The True Cost to Originate a Loan (and Where It Hides)

The Mortgage Bankers Association's Quarterly Mortgage Bankers Performance Report has tracked loan production expenses climbing for several consecutive quarters, now running well above $10,000 per loan across the industry. That figure is real, and it is directional: costs have been rising for some time. It is also incomplete, because it measures cost the way accounting measures cost, as a set of line items, and the largest cost inside mortgage origination has never been a clean line item. It is rework, and it is distributed across five or six categories where nobody is looking for it as a single number.

What the MBA benchmark actually measures

MBA's cost-to-originate figure, drawn from its quarterly Performance Report and built jointly with STRATMOR's Peer Group Roundtables Program, aggregates commissions, compensation, occupancy, equipment, corporate allocations, and other production expenses, divided across funded loan volume. It is the industry's standard reference point, and it is genuinely useful for peer comparison on the categories it captures.

The number moves quarter to quarter, and it has trended upward over the long run. Current costs sit meaningfully above the multi-year historical average MBA reports for the industry. Regional and institutional variance is substantial too. MBA's data consistently shows higher-cost-of-living regions such as California and the Northeast running well above the national average, while the Midwest tends to run below it, a gap partly explained by average loan balance, since commissions scale with loan amount, though loan balance alone does not fully account for it.

Institution type matters as much as geography. MBA and STRATMOR's benchmarking of the retail production channel has consistently shown depositories running meaningfully higher per-loan costs than independent mortgage companies, largely because depositories carry heavier corporate cost and production support allocations. The spread between well-run and poorly-run operations is widening as well: MBA's own year-over-year comparisons of top-quartile companies against bottom-quartile companies show that gap growing substantially over the past several years, well beyond what it was a decade ago.

That widening gap is worth sitting with. Two companies can operate in the same market, under the same rate environment and the same regulatory regime, and still show a growing difference in what it costs them to produce a loan. Something structural is diverging between them, and it is not fully captured by any single line item on the MBA report.

Where personnel expense actually goes

Personnel cost is the largest single input to the MBA figure, typically representing well over half of total production expense. That tells you something the benchmark itself doesn't break out: mortgage origination is still, structurally, a labor-intensive business, even after two decades of technology investment aimed explicitly at making it less so.

What the benchmark doesn't separate is why personnel cost stays this high relative to loan volume. It does not distinguish between labor spent on work that had to happen once, such as underwriting judgment calls and exception handling that genuinely required a person, and labor spent redoing work that had already been done correctly somewhere upstream. Both show up in the same personnel-expense line on the MBA report. Only one of them is necessary.

The cost the benchmark doesn't separate out: trust friction

Every loan file moves through pre-settlement review, warehouse funding, post-settlement or TPR review, buyer or investor diligence, custody, and eventually servicing boarding. At most of those handoffs, someone re-establishes a fact that a previous participant already established. Income gets re-verified a second time. A cleared condition gets re-checked. A certification gets re-derived because it didn't travel with the file in a form the next party could rely on.

That repetition has a name: trust friction. It is the cost of manufacturing the same trust more than once, and it is nearly invisible on a standard cost report because it is spread across origination, QC, and delivery functions rather than concentrated in one place where a controller would notice it as its own line item.

Illustrative modeling of a representative loan puts the review-cost impact of eliminating redundant verification in the mid-hundreds of dollars per loan. That figure comes before accounting for time. A file that has to be independently re-verified at each stage takes longer to move through the pipeline than one where later participants can act on earlier certification, and every extra day on a warehouse line carries a cost. Compressing average warehouse dwell time on a representative warehouse line releases capital and avoids a meaningful amount of fully loaded capital carry per loan: funding cost, hedge cost, liquidity charges, and capital allocation, all of which accumulate while a loan sits waiting on verification that has often already happened.

Combined, illustrative modeling puts direct value in the range of $1,500 to $1,700 per loan, in addition to whatever the MBA benchmark is already capturing. At meaningful scale, that adds up to a substantial, unlabeled amount sitting inside the cost structure because nobody built a line item for it. These figures are representative modeling assumptions, not audited results, and actual results vary by institution, loan mix, and operating model.

Why cost per loan keeps rising even as technology spend goes up

This is the pattern that should be more alarming to the industry than it currently is. Technology investment in mortgage has grown substantially over the past decade, and cost per loan has not fallen in proportion. According to MBA's own longitudinal data, it has instead risen well above its historical average. The relationship between investing in technology and lowering cost per loan has been far weaker than the industry expected when that investment began.

The technology itself was rarely the problem. Most of it targeted the wrong layer. Faster document review, faster income verification, and better-informed underwriters all make an individual step quicker. None of that changes whether the next participant in the chain has to independently re-verify the output before acting on it. If they still do, the trust friction is still there, just running slightly faster than before. A faster re-verification is still a re-verification. It still shows up as personnel cost. It still adds to cost per loan on next quarter's MBA report.

Where the real number lives, and how to find it

If you want your actual cost to originate, the real figure rather than the benchmark version, three questions will get you closer than any dashboard currently does.

What percentage of files touch a given function more than once? Set exceptions aside. Look at ordinary files coming back around for a second pass through underwriting, QC, or closing for reasons that shouldn't have required it.

How many times is the same fact independently re-verified before close? Income, assets, and conditions are the usual suspects. Count re-establishment from source, not simply views of a prior finding.

What does your team maintain outside the system of record, and why? Every shadow spreadsheet a processor or underwriter keeps "just to be safe" is a control point someone built because they did not trust the system that was supposed to make it unnecessary, and it is usually a proxy for exactly where trust friction is concentrated.

What changes the number

The MBA benchmark will not move by hiring differently or negotiating vendor contracts down. Both of those approaches have been tried across the industry for the better part of a decade, and the trend line has moved the wrong direction regardless. It moves when verification stops being repeated at every handoff, when the evidence, certification, and lineage behind a finding travel with the file so the next participant can act on it instead of re-deriving it from scratch.

This is a redesign of where the cost was actually coming from, a cost that, for most institutions, has never been fully counted, let alone addressed, because the MBA benchmark was never built to isolate it in the first place. It was built to compare institutions against each other on the costs everyone already tracks. Trust friction is the cost that shows up inside all of those categories at once, in a way no single peer comparison will ever surface.