Why checking a landscape package takes about 2 weeks
Before a firm can price a job, someone has to check what the developer sent. On a landscape package, that takes about 2 weeks by hand. I've seen it done, and the idea for LandSight came out of conversations with property developers. Landscape is where LandSight started, so that's the package I'll describe here.
From the outside, it sounds like reading some drawings. It isn't.
What arrives is a folder, not a document
You get whatever the developer's team had on hand: CAD sheets, scans, Excel bills of quantities, backups and old revisions, often all in the same folder. Nobody cleaned it up for you. The first job is working out what each file is and which bill actually counts.
The counting is the slow part
Then someone counts every tree, measures every planter and checks every line of the bill, sheet by sheet. The drawings don't make it easy. On a CAD sheet, a tree is often 12 loose lines. A hedge can be a single text note that stands for hundreds of plants. A colored presentation plan usually has no CAD behind it at all, so the picture is all you have to measure.
The mistakes sit between the sheets
Most sheets look fine on their own. The errors show up when you compare: the bill against the drawing, the plan against the detail, each revision against the last, the price against the firm's own list, the planting against the local building code. That's why you can't skim it. The work isn't reading a page carefully. It's holding many pages against each other.
Why it has to happen before pricing
A mistake you find before pricing is a question you can send back to the developer. Find it after you've priced and it's your problem. So the check gets done, however long it takes.
What we changed, and what we didn't
LandSight does the counting and the cross-checking. It gives every file a verdict, counts and measures what's drawn, and checks the bill against the drawings, the firm's price list and the local building code. On a sample landscape package with 30 planted errors, it found 28, flagged 2 for a person and missed 0, in about 10 minutes.
That package is synthetic. Real drawings are under NDA, so I built sample packages with planted errors. We don't have results on real customer drawings yet, and these numbers say nothing about other trades.
What we didn't change is who decides. Nothing counts until a person confirms it, and every number opens on the drawing it came from. The judgment stays with the estimator. The counting is the part we want to take off their desk.