The file everybody remembers from last month closed on a Thursday. Sixty-eight days from application. Two lock extensions. A tolerance cure at the table. Your LO gave back a quarter point in week nine to keep the borrower from walking, and your processor stayed late twice to save it. When the wire went out, the team clapped.
That file lost money. Nobody in the building knows, and nothing you will ever be handed says so. On the 14th your margin comes back at 244 basis points against a posted 250, and somebody tells you to watch your pricing.
Here is why losing files hide instead of announcing themselves. Revenue on a file is a percentage of the loan amount. Cost on a file is a function of how many times a human touched it. Those two numbers have nothing to do with each other. Loan size is set by the house the borrower picked. Touch count is set by their documents, the appraiser, the agent, and your own handoffs. Average margin blends both and destroys the only signal you can act on.
The short version
- Revenue scales with loan amount. Cost scales with touch count. Nothing links them, which is why a respectable average can sit on top of files that cost more to make than they sell for.
- Two numbers per file, never one: contribution (revenue minus attached costs) and an overhead hurdle (fixed cost divided by files funded). Never smear rent across files.
- The attach test: a cost attaches to a file only if it would not exist had that file never opened.
- Divide fulfillment cost by files funded, not files worked. The gap is your fallout tax.
- No file is final until its EPO window closes. Grade every file twice.
Two files, same month, same 250 basis points
Say corporate credits your branch 250 bps on funded volume and your LOs sit at 110. Two files out of the same month.
The first is a $520,000 purchase. W-2 borrower, eleven conditions, twenty-four days from application to wire. Revenue $13,000. LO comp $5,720. One file's worth of processing capacity, which we will price at $375 in a minute. Contribution: $6,905.
The second is the file everybody remembers. A $165,000 refinance, self-employed borrower, sixty-eight days. Gross revenue $4,125, less a quarter-point concession — $413 — so $3,712 of revenue that actually existed. Then the costs that file and only that file created: comp, three files' worth of processing capacity, two lock extensions, the cure, and the reruns nobody logged.
Contribution: negative $155. Sixty-eight days of branch capacity spent to move backwards by a hundred and fifty-five dollars.
Now blend them. $16,712 of revenue on $685,000 of volume is 244 bps. Six basis points off plan. Nobody investigates six basis points. The average did not lie to you. It answered a question you should stop asking.
The attach test
One question sorts every dollar the branch spends: would this cost exist if this file had never opened?
Yes, and it attaches. LO comp on the deal, already computed per file and the biggest attached cost you have. The fulfillment capacity the file consumed. Lock extensions and re-locks. Tolerance cures. Hard costs you ate instead of passing through — a credit rerun, a rush verification, a second appraisal. And clawbacks, whenever they arrive.
No, and it pools. Rent. LOS seats and the rest of the tech stack. Branch marketing. Your salary. Licensing, E&O, the corporate allocation.
Two of these get booked wrong almost everywhere. Clawbacks. An early payoff inside the investor's EPO window pulls the premium back months after funding. Book that as a general branch expense and the file that caused it stays profitable in your records forever, along with whatever pattern produced it. Fulfillment. If your processor is salaried you have to build a per-file rate, and how you build it decides whether this whole exercise tells the truth. More on that shortly.
Contribution first, hurdle second, and never smear
Carry two numbers per file, not one.
Contribution is revenue minus attached costs. It answers exactly one question: was this file worth doing at all?
The hurdle is total monthly fixed cost divided by files funded. Say your fixed cost is $50,000 and you fund twenty files — $2,500 a file. A file that clears its attached costs but not the hurdle is not a losing file. It still helped pay the rent. A month full of them is a losing month.
The temptation is to skip the second number and allocate overhead into each file instead. Don't. Smearing rent per file makes every small loan look like a loser, and it moves every file's "profit" whenever volume changes even though nothing about those files changed. Keep contribution clean, keep overhead visible, and reconcile the stack against your branch P&L monthly. Where bottom-up and top-down disagree, you have found something.
The fallout tax
Here is the number that decides whether your per-file math is honest, and nearly everybody gets it backwards.
Say your processor costs $7,500 a month fully loaded, works 26 files, and 20 of them fund. The instinct is $7,500 divided by 26, or $288 a file. That is the cost per file worked, and it is the wrong denominator, because only funded files carry revenue. The right number is $7,500 divided by 20 — $375 a funded file.
The $87 gap is not an accounting nicety. It is the cost of the six files that died, redistributed onto the twenty that lived. Add the hard costs you ate on the dead ones before they died, and the fallout tax on this illustrative month runs about $144 per funded file.
That money appears on no file's ledger. It sits inside overhead, where it reads like the price of doing business. It isn't. Every dollar of it has a source, a product, and a stage attached, so the fix is attribution rather than accounting: tag every dead file with where it came from and where it died. A relationship that sends two files a month and kills one of them is not neutral. Pull-through is a profitability number, not just a pipeline number.
No file is final until the EPO window closes
Grade every file twice.
Provisional, at funding. Revenue, attached costs, contribution. This is the number you manage with, and it is an estimate.
Final, when the window closes. Know your investor's EPO window and your own comp clawback terms cold. They are not always the same length, and the gap between them is branch exposure nobody wrote down.
The cruelty is that the two lists overlap. The file you conceded on to win a rate shopper is the same file a competitor refinances the moment pricing moves. You paid to win a borrower who was never loyal, then paid again when they left. So if you fund an aggressively priced, refi-heavy month, don't spend it. Keep a pending column and let the windows clear before you call it a good quarter.
Read the bottom five by pattern, not by name
First Friday of the month, funded files in a sheet, sorted by contribution. Read the bottom five. Four cuts, in this order:
- By source. Which relationships and channels show up disproportionately at the bottom of the funded list — and in the dead-file log.
- By product. Some programs carry touch time your pricing never covered at your average loan size. That is a menu decision, not a coaching one.
- By originator. If one LO's files carry most of the concessions, the problem is at the front of the file, not the end.
- By stage. If extensions and cures cluster at the same handoff, the fix is process, and no coaching plan will touch it.
The originator conversation only works if you open on the pattern instead of the file:
"I'm not asking about that one file. Three of your last five ended with a concession. Something is happening at application that we're paying for in week nine. Walk me through how you set expectations on the last one."
Some losing files are worth losing. The first deal from an agent you're courting. A small refi for a past client who has sent you three referrals. Fine — lose that money out loud, on purpose. The failure mode is never the strategic loser. It is the accidental one nobody priced, nobody caught, and everybody repeats.
Thirty minutes, and keep it off the wall
You don't need software to start. One row per funded file, eight columns: loan amount, revenue, concessions, comp, fulfillment, extensions, cures, hard costs. Contribution falls out. Second tab for dead files, two columns, source and cause. Thirty minutes the first month, ten after that.
Two cautions. Per-file profit exposes comp, and comp is not team material. This is owner and sales-manager information. In MAVYN the branch P&L lives behind the same login as the pipeline, but financials render only for owner and sales-manager seats, enforced at the database row level. Whatever you use, enforce it somewhere real — a tab nobody is supposed to open is not a control.
And resist turning the monthly read into a trial of whoever's name sits on the worst file. One ugly file is a story. The same ugly file three months running is a system, and the system is yours.
Your average margin tells you how the month went. Per-loan profitability tells you which decisions to stop making.
Compute both. Only one of them is actionable.