Most software bolted a chatbot on the side. Crewplate's AI lives inside the ledger — it reads your real numbers, drafts real documents, and every action prints a receipt you can check. When it can't help, it says so and adds four honest words: Books untouched.
"2-story vinyl, about 2,400 square feet, gutters, 600 of driveway." The composer drafts the lines from your price book — it estimates quantities and never invents prices. You adjust, client signs.
Walk the property with your phone and let the scan read it. All four trades: stains and surfaces on a house, defects panel by panel on a vehicle, roofline feet and window counts on a lights job, turf and bed square footage on a yard. Those measurements become quote lines priced from your book — and they come out deliberately short, because it underestimates rather than over. (A quick check spends one AI credit; a full walkaround spends two.)
A full walkaround runs on the deeper model, and it doesn't hand you the first draft. It re-examines every finding against your photos before answering — can it still see that stain, on that wall, in the photo it cited? What it can't confirm gets dropped; what's genuinely ambiguous comes back marked uncertain instead of confident. A short list you can stand behind beats a long one you have to walk back. Checking finished work is the same discipline: three before photos against three after, and an honest list of what got missed.
It won't pretend a picture is a tape measure or a paint gauge. Dent depth, coating thickness, whether a scratch went through the clear — those come back as a plain line about what the photos cannot confirm, not a number you'd have to defend in a driveway. Glare and water drops imitate damage, so a doubtful finding gets flagged rather than invented. And when you delete something it got wrong, that correction is logged — real misses are what sharpen what it looks for next.
Pick friendly, firm, or final — it writes the payment nudge with the invoice number and balance, under 55 words, no threats, ready to send from your own phone.
Three bullets before your first coffee: the money that moved, the schedule that matters, one concrete action. Numbers first, zero fluff.
Job's done — how did the money actually land against the quote? Hours, costs, and one specific price-book suggestion, or an honest "no change needed."
Before a quote goes out, a second set of eyes for typo smells: quantities off by 10×, unit mismatches, a missing deposit on a big job. It knows your catalog, not "market rates" — and says so.
"Add the Hendersons, then quote them a house wash." Two receipts appear under the answer — client added, quote drafted — because the Assistant runs the same books you do. (The ✦ Assistant dock is live on every plan.)
Teach it once — your prices, policies, warranty terms, the way you word things — and every ✦ answer speaks as your business, not a generic bot. "We soft-wash roofs, never pressure-wash them" sticks, across quotes, replies, and marketing.
Draft a before/after social post, a segmented email blast, or a win-back note for a client who's gone quiet — in your voice, from your actual jobs. Nothing sends until you say so.
The equipment AI: log a repair by talking, work through a symptom before you call the shop, or plan the next maintenance — right on the Service page, so a pump doesn't die mid-driveway.