Imagine coming across a Hinge profile with the just-right bio and photos. You match, and both agree to meet. The conversation over coffee is fine, but as the check arrives, you realize this match isn’t so interested in you as in “closing” the date. Subtle, unmistakable and it takes the relationship in a very different direction.
Doesn’t this sound like getting a mortgage?
The industry has spent decades optimizing the profile – the rate sheet, the loan estimate, the digital point-of-sale experience – without much thought about whether any of it builds trust. Rate is the swipe-right mechanism: It attracts attention, generates volume and builds the funnel. What it does not manufacture are the signals that turn a transaction into a relationship. Until recently, it frankly didn’t matter. Volume covered for the absence of relationship when rates were low, and lenders were swimming in transactions. Why plan for the future when the “seven-year itch” crops up by year two? The industry optimized for the close – the honeymoon period.
Alas, that era is over. With volumes compressed and competition intensified for a smaller pool of borrowers, the industry is discovering it built its entire customer model on levers that no longer differentiate. The trust infrastructure was never built because it was never required. Until now.
The four mechanisms the industry must heed
Trust is not a single thing. Research on high-stakes digitally mediated relationships – online dating is the clearest laboratory, and yes, there is a body of academic research on this – identifies four distinct mechanisms, each doing different work at different points in a relationship. This industry has built its customer model almost entirely around one of them and has largely ignored the other three.
1. System confidence
The first is system confidence – does the platform work as promised? In mortgage, the GSEs and the compliance regime provide this: disclosure requirements, underwriting standards, the rep and warrant framework. Most borrowers take it for granted. Lenders are mistaken to assume this automatically transfers to them as an institution.
2. Trustworthiness
The second is trustworthiness – can I trust this counterparty before we have any history together? This is what a good loan officer actually does. Not processing applications, but manufacturing trust signals. Most lenders depend almost entirely on this mechanism, at the individual level, which is why it too often follows the LO out the door when they leave.
3. Relational trust
The third is relational trust – the confidence built through a track record of the other party acting in your interest over time. Political scientist Russell Hardin called this encapsulated interest: The idea that trust is earned not through promises but through demonstrated alignment of interests across repeated interactions. This is what servicers are best positioned to build and have most consistently failed to.
Thirty years of monthly payments is not a track record of shared interest – it is a billing relationship. J.D. Power’s 2025 Mortgage Servicer Satisfaction Survey puts a number on it: average servicer satisfaction runs 131 points below originator satisfaction. Servicers should not confuse the size of a book with the depth of a relationship.
4. Dispositional trust
The fourth is dispositional trust – the baseline openness a borrower brings to the transaction. Some extend trust readily. Others require every signal to be earned. The industry has never designed for this range, which is why the same process feels adequate to one borrower and adversarial to another.
The invisible failure
Trust failures in online dating are immediately visible – you get stood up, in-person looks nothing like the photo – because the feedback loop is tight and unambiguous. In mortgage, failures are mostly invisible at the moment they occur. An LO overpromises on a rate lock. A pricing model incorporates a factor the borrower never knew was in play. A servicer misapplies a payment. None of these announce themselves as trust violations.
The borrower attributes the outcome to the market, to bad luck, to the complexity of a process that was never designed to be understood. The borrower has no signal that a trust violation occurred; only that the outcome was worse than they hoped.
This invisibility is what has allowed the industry to substitute rate for trust for so long without consequence. The feedback loop that would discipline a dating app to build better trust infrastructure – that first-date verification moment where claims are checked against reality – simply does not exist in mortgage. Most borrowers transact once or twice in a lifetime. The signal never returns. And the closing table is too late and too costly to walk away from.
What AI changes
AI does not introduce new failures into this system. It makes the ones already there impossible to ignore at scale. Operating across thousands of decisions simultaneously, AI converts what were previously individual, invisible failures into systematic, discoverable patterns – visible to a plaintiff or regulator with the right data tools, even when invisible at the individual level.
The deeper problem is this: The industry’s one functional trust mechanism – the LO relationship, the human touchpoint where trustworthiness signals were manufactured and relational trust was initiated – is precisely where AI deployment is most often designed to reduce. Efficiency gains in origination are largely gains extracted from the interaction layer. What gets automated is the conversation. What gets lost is the mechanism.
Dating apps learned through competitive pressure that trust is a product feature. Platforms that manufacture it retain users; platforms that don’t are lost to history. The mortgage industry has not had that disciplining mechanism because its failures are invisible and its customers don’t repeat often enough to wise up. AI ends that insulation at exactly the wrong moment.
The architecture the industry needs to build
My last piece argued that the mortgage industry’s accountability architecture stops at the wrong point in the stack – governance drawn at the document layer misses the decisioning layer where consequential choices are actually made. The customer trust problem is the same argument from the borrower’s side. The industry’s trust architecture stops at the transaction layer. The relationship layer, where the thirty-year commitment actually lives, remains unbuilt. That is the opportunity.
For the C-Suite, the strategic picture is fairly straightforward (if uncomfortable). Trust is a balance sheet asset the industry has never capitalized. The lenders who build it early will see it compound. The rest will find themselves competing on rate in a market where rate alone no longer closes the gap.
Each of the four mechanisms described above is a design decision with direct implications for retention, recapture and referral economics. System confidence needs to be made visible to borrowers who currently can’t see it. Trustworthiness needs to be owned at the institution level, not left to walk out the door with individual LOs. And relational trust requires servicers to behave like relationship managers – proactive, borrower-interested and present between problems – rather than billing agents.
AI is fundamental to each. Deployed carelessly, it automates away the touchpoints where trust is manufactured. Deployed deliberately, it is the first tool the industry has had to scale trust-building beyond what any individual loan officer can do alone.
None of this shows up in pull-through metrics or cost per loan. It shows up in recapture rates, in borrower-initiated referrals, in the spread between what a lender charges and what a borrower believes they deserved. Rocket’s recapture rate, running at three times the industry average, is not a technology story. It is a trust story, and the technology is simply how they built the architecture to produce it.
The mortgage industry already knows how to optimize a transaction. The question is when it will recognize that thirty years is not a relationship.
Because AI is about to make the difference impossible to ignore.
Marvin Chang is the Executive in Residence at Duke University Pratt School of Engineering and Principal at Mercer Knoll Strategies.
This column does not necessarily reflect the opinion of HousingWire’s editorial department and its owners. To contact the editor responsible for this piece: [emailprotected].