The Mortgage Experience Borrowers Are Really Asking For

What borrowers actually want splits pretty cleanly depending on whether they’re purchasing or refinancing, and I think the industry still underrates how different those two psychologies are. A purchase borrower, especially a first-time buyer, mostly wants certainty that this thing gets done on time and someone else project-managing it for them. A friend of mine recently told me the most annoying part of her mortgage was that the lender kept insisting on calling her instead of texting, which sounds trivial but says a lot: she wasn’t looking for more or less touchpoints, she was looking for less to worry about. On the refi side, people care far more about rate and speed and much less about hand holding. But underneath both of those, every borrower shares the same core frustration, which is ending up as the de facto project manager of their own loan because the lender isn’t communicating proactively, so they’re left chasing updates and wondering if everything is actually fine. There’s an interesting pricing wrinkle here too. People are demonstrably willing to pay more when they perceive that someone did real work on their behalf, which means you can likely price a loan so the borrower feels like they’re paying for the capital markets instrument itself, not the manufacturing labor behind it, and that reframing changes how the fee feels even when the economics stay the same.

On the industry conversation shifting from streamlining loan processing back toward earlier borrower engagement and relationships, I think that’s a real pendulum, and it swings for a predictable reason: once everyone is doing the same optimization, doing it stops being an advantage, so a vocal cohort of lenders goes looking for whatever the next differentiator is. But there’s a second, more concrete reason this shift is happening right now, which is that lenders have genuinely underestimated what current AI can do. For a decade, the industry’s mental model of AI was reading documents, that was the safe, obvious, regulatory-friendly use case, so that’s where everyone started. What’s changed in the last three to six months specifically is that voice demos have gotten good enough that lenders who would have laughed off the idea of an AI handling a real borrower conversation a year ago are now sending each other clips saying it’s crazy how well it works. The technology moved, but just as importantly, people’s understanding of the technology moved, and that has reopened conversations about automation that felt closed a year ago.

I’d be careful, though, about how much weight to put on the public version of this conversation, because a lot of what gets said on an earnings call or a conference stage is shaped by incentives that have nothing to do with what’s actually working. Companies talk up their AI usage because that’s what recruiting narratives and stock analysts want to hear, and when you actually dig under the announcement, you often find a human still reviewing every case, or the deployment limited to one narrow product line like HELOCs. The real, nuanced picture of what’s working almost never shows up in a public statement, but rather in one-to-one conversations with people you trust. If you build your strategy by chasing whatever the loudest voice on social media claims to be doing, you’ll find yourself pivoting constantly, because those same people will be announcing a different pilot next quarter.

On the bigger question of when AI-driven underwriting and processing become genuinely ubiquitous, I think we’re further away than the hype suggests, and the reason is the same one that’s slowed every technology diffusion cycle in this industry: it’s not enough to deploy the tool, you have to re-architect the process around it. Right now, plenty of lenders have AI agents doing real work, but they still have a human reading every single case the agent touches to confirm it did what a person would have done. That’s not a streamlined process, that’s a QC layer wrapped around automation. Getting from that state to genuinely removing people from the loop on most loans is a 12- to 24-month journey for an individual lender, and probably a three- to five-year diffusion curve for the industry as a whole. The trap is that three to five years sounds like plenty of runway, so people don’t start soon enough, and then two years in they realize competitors who started earlier are a year out while they’re still three years out.

Once that manufacturing layer does get commoditized down to essentially the cost of the technology itself, the real question becomes what a lender actually competes on next. Cost of manufacturing is where a lot of the current conversation is focused, understandably, because it’s a commodity product and cost is the obvious lever. But once everyone’s manufacturing cost converges, the differentiation moves elsewhere: brand, existing customer relationships, and partnerships that put you inside an ecosystem where consumers already are, rather than trying to acquire them cold. I’d actually push back on one popular claim here, that customer experience alone will be the deciding factor, since buying a home is infrequent enough that I’m skeptical “experience quality” moves the needle on its own the way people assume. Cost of capital is the other lever worth watching, since banks are candid that regulation makes their cost of capital less competitive today, and you’re already seeing alternative capital sources, hedge funds, insurance funds, and platforms like Figure, start to reshape that side of the equation in ways that could matter more across the next decade than any front-end feature.

After hundreds of conversations that I’ve had across the Daily Mortgage News Podcast and various Chrisman Commentary video shows, the one piece of advice I’d give any lender trying to deploy AI thoughtfully rather than just throwing it at everything is: anchor every deployment to a specific business problem, not a general ambition to look cutting edge. If your actual problem is that you’re over-conditioning loans and asking borrowers for 20 percent more documentation than necessary, that requires a completely different AI solution than if your problem is catching a ring of borrowers running sophisticated bank statement fraud. Once you know the problem, you know which metrics in the business should move if you’ve solved it, and which stakeholders should notice. Measure success against that, not against tokens burned or how many loans are running through the new tool, because those are proxy metrics that tell you adoption happened, not that the problem actually got solved. Solving the problem versus deploying the technology, is the one facet most of the industry’s current AI conversation is still missing.