The full build, start to finish — including the parts that went wrong.

I needed a seller net sheet.

Not eventually. That week, in front of a seller, with the clock already running.

My title company has an app that produces one. Good app. I’ve used their output for years. But using it meant waiting on someone over there to grant me access — and access is somebody else’s to give, on somebody else’s timeline. Deal timelines don’t pause while a vendor works through an onboarding queue.

So I stopped waiting and built my own in Claude.

Here’s the part I want you to take away, because it’s the whole thing:

I uploaded a sample net sheet.

Not a description of one. Not a list of what I wanted. An actual net sheet from a real transaction — the same fields, the same line items, the same calculations I’d have gotten out of the title company’s app.

That one upload did two jobs at once. It was the specification, so nothing had to be guessed at or invented. And it was the answer key, so there was a known-correct result to check the finished tool against.

That’s the difference between “build me a net sheet calculator” and getting something I could actually put in front of a client.

The output matched the title company’s exactly. No discrepancies.

Every line item, to the penny. The owner’s title policy. The per-diem interest on the payoff. The prorated taxes. Both commission lines. Total closing costs. Net at close.

Not close enough. Identical.

That’s the check that mattered, and it’s the one I’d insist on before trusting any tool with a client’s numbers — including one I built myself.

It also calculates the owner’s title policy straight from the Texas promulgated rate schedule, so the premium is right the moment I type a sales price. It exports a clean one-page PDF with two signature lines for two sellers. And it’s available at 9pm on a Sunday without asking anyone.

The title company’s app produces the title company’s document.

Mine produces mine.

That sounds small until you think about what a net sheet actually is in our business. It’s not a worksheet. It’s a document a seller keeps. They read it at the kitchen table. They forward it to a spouse, to their accountant, to the attorney handling the estate. Sometimes to another agent they’re still deciding between.

Whatever they forward carries somebody’s name and face on it. For years, that wasn’t mine.

Now it is. My headshot, my seal, my contact information, my colors — sitting at the top of the page a seller is looking at while they decide whether the numbers work. Same math as before. Completely different document.

That’s the piece I’d underline for anyone in this industry. We spend real money on branded listing presentations and branded market reports, and then hand a client the single most consequential page in the transaction with someone else’s logo on it. The numbers were never the only thing on that page.

Claude matched my branding exactly — the headshot, the seal, the palette, the layout — because I gave it my brand assets and told it what to do with them. That took one round.

It wasn’t one prompt and done, and I’d be selling you something if I said it was.

Three things went wrong before it was right, and one of them is worth your attention:

A field quietly stopped updating. To get started, the tool pre-filled the prorated tax figure from my sample. Correct for that deal. But it had frozen the field — I changed the annual taxes and the number didn’t move. No error. No warning. Just a stale figure sitting there looking finished while everything around it changed.

That’s the failure mode to watch for. Not the ones that break loudly. The ones that keep working and keep being wrong.

The PDF came out unusable. The layout collapsed into the narrow phone version and got shrunk onto one page. Type the size of a footnote. The build reported success — the checks it was running all came back clean. They were checking the wrong thing. It took me sending the bad file back and saying look at it before the real cause got found.

My branding vanished after an update, because settings saved from an earlier version overwrote the new ones.

All fixed. But every one of them surfaced because I checked, not because anything flagged it.

Bring a document, not a description. The sample net sheet is the reason this worked. If you’re building something that already exists in some form — a checklist, a report, a spreadsheet you get from a vendor — upload the real one. You’ll get something that matches instead of something that approximates.

Test against an answer you already have. I didn’t ask whether the output looked right. I asked whether it matched a document I could check it against, line by line. If it can’t reproduce a known result, don’t trust it on an unknown one.

Open the actual file. Not the summary. Not the confirmation that it worked. The file. I lost a round trip accepting “verified” instead of opening the PDF myself.

This isn’t really about net sheets.

It’s about how many things in this business you’re waiting on someone else to hand you. A login. A template. A report that takes three days. Every one of those is a place where your timeline is set by somebody else’s queue.

Some of that waiting is unavoidable. A lot more of it isn’t than I assumed a year ago.

Systems built by a producing broker.

I document these builds as I do them — what worked, what broke, and what it actually took. Tools like this one, transaction workflows, and the admin systems that got me more than 130 hours back across five files.

Sign up at backbyfive.com

Real estate doesn’t end at five. Your admin should.