially different starting points. It does, however, leave us unable to reconcile the two or understand what either calculation includes.

Better's ‘26 Q2 10-Q reports $3.5M of "loan origination expense" for the quarter, but that is a specific expense category, not a fully loaded cost to originate. Compensation, technology, marketing, & general & administrative expenses are all reported separately. And Better doesn’t provide a per-loan breakdown of it anywhere in the filing, Betsy specific or otherwise.

Then on Aug. 17, the Garg Group filed its preliminary consent solicitation w/ the SEC as part of the push to reconstitute Better's board. The filing argues for continuing Better's "cost transformation" & recounts Garg’s proposal for the “continued scaling of Tinman AI” as core to the turnaround case.

The $3,000 number isn't there either. That seems like a pretty obvious place to include it. If Betsy is really cutting Better’s comparable production cost to roughly 75% below the industry benchmark, that is a powerful part of the case to shareholders that Garg's strategy is working. Yet the number and the methodology behind it aren't included.

That doesn't mean the number is wrong. It means we still don't know what $3,000 measures.

But also…$3,000 compared w/ what? MBA's calculation does include commissions, production compensation, occupancy, equipment, other production expenses, & corporate allocations. For additional context, MBA's retail-channel study showed lenders averaging $12,209 per loan in 2025.

Those are averages, not some theoretical floor. There is nothing mathematically impossible about Better achieving a way to operate well below them. Technology & automation are supposed to drive exactly these kinds of costs down. And Better has spent years — & a lot of money — building Betsy & Tinman around that thesis. On the same Q2 call, management raised its annualized cost-savings target from $25M to more than $45M.

On Aug. 19, I sent Better’s PR team a specific set of questions: what the $3,000 figure includes & excludes, whether it’s methodologically comparable to Freddie Mac or MBA’s numbers, whether the loan population includes HELOCs or other second-lien products that could skew a per-loan average, & how the 41% figure reconciles with the ~75% implied by the headline number.

Better did not respond. That doesn’t resolve anything either way. It just means questions aren’t answered yet. Maybe they really have broken the cost curve. But before we say they have, I want to know what's inside the number.

Does the $3,000 include sales compensation? Marketing & customer acquisition? Processing, underwriting, & closing? Technology? Corporate allocations? Does it include the cost of building & maintaining the tech itself?

And how are costs being allocated now that platform volume represents 55% of Better's total loan volume? A loan Better originates itself & a loan running through Tinman for an outside lender aren't necessarily carrying the same economics.

Those aren't technicalities. They're the difference between: ‘Betsy gets some defined portion of the origination cost down to $3,000,’ and ‘We can produce a mortgage for less than $3,000 while the rest of the industry averages around $11,000 to 12,000.’

I’m not trying to knock the number; I simply want it explained. Because if that is truly an apples-to-apples, fully loaded production cost, the conversation around Better should change from intriguing incremental improvement to nailing a totally different mortgage manufacturing model.

The more extraordinary the number, the more important it is to understand how it was calculated. Because this is how numbers become conventional wisdom in the mortgage industry. Someone says it. Someone else quotes it. It gets paraphrased. It ends up in a LinkedIn post, then a deck, then another article (like this one!).

A few repetitions later, nobody remembers where the number originally came from, like the game “Telephone.” They just remember that Better originates loans for $3,000.

If the $3,000 is a narrower manufacturing number that excludes customer acquisition, sales comp, technology, & overhead, tell us. That's still useful, & possibly impressive information.

If it's fully loaded & comparable to the MBA definition, tell us that too. Because then I don't want to debate whether $3,000 is believable. I want to have a completely different conversation about Betsy…& Tinman.

About the Author

Karen Postiglioni is a mortgage and fintech executive focused on how technology, operations, and customer experience can reshape lending.

In July I reported that ICE Mortgage Technology had blinked on charging lenders for SDK on Encompass past its latest deadline.

Well, ICE has officially confirmed it & emails are making the rounds on LinkedIn. Back in July, a source told me then that a billing extension was in the works, that it would only impact the ~50 biggest lenders, & that ICE wouldn't call it an "extension."

Why the top 50? They're the ones w/ real SDK volume, & the ones ICE, I imagine, would rather not needlessly antagonize.

The Trump administration is still going after Fed Gov. Lisa Cook on mortgage fraud charges. And Cook is fighting back.

Per the NYT, in a defiant letter, her lawyer Abbe Lowell said that listing the Atlanta condo as a primary residence was an “entirely inadvertent oversight,” not fraud. He noted that Cook had correctly identified the property as a vacation home earlier in the mortgage application process, & said she had not received a preferential rate on her loan as a result of the error.

I did a deep dive into the case against her back in September. I was the first journalist to identify loan-level data tied to her Atlanta condo via HMDA. Here’s an excerpt:

(🙏 If you like what you’re reading, tell a fellow mortgage junkie to sign up here.)