Skip to content
Book AI Assessment

[BLOG / TRADES]

AI Estimating for Contractors: Setup, Price Book and Approval Rules

The setup guide for an AI estimating agent at a contracting company: what goes in the price book, which quotes send on their own, how the estimate reaches your CRM and the customer, and a worked panel-upgrade example.

Sebastian Alidad · September 24, 2026 · 9 min read

A blank clipboard, a coil of copper wire and a tablet glowing orange sit on a workshop bench below a closed breaker panel.

Questions this post answers

The short answer

Setting up an AI estimating agent for a contracting company takes four decisions: a written price book, a list of intake questions per job type, rules for which quotes need your approval, and a place in your CRM where drafts land. The agent then prices from your rules, not from guesses.

This is the setup guide, not the buying guide. If you are still deciding what an AI quoting agent is and whether you need one, start with what to look for in AI quoting software. Here we assume you have decided, and we cover what you write down, what you approve, and what the first month looks like.

It is also not a takeoff tool. Nothing here reads blueprints or counts fixtures from a plan set. It covers the residential and light-commercial jobs priced from a menu of work, a few site facts and your rules.

How do I set up an AI estimating agent for my contracting company?

Four decisions get an estimating agent running: the price book, the intake questions for each job type, the approval rules, and where drafts land. The build follows from those decisions, which is why the first week is mostly you and a spreadsheet.

For context, the Census Bureau's AI use survey of businesses shows that between December 2025 and May 2026 AI use "didn’t change significantly among firms with fewer than 20 employees." Among commercial contractors, ServiceTitan and Thrive Analytics found that contractors are applying AI to "cost estimation and budgeting (24%) and bid management (22%)". Estimating is one of the first places contractors reach for it.

The four setup decisions

  1. Price book. Every line the agent may charge for, with the rule that decides the quantity or the rate.
  2. Intake script. The questions for each job type, and which answers are required before a draft is possible.
  3. Approval rules. The conditions that send a draft to you before the customer sees it.
  4. Destination. The CRM or job record the draft is written to, and the message that carries it to the customer.

What goes into the price book an AI estimator works from?

Everything your best estimator keeps in their head, written as lines and rules. A price book for an agent has five parts: labor rates, material costs with your markup, fixed charges such as permits and trip fees, add-ons you always offer, and the conditions that change a price. If a line has no rule, the agent has no way to price it, and a good build says so instead of guessing.

Labor and materials

Write labor as a rate per job type or per hour, and say which. Write materials as your supplier cost plus your markup, and note the date you last checked supplier pricing. Material costs move, so the book needs an owner and a review date.

Fixed charges and add-ons

Permit fees, trip charges by zone, disposal, minimum job charges and the add-ons you list on most jobs (surge protection on a panel job, a haul-away on a demo). Add-ons matter more than they look. Housecall Pro's 2026 State of Home Spending report says "68% will shop around if a price comes in higher than expected," and advises that "A single-line estimate invites comparison. A detailed estimate with good-better-best options changes the decision from whether to hire you to which solution to pick". A price book with tiers lets the agent draft that second kind of estimate every time.

The conditions that change a price

These are the rules people forget to write: access is tight, the panel is in a finished wall, the home is pre-1980, the work is above a certain floor. Each one becomes either a line (a labor factor) or a stop (the agent asks for a site visit instead of a price). Writing them is also the quickest way to find where two of your estimators price the same job differently.

Which quotes can the AI send on its own and which need my approval?

Start with every quote needing your approval, then loosen one job type at a time. A fixed-price, low-variance job where the agent's draft matched yours over several weeks can earn auto-send. Anything with a site condition, a permit, a large total or a customer who has been quoted before stays with you. The rules below are an example, not a standard.

An example approval ladder

Example rules, not data from a real company:

SituationWhat happens
Fixed-price service, total under the auto-send ceiling, all intake answers present, no flagsSends on its own, copy to you
Total above the ceilingDraft held for your approval
Any intake answer missing or "not sure"Agent asks a follow-up, or books a site visit
Any flagged condition (old wiring, tight access, finished walls)Draft held, flag shown at the top
Repeat customer or open disputeHeld for you
Job type not in the price bookRouted to you, no price drafted

The ceiling is your call. Pick a number you would sign without reading, and start it at zero for the first weeks.

Expect the rules to need adjusting in the first weeks, and treat approval as the control that lets you adjust them safely.

How does an AI estimate get into my CRM and in front of the customer?

The agent files the draft on the customer's job in your CRM or field-service system, where that system offers a way to write records; the first thing we check is which fields can be written and which can only be read. It then sends the approved version to the customer by the channel they used to ask. That keeps intake, quote and customer on one record. The customer gets an estimate that looks like yours, with your logo, your wording and your terms.

