Seven days to get a number out the door

Friday morning I'm on a call with a landscape contractor here in the Bay Area, looking at an estimate for a job he's pricing. Not one he wrote. One the tool wrote, straight off the plan, with nothing behind it but the drawing.

First pass comes in around $400,000. He looks at it and says it's not far off.

Now I've never priced a yard in my life. I don't know what steel edging costs or how long a crew takes to set a base, and neither does the thing that built that estimate. It read a plan and did arithmetic. And it landed close enough that a man who prices these for a living recognised the number.

Everybody wants to talk about whether the number's right. His actual problem is seven.

Seven days. That's how long it takes him to go from walking a site to an estimate leaving his inbox.

The four month job

He took a job once that he estimated at eight weeks. It ran four months. He ate the difference.

So his rule now is to pad the big ones. Deliberately, and he says it flat, the way a man quotes a policy he wrote himself. He'd rather miss the work than come in under on a half million dollar contract.

Fair play to him, that's a scar. And any tool that turns up wanting to shave the padding off is going to get walked to the gate, and it should be! The padding is the only thing standing between him and that four month job happening to him twice.

So what I'm building doesn't touch his numbers. It doesn't decide what a job is worth and it doesn't know what his crews cost. It reads the plan, builds out the line items, and puts a draft in front of him. He pads it exactly the way he always has. He just gets to do it on Tuesday instead of the following Monday.

The forty foot bed

The part of the call that actually mattered was one correction.

I'd demoed a voice memo workflow. You walk the site with your phone, talk it through the way you'd talk it through to a foreman, and it comes back with an estimate and a draft email to the client. On the recording I'd called a bed at forty feet, then corrected myself to fifty.

The output took fifty. One man changing his mind out loud, and the thing going with the change instead of the first number it heard.

Because what an owner is quietly waiting for is the moment it invents a number on him. Buries a figure in a line item he'll never check, and he only finds out when the crew is on site and the material is short. You can't talk him out of that fear with a page about accuracy. You can only let him watch it not happen.

Eight pages was the wrong instinct

Then he told me the estimate was too long.

Eight pages. I'd built the detail in on purpose, because detail reads as rigour, and I assumed a client seeing every line would trust the number more.

Wrong way round entirely. He wants the line items grouped. Hardscape. Lawn. Planting and prep. Irrigation. Steel and fencing. Five sections, and a homeowner reading it can't sit there picking at individual numbers.

And he's not hiding anything. He's stopping a line by line negotiation over a number he already padded for good reason, with someone who has never built a retaining wall and is about to ask why the irrigation costs what it costs.

That reframed the build for me. I'd been treating the estimate as a document that proves the tool works. It's a document that has to survive a Saturday morning at a kitchen table, with a homeowner and a printout and plenty of time to read it.

What this looks like for a landscape contractor in the Bay Area

If you're running a landscape or hardscape crew anywhere around Marin, the week is the thing worth attacking. Not the number. You already know the number, or near enough, within about an hour of walking the site. What you don't have is a free evening to type it up, because the typing up is the bit that waits until Sunday.

So the honest version of what AI does for a landscape contractor in the Bay Area right now is boring and worth money. It turns a plan set and a three minute voice memo into a first draft with the line items already grouped the way you like them. You correct it. The corrections go in as rules, so the next one comes out closer. And the won and lost jobs already sitting in a folder on your machine become the context for the next estimate, so after a while it can flag where your pricing drifts by area. Nobody's replacing your pricing judgement, and if anybody tells you they are, they've never lost four months on an eight week job.

Seven days down to one or two.

If you want to see it on one of your own jobs

Send me a plan set and the estimate you built off it. I'll run the plan through and send you back what comes out, and you can tell me how far off it is. No charge, and you keep the output either way.

I'm in Tiburon. So if you're anywhere in Marin or the North Bay and you'd rather watch it run than read about it, say the word and I'll drive to your yard with a laptop and run it there in front of you, on a job you already priced and already know the right answer to. You'll know inside twenty minutes whether it's any use to you.

There's another estimate going through it this week. I'll know then whether one or two days was optimistic.