AI for New Zealand construction.

Where AI earns its keep in a construction business

AI is most useful in construction where skilled people spend hours reading and rewriting documents: estimators working through tender packs, project managers chasing subcontractor paperwork, and office staff watching inboxes for claims and variations. It is weak at the physical judgement that makes building work safe and profitable, so good projects take the reading and first drafts off people and leave the decisions with them.

The sector is large and uneven. Construction contributed about $12.6 billion to Auckland’s GDP and $4.7 billion to Canterbury’s in the year to March 2024, according to Stats NZ regional GDP data (our ranking of its industry rows). Adoption follows firm size. MYOB’s Business Monitor, a vendor survey of 1,026 New Zealand business owners and managers reported by IT Brief on 4 August 2026, found 30% of businesses with one to five staff proactively using AI, against 64% of those with 20 or more. There is no official figure for construction specifically.

Construction workflow map

The first three rows have worked guides: extracting obligations from tender documents, preparing quotes from job notes and building a site safety knowledge assistant.

Which New Zealand rules shape AI use on site and in the office

Neither WorkSafe nor any construction body has issued AI-specific guidance. Existing law applies, and several rules bite harder once software reads your inbox or answers your workers.

• Health and Safety at Work Act 2015. The primary duty in section 36 stays with the PCBU, so far as is reasonably practicable. A safety assistant is part of how you manage risk, not a transfer of that duty. The Act also requires you to engage workers when proposing changes that may affect their health or safety, which includes new tools on site. WorkSafe’s psychosocial risk guidance (April 2025) lists over-reliance on technology and constant digital surveillance as hazards, which matters for camera analytics and worker monitoring.
• Construction Contracts Act 2002. Under section 22, a payer who does not provide a payment schedule within the contract’s timeframe, or 20 working days after the claim is served if the contract is silent, becomes liable to pay the claimed amount. An AI triage step that files a payment claim as general correspondence creates real exposure.
• Quotes and estimates. The Commerce Commission says “A quote is an offer to do a job for a certain price”, and that changing an accepted quote needs the customer’s agreement. AI-drafted pricing must be labelled correctly before it leaves the office.
• Residential building contracts. The Building Act 2004 requires prescribed disclosure information and a checklist before some residential contracts (s 362D), which must be in writing and dated (s 362F). Check generated contract packs against both.
• Privacy and hiring. Incident reports are personal information. The Privacy Commissioner expects a privacy impact assessment before AI tools are used, and human review before acting on outputs. MBIE’s Responsible AI Guidance for Businesses uses a hypothetical construction firm, “BigBuild”, trialling AI CV screening to show how hiring tools can discriminate.

This is general information. Get legal advice on contract terms, claims disputes and employment matters.

The software NZ builders already run

Most firms hold their data in a few systems: Procore or similar for commercial projects, Simpro or ServiceM8 for trade jobs and quoting, Xero or MYOB for accounts, and Microsoft 365 or Google Workspace for documents. AI works best connected to these, not as a separate place to paste text.

Two vendor developments are worth knowing, as listed on 19 October 2026. ServiceM8 says it is available as an approved plugin in ChatGPT and can be added to Claude as a custom connector, using MCP, so an assistant can search jobs and clients with the user’s permissions. Simpro’s AI pledge states it will not share customer data with third parties to train AI models without express permission. Check your own plan’s terms. Our integration notes for Simpro, ServiceM8 and Xero cover the APIs and permission settings.

A first project worth doing: tender document extraction

For a commercial builder, tender extraction is the cleanest start: bounded inputs, an estimator who already does the work, and last year’s tenders as a test set with known answers.

1. Build the answer key. Take 10 recent tender packs and the estimator’s own notes on each. Agree a checklist of 20 to 30 items: dates, retentions, liquidated damages, insurance, separable portions, special conditions, exclusions.
2. Run extraction on the old packs. Score each item as found, missed or wrong, and check every page reference.
3. Fix the misses. Most come from addenda, scanned pages and tables, so fix document handling before the prompt.
4. Pilot on live tenders. The estimator reviews every output against the source for a month and logs corrections.
5. Decide with numbers. Keep it if high-consequence recall is complete and review beats manual reading.

Measure three things: recall on the high-consequence items (liquidated damages, retentions, insurance and time bars should never be missed), estimator hours per tender, and items the system found that the estimator had missed.

Illustrative example: a 30-person Wellington commercial builder prices six tenders a month, and reading and summarising each pack takes an estimator about six hours, so 36 hours a month. If a reviewed first pass cuts that to 2.5 hours per tender, the saving is 3.5 × 6 = 21 hours a month. At an assumed loaded cost of $95 an hour, that is about $2,000 a month (NZD) before running costs. Model fees are small next to estimator time; integration and review are the real costs. Our AI document processing service and the AI pilot scorecard set this up in more detail.

Where AI gets construction work wrong

Related work: PainPal

Levy Street worked with BetterWorx on PainPal, an AI “digital workmate” for musculoskeletal pain and injury prevention, launched by BetterWorx (a subsidiary of Construction Health and Safety New Zealand, CHASNZ) in December 2025. It answers from expert-reviewed guidance and knows when to send someone to a professional, the same discipline a site safety assistant needs. To choose between these workflows, start with an AI consulting conversation.

Workflows we map in this sector

• Tender document extractionTender packs turned into a dated brief, a returnables matrix and a risk register, each line linked to its page.
• Site safety knowledge assistantWorkers ask about the site’s own safety rules and get a cited answer, or a clear pointer to their supervisor.
• Quote preparation from job notesSite notes, photos and voice memos turned into a draft quote from your own price book, checked before it goes out.

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