Bidding a Lawn From Space, and Invoicing the Day It's Cut

Quoting a new commercial property meant getting in a truck, looking at the lawn, and guessing at the areas that matter. That put a hard ceiling on how many properties could be bid in a week, and it made bidding a large portfolio impractical.

We built an estimating pipeline that starts from an address. It geocodes the property, pulls the county's own recorded parcel polygon, fetches satellite imagery sized to the lot, reads that image with AI to separate turf from canopy and beds, samples a USGS elevation grid to derive slope, and prices the result against a transparent model.

Properties are bid from a desk in seconds, individually or as a pasted batch, with a per-cut price and an exportable spreadsheet.

A pricing model that depends on a language model reading an aerial photo has a known failure mode: it will occasionally call a large house on a small lot a commercial property, and misprice the job.

We cross-check the model against physical reality. Parcel acreage overrides a commercial classification when the land does not support it, because a real commercial mowing job sits on real land. Every measurement carries a method and a confidence, and a low-confidence bid carries an internal advisory to walk the property before it goes out, so the estimator never has the tool's uncertainty hidden from them.

Every measurement carries its method and a confidence level, and the bid surfaces its own weak spots to the estimator instead of presenting a shaky number as a certain one.

A route is only right if it holds exactly the properties the client is being billed for. Hold the drive order and the billing roster in two places and they drift: a property gets added to the account and never makes the drive, or a stop keeps getting cut after it comes off the books. Stops in the order they were added also produce long backtracking drives between neighbors.

We brought routing in-house and tied it to the customer records. A 2-opt optimizer refines the drive order on an open path, and adding a property automatically slots a stop into the route while removing one archives it, so the route and the billable roster can never drift apart.

The daily drive order is derived from the same records that bill the client, so a property can't sit on one list without sitting on the other.

Cuts were recorded one place and invoiced another, by hand, on a monthly cycle. For a per-cut commercial account that delay pushed payment out by weeks.

The crew works the route in an installed phone app that tolerates dead zones, tapping Navigate and then Complete, which stamps the cut with GPS and queues it. An end-of-day digest emails the client one invoice per cut, while monthly accounts stay on a scheduled batch. Both cadences read one field, so no account can ever be billed twice.

For per-cut accounts, invoices leave the same day the work happens instead of at the end of the month.