AI quote preparation from job notes.

From site visit to draft quote

Quote preparation turns what a tradesperson captured on site into a quote the owner can check in minutes. The inputs are messy: a voice memo recorded in the van, six photos, a measurement or two in the notes app, and the client’s original email. The output is a draft in the job management system with scope lines mapped to the firm’s price book, labour estimated from similar past jobs, assumptions and exclusions written out, and a short list of questions where the notes are incomplete.

The aim is a faster, more consistent reply without the owner retyping notes late at night. This is workflow automation with a person at the price.

How the drafting works

Inputs

Voice memos, typed site notes, photos, measurements, the client’s request, the firm’s price book and labour rates, current supplier price lists, standard exclusions by job type, and past quotes for similar work.

Systems

Simpro or ServiceM8 through their developer interfaces with permission to create draft quotes, a transcription model, a language model, and the supplier catalogue.

Steps

Transcribe the memo; extract scope items and measurements; send a missing-information checklist back to the tradesperson; map items to price book codes; estimate labour from past jobs of the same type; add exclusions and assumptions; check the GST-inclusive total against the residential threshold; create the draft.

Human review

The estimator or owner checks every line, compares quantities with the photos, sets margin, and approves anything unusual. A person sends the quote.

Outputs

A draft quote with mapped lines, an assumptions and exclusions list, questions for the client, and flags for consents, licensed work or the residential threshold.

Measures

Hours from site visit to quote sent, share of lines edited by the reviewer, and on completed jobs, variations caused by scope the quote missed.

Risks

Guessed quantities, stale prices, missing exclusions, provisional sums presented as fixed prices, and skipped regulatory steps.

Illustrative example: a seven-person Palmerston North plumbing firm

Illustrative example, synthetic numbers: the firm sends 140 quotes a month. The owner spends about 35 minutes per quote typing up notes and pricing. With drafts, checking takes about 12 minutes, and 30% of jobs need a 5-minute follow-up for missing details.

Transcription assumes a 4-minute memo on gpt-transcribe at US$0.0045 a minute. Drafting assumes 18,000 input tokens (six photos at an assumed 1,500 tokens each, notes, a price book extract and instructions) and 1,200 output tokens on gpt-6.1-sol at US$2 and US$10 per million tokens. Prices are as listed on 19 October 2026.

NZ rules a residential quote has to fit

• The $30,000 threshold. Regulation 4 of the Building (Residential Consumer Rights and Remedies) Regulations 2014 sets the prescribed minimum price at $30,000 including GST. The workflow checks the GST-inclusive total, not the exclusive one.
• Disclosure before contract. Regulation 5 requires the prescribed disclosure information and checklist to be given before entering a residential building contract when the price is at or above the threshold, or whenever the client asks.
• Contract content. Regulation 6 lists what such contracts must contain, including the price or how it will be calculated, the number, amounts and due dates of payments, and how variations will be agreed. A draft quote should already separate fixed lines from provisional sums so the contract can say which is which.
• Photos and privacy. Photos of clients’ homes can capture people, documents and belongings. Keep them in the job system and follow the Privacy Commissioner’s guidance on AI and the privacy principles for what you send to a model.

What the owner or estimator keeps

The price and margin, judgement about site conditions (access, the state of existing pipework, possible asbestos), whether a consent or licensed practitioner is needed, the wording sent to the client, and every later variation. Our guides to Simpro and ServiceM8 cover keeping the system’s permissions to drafts only.

Where AI quoting goes wrong

• Plausible invented measurements. Notes say “bathroom, standard size” and the model writes 2.4 by 1.8 metres. Forbid invented quantities; a missing number becomes a question.
• Photos overrule notes. A mirror doubles the basins in a photo. Quantities come from the notes and the tradesperson’s answers; photos support exclusions and condition notes.
• Stale supplier prices. Copper went up last month and the price book did not. Flag every line priced from data older than your review period.
• Missing exclusions. Making good, rubbish removal and after-hours work drop out of drafts because nobody said them aloud. Use a fixed exclusions checklist per job type.
• Hovering near the threshold. A job drafted at $28,500 including GST can cross $30,000 when one line is added during review. Recheck the total every time the draft changes, before any contract is entered into.

Quoting connects to the other construction workflows we map: tender document extraction for larger commercial bids and a site safety knowledge assistant for crews on site. The construction sector guide sets out where to begin, and working with images in business AI explains what models can and cannot read from photos.

Written by the Levy Street team in Wellington. Last updated 19 October 2026.