AI for estimating and quotes: let it write, never let it set the price

Estimating is where a contractor makes or loses money before the job starts. It combines things AI can't see (the condition of the house, the access, your crew's speed, this month's material costs) with things it handles well (organizing notes, writing clearly, spotting gaps in a list). The safe split is simple. You decide the numbers. AI helps you present them, check them and follow up on them.

Why can't an AI tool set your price?

A price is your costs plus your margin, adjusted for risk. An AI chat tool knows none of the inputs. It doesn't know what you pay your crew, what your truck and insurance cost per day, what your supplier charges this week, or that this particular driveway slopes toward the garage and needs extra prep. When you ask it to price a job anyway, it produces a number that sounds reasonable, built from general text it was trained on. NIST's generative AI profile calls this kind of output confabulation: confidently stated content that is wrong.

A wrong price fails in one of two directions. Too high, and you lose the job to someone who priced from real numbers. Too low, and you win a job that loses money. Neither shows up until it's too late. That's why pricing stays with your own numbers. Our guide to pricing service jobs walks through building those numbers, and job costing covers checking them against what jobs actually cost.

Where does AI actually help with estimates?

Notice the pattern in that table. AI sits next to every step that involves words or arithmetic you can verify, and is absent from every step that involves judgment about cost, risk or the law. If a tool you're evaluating wants to cross that line, for example by suggesting a price from a photo and your zip code, treat its number as one more opinion to check, not as an estimate. The estimate is the document you'll be held to, so it has to be built from information you can stand behind.

How do you turn site notes into a clean scope?

Most estimators come back from a walkthrough with a phone full of half-sentences and photos. That's the most useful place to start. Dictate or type the notes, strip out the customer's name and address, and ask the assistant to organize them into a numbered scope with one task per line, using only the tasks and materials you mentioned.

Then ask a second question: "What in these notes is unclear or missing?" This is where AI earns its keep. It's good at noticing that you listed tear-off but no disposal, primer but no caulk, a new panel but no permit line. It doesn't know whether those items apply to this job. You do. Treat its list as a prompt for your own memory, and decide each item yourself.

Some estimators keep a written checklist per service (prep, protection, materials, labor, disposal, permits, cleanup, warranty). Paste that checklist into the request and ask the assistant to compare the scope against it. That way the check is based on your standards instead of whatever it has read online.

Can AI measure from photos or plans?

Photo-measurement and plan takeoff tools exist for roofing, siding, paint, flooring and similar trades, and many contractors find them handy for a same-day ballpark. Use their output as a first pass. Before you order materials or sign a fixed-price contract, verify the measurements that drive cost: square footage, pitch, linear feet of trim, number of openings. Photos hide things. A second layer of shingles, rotted sheathing, or a wall that isn't square won't show up in a measurement report.

General chat assistants are weaker still at reading measurements from an ordinary photo. They can describe what's in a picture. They can't reliably tell you how big it is. Don't let a chat assistant's guess at square footage reach a quote.

Where a chat assistant is reliable enough to use is arithmetic you can verify: converting square feet to squares, adding waste factors you specify, totaling line items. Even then, spot-check the math. Language models can slip on arithmetic, and a spreadsheet or your estimating software is the better tool for totals.

How should AI help write the quote itself?

A homeowner holding three quotes tends to trust the one they can actually read. AI is good at translating trade language into plain English. Give it your line items and ask for a one-sentence explanation of each, written for someone who has never done this kind of project. Then read every line. It will sometimes describe a step you don't perform or a material you don't use, because that's what most quotes it has seen included.

It can also help you present options. If you already have three tiers with prices, it can write short descriptions explaining who each tier suits. Our good, better, best quotes guide covers how to structure the tiers. The prices and contents of each tier stay yours.

Two places to slow down. Exclusions and change-order terms are contract language, so have an attorney in your state review the standard version once, then reuse it. And warranties: if AI writes "lifetime warranty" when your manufacturer covers 25 years and you cover 2, you've made a promise you can't keep.

What estimate details should stay out of AI tools?

An estimate carries a customer's name, home address, phone, sometimes photos of the inside of their house, and occasionally details about their finances or health ("needs grab bars, recovering from surgery"). None of that is needed to get a good draft. Replace it with a description ("two-story colonial, north elevation") before you paste anything into a consumer AI tool, and read that vendor's data policy before you rely on it for business work. Some vendors are explicit that people may read what you type: Google's Gemini privacy hub notes that trained reviewers look at some conversations to improve its services.

Photos deserve the same care. A photo of a front door with a house number, a car plate in the driveway, or a family picture on the wall identifies a customer just as well as a name does. Crop or skip them.

How can AI help after the quote goes out?

A quote nobody follows up on is a quote you paid to write for nothing. AI can draft a short, low-pressure follow-up for day two, a reminder with a real scheduling detail for day seven, and a final check-in later. Write them once, edit them until they sound like you, and save them as templates. Our quote follow-up guide covers timing.

The better long-term setup is to keep these templates and your price book inside your own system, so a quote is built from your stored line items and the follow-ups go out on schedule without anyone copying details into a chat window. That's the kind of setup we build into a custom CRM: your numbers, your wording, stored on systems you control.

A simple estimating workflow with AI in it

1. Walk the job. Take notes and photos as you always have.
2. Back at the truck, strip personal details from the notes and ask an assistant to organize them into a scope and list anything unclear or missing.
3. Decide each flagged item yourself. Measure again if needed.
4. Price the scope from your price book or estimating software. No AI in this step.
5. Ask the assistant to rewrite line items in plain English. Read every line.
6. Add your standard terms, exclusions and warranty, unchanged.
7. Send it, then let your saved follow-up templates do the reminding.

Done this way, AI saves time on the writing without ever touching the number that decides whether the job makes money.

Frequently asked questions