What lands in the CRM

  • The intake answers, in fields, not buried in a note
  • The drafted line items with the price-book rule that produced each one
  • Any flags and the reason for the hold
  • The approval status and who approved it
  • A follow-up task with a date

Because the rule behind each line is stored, you can answer "why is this $400 higher than last month" in one click. That audit trail is what makes it safe to loosen the approval ladder later.

What the customer receives

An email or text with the estimate, ideally in tiers, and a one-line way to accept. Then the second job starts: the estimate going quiet. ServiceTitan's 2026 roofing and exterior report found "Only 16% of contractors follow up with homeowners the same day for unsold estimates, despite most leads expecting rapid re-engagement." The quote follow-up agent covers that stage, and our guide to a quote follow-up system covers how the cadence works. If you run a trades business, the trades and home services page shows the other systems that connect to the estimate.

What should the first weeks of an AI estimating agent look like?

Slow on purpose. Treat the first weeks as a trial where you check every draft against what you would have written, and the agent's job is to match you. A rough sequence is below. The build timeline for your company comes out of discovery, so treat this as the shape, not a schedule.

Week one: the book and the questions

You write the price book and intake script. We turn them into rules and test them against old quotes you already sent, so the first drafts are compared with known answers instead of live customers.

Weeks two and three: shadow mode

The agent drafts on live requests, and nothing goes to a customer without you. You mark each draft: matched, close, wrong. Every "wrong" becomes either a missing price-book line or a missing rule. Expect most of the early misses to be rules you never wrote down, not model mistakes.

Weeks four onward: loosen one lever

Pick one fixed-price job type with a clean record, set a low auto-send ceiling, and keep copies coming to you. Watch three numbers: how many drafts you change, how long a request waits before the customer gets the estimate, and how many estimates get a reply. Change one rule at a time so you know what moved the numbers.

If a job type keeps producing edits, it stays on approval. That is the system working, not failing.

An electrical contractor asks: "How would an AI agent quote a panel upgrade?"

A homeowner texts: "We want to swap our 100 amp panel for 200 amp. House is from 1978." Here is the walk-through with every figure an example.

Made-up figures for one job (yours will differ):

  • Home built 1978, panel in the garage, mast and meter on the same wall
  • Price-book line: 200 amp panel replacement, $2,800 base
  • Add-on: whole-home surge protector, $350
  • Rule: homes built before 1980 add an inspection allowance, $400
  • Permit and inspection fee line: $250
  • Approval rule: any panel job is held for the owner

Step 1: intake

The agent asks what the price book needs: current amperage, new amperage, panel location, whether the meter and mast are moving, and a photo of the existing panel and the wall around it. The customer answers by text in about five minutes.

Step 2: price-book lines and rule checks

From the answers the agent builds the draft:

LineSourceExample amount
200 amp panel replacementBase line$2,800
Pre-1980 inspection allowanceCondition rule$400
Permit and inspection feeFixed charge$250
Whole-home surge protector (optional)Add-on$350
Total with add-on$3,800

The rule check runs next. Garage location and no meter move pass. The photo shows a cloth-sheathed cable entering the panel, which matches the "old wiring" flag, so the draft gains a note: "Possible additional work if wiring needs replacing; confirm on site."

Step 3: the approval threshold

Panel jobs are held, and the flag makes it doubly so. The draft lands on the owner's screen with the intake answers, the photo, the flag and the rule behind each line. They change nothing, approve, and the estimate goes out.

Step 4: the estimate and the follow-up

The customer receives two options, not one: the base upgrade at $3,450 and the upgrade with surge protection at $3,800, each with scope notes and the wiring caveat. A follow-up task is created in the CRM for the next day. If the customer has not replied, the follow-up agent sends a short check-in that references the option they looked at.

The owner's time on this request: a minute to review. The customer's wait for the estimate: the time it took the owner to open a phone. The agent did not quote anything the price book did not cover, and it did not hide the one condition that could change the price.

Unsure whether your rates are ready to be written into a price book? Talk it through with us. Should a discovery engagement make sense for your estimating, you will hear it directly. The call itself stays a conversation, with nothing reviewed or priced.

Sources

[WRITTEN BY]

Sebastian Alidad

Founder of Built to Spec AI, an Irvine, CA studio that specs, builds, and ships custom AI systems for small businesses.

See where AI can save you time

Book a free AI audit. Tell us where the work piles up, and we’ll talk through whether automation belongs there. No forms. No waiting on us.

Build notes, occasionally