The agent called me instead of the loan officer, which is never a good sign. She wanted to know whether we were still closing Tuesday.
I opened the pipeline sheet: in processing. I looked at the whiteboard by the printer: CTC. I scrolled the group text: appraisal still out, posted the day before. Three answers, each written by someone telling the truth at the moment they wrote it. The real answer took eleven minutes and two phone calls — the appraisal had come in low that morning, the LO was on with the listing side, and nobody had updated anything because they were busy doing the thing that changed the answer.
That's the whole problem in one line. The drift isn't caused by sloppiness. It's caused by the work.
The short version
- Truth settles in whatever container is cheapest to write to. If your system of record is slower than a group text, the group text is your system of record for everything urgent.
- Every duplicate copy was born from one of four triggers: friction, blindness, silence, or confidentiality. Each takes a different fix.
- Consolidation fixes none of them. One bigger spreadsheet is still slow, still blind, still silent, and still can't show a manager the P&L without showing an LO the comp splits.
- Run the twenty-minute inventory: every surface where status got written last week, who writes it, who reads it, and the last time it was wrong.
- Then retire exactly one copy — the one behind your most recent miss — and give the person who owned it a real job in whatever replaces it.
Truth settles wherever it is cheapest to write
Nobody decided to keep five copies of the pipeline. What people chose, a few hundred times, was the fastest surface available in the ninety seconds they had.
The group text wins because it costs one thumb. The LOS milestone loses because it costs a login, a search, a screen, and a save, squeezed between a callback with the underwriter and the next borrower on hold. So the fact lands in the thread, and the thread becomes your system of record for anything urgent — exactly the category you can least afford to lose.
Which is why "just keep the sheet updated" has never worked. You're asking a person to choose the slower container while a borrower waits. Truth runs downhill. You don't lecture the water. You move the drain.
The four reasons a copy gets born
Before you evaluate any mortgage branch software, get specific about why each copy exists. There are four triggers, and telling them apart matters — they don't share a fix.
Friction. The master is slower to write than the copy, so the fact lands somewhere else. Sticky notes. Texts. The note in an LO's phone about what the borrower said about the gift funds.
Blindness. Someone who needs the fact can't see the master. The whiteboard exists because the spreadsheet isn't visible from the floor. The forwarded PDF exists because the LO can't see the processor's condition queue and got tired of asking.
Silence. The master can't tell anyone anything changed. A lock with four days on it is only urgent if something says so out loud. So somebody builds a reminders tab, a Friday text, a calendar block — a second record whose entire job is to notice.
Confidentiality. The fact can't sit beside the other facts. Branch P&L can't live in the pipeline sheet, because margin and comp detail aren't every LO's to read. So a private second sheet is born. Recruiting notes can't sit next to production numbers, so a third appears. Every time the answer to "who can see this?" is "not everyone," a new version of the truth starts drifting.
Three of those look like tooling problems and one looks like a trust problem. They're the same problem: at the moment of writing, the master didn't fit the job.
A branch doesn't have a spreadsheet problem. It has a permissions problem and a notification problem, wearing spreadsheets.
Watch one fact drift for a week
Pick the fact with the shortest fuse. Lock expiration.
The confirmation is right — it always is. It's the copies that rot. The sheet got the original date typed in by hand and never touched again. Wednesday the file takes a fifteen-day extension. The lock desk knows. The LO knows. The sheet still says there's cushion, and Monday somebody reads that row out loud and the room believes it.
Price it. Say your branch funds $6M a month across twenty files, call it $300K average, and an extension runs an eighth for a week: $375 on that file, out of margin. Two avoidable extensions a month is $9,000 a year — the cheap version. The expensive version is fallout, where you don't lose basis points, you lose the file, the pull-through, and the agent who sent it.
The lead side rots more quietly. If leads live in three places your denominator is a guess, so your pull-through is a guess, so every capacity decision downstream is a guess wearing a number's clothes. And sooner or later two LOs call the same lead an hour apart, which teaches that borrower one thing about your shop.
The symptom you can check this week is your Monday meeting. When status lives in five places, the first stretch goes to reconciliation — reading rows aloud until everyone's copy agrees — instead of decisions. If your Monday pipeline meeting feels like an audit, sprawl is why.
What one source of truth actually requires
The tempting fix is consolidation: one big sheet with everything in it. It doesn't hold, because sprawl was never caused by having too many files. Three properties do the work; anything missing one re-sprawls inside a quarter.
- One record per fact. A lead exists once. A lock date exists once. Views can differ — the LO sees their files, the processor sees the queue, the manager sees the branch — but every view reads the same row. The moment a fact gets typed in two places, you've scheduled a future disagreement.
- Per-seat slices, enforced below the interface. One source of truth doesn't mean everyone sees everything. It means one record showing each seat exactly the slice it's entitled to, enforced by the system instead of by which attachment somebody remembered not to forward. We built MAVYN so financials render only for owner and sales-manager seats, with access enforced at the database row level — because hiding a menu item is not access control.
- It comes to you. A source of truth nobody opens decays into one more file. The record has to arrive: the morning state of the branch, the file that stopped moving, the lock inside its window.
That third property is why MAVIS composes the branch's morning and watches files for stalls instead of waiting to be asked. A truth you're handed daily stays the truth.
The twenty-minute inventory
Do this before your first demo.
- List every surface where a loan's or a lead's status got written down last week. Sheets, boards, texts, the LOS, the sticky note on the monitor, the CRM one LO pays for personally.
- For each, write who updates it, who reads it, and the last time it was wrong. That third column does most of the work — people defend a surface in the abstract and stop defending it in the specific.
- Give every row a verdict. Retire — its facts already live somewhere better. Feed — it's a view, so make it read the master instead of holding a copy. Keep — nothing else can hold this fact yet, and that's an honest answer.
- Read the finished list before acting on it. Two surfaces and you don't have a sprawl problem; go back to work. Six or more and you pick one — the copy behind your most recent miss — and retire it this month.
You'll learn more about what your branch needs from killing one spreadsheet than from three sales demos.
The switching costs nobody prices
Migration is the easy part. Pipelines and lead lists are small data.
The expensive part is that the sheet's real content is habits, not rows. The column where "!!" means rush. The tab only the processor understands. The unwritten rule that a blank close date means the agent hasn't confirmed. Those took years to settle and none of them import.
The dangerous part is two-system limbo. Run both with no end date and you're strictly worse off — one more source of truth than you started with. Parallel is fine for a defined window. Then the sheet dies on a date, and one named person owns the cutover.
There's also a political cost nobody budgets for. Whoever owns the pipeline spreadsheet holds quiet power — they're the person everyone has to call. Take that away without replacing it and the sheet lives on in a drawer. Say it early, out loud: "I want you running the board in the new system, and I want the sheet dead on the 30th. Same job, better tools — and I need you to be the one who kills it."
The honest counterpoint: if you're a three-person shop, a sheet may genuinely be fine. Sprawl becomes a tax where handoffs multiply — several LOs, a processor or two, a recruiter, a manager who isn't in production daily, all writing status at once. That's when reconciliation minutes cost more than any switch would.