# AI for New Zealand manufacturing. URL: https://levystreet.com/industries/manufacturing/ Updated: 2026-10-19 AI for manufacturing in New Zealand: supplier documents, quality reports and maintenance knowledge, the data foundations, HSWA duties and failure modes. ## Short answer AI in New Zealand manufacturing is most useful on the paperwork around production: extracting data from supplier documents and certificates, triaging quality inspection reports, and answering maintenance questions from manuals and past job cards. Quality, safety and engineering decisions stay with people. Results depend heavily on data foundations, because many mid-sized NZ manufacturers still run core operations across spreadsheets and older ERP systems. - Supplier document extraction is a strong first project with a clear accuracy measure. - Only 36% of NZ mid-sized firms report a modern cloud ERP, per MYOB's 2026 research. - Keep AI away from machine controls; read from operational data, never write to it. - Introducing AI on the shop floor is a change that HSWA requires you to discuss with workers. ## Where AI helps a New Zealand manufacturer AI helps manufacturers most with documents and knowledge rather than machines: reading supplier certificates and packing lists, sorting inspection reports and non-conformances, and answering "how did we fix this last time" from manuals and job cards. Those tasks sit with planners, quality staff and maintenance leads who are expensive to tie up with typing and searching. The limiting factor is usually the data underneath. MYOB research with more than 500 New Zealand mid-sized business leaders, reported in April 2026, found only 36% ran a modern cloud-based ERP; 30% ran core operations across multiple systems and spreadsheets, and another 30% were on desktop or on-premise ERP software ([IT Brief, 27 April 2026](https://itbrief.com.au/story/new-zealand-mid-sized-firms-lag-on-ai-readiness-myob-finds)). The same research found 37% of firms with more than 100 staff reported improved margins from AI, against 11% of firms with 20 to 49 staff. Scale and systems both matter. Manufacturing is also concentrated. MBIE's 2026 factsheet notes that Auckland, Waikato and Canterbury [continue to be the main centres](https://www.mbie.govt.nz/dmsdocument/32448-small-business-and-manufacturing-fact-sheets-2026-pdf) for New Zealand manufacturing, so most of this work happens close to engineers who can walk the floor. Manufacturing was Canterbury's largest industry by GDP in the year to March 2024, at about $5.0 billion, on our ranking of [Stats NZ regional GDP data](https://www.stats.govt.nz/assets/Uploads/Regional-gross-domestic-product/Regional-gross-domestic-product-Year-ended-March-2025/Download-data/regional-gross-domestic-product-year-ended-march-2025.xlsx), and it employs about 77,300 people in the Auckland region (Stats NZ business demography, February 2025). ## Manufacturing workflows mapped | Workflow | What AI does | What a person still decides | Data involved | Difficulty | |---|---|---|---|---| | [Supplier document extraction](/industries/manufacturing/supplier-document-extraction/) | Reads certificates, packing lists and invoices into ERP fields, matches to purchase orders | Accepting stock with mismatches | Mill certs, CoCs, packing lists, POs | Low: fixed fields, checkable | | [Quality inspection report triage](/industries/manufacturing/quality-inspection-report-triage/) | Sorts inspection reports by defect type and severity, links to similar past cases | Disposition: use, rework or scrap | Inspection sheets, photos, NCR history | Medium: severity judgement | | [Maintenance knowledge search](/industries/manufacturing/maintenance-knowledge-search/) | Answers fault questions from manuals and job cards with citations | The repair and any lock-out | OEM manuals, job cards, technician notes | Medium: tacit knowledge gaps | | RFQ to quote drafting | Extracts specs from customer RFQs, drafts a bill of materials and questions | Price, lead time, feasibility | RFQs, drawings, past quotes | High: drawings and tolerances | | Non-conformance report drafting | Drafts the NCR from inspection data and operator notes | Root cause and corrective action | Inspection data, operator notes | Medium: cause stays human | | Shift handover summaries | Summarises downtime, issues and open jobs for the next shift | Priorities for the shift | Shift logs, downtime records | Low: internal summary | | Certificate tracking | Tracks expiry and coverage of supplier certificates and approvals | Whether a supplier stays approved | Certificates, supplier register | Low: dates and lists | | SOP questions | Answers operator questions from current procedures | Training sign-off, procedure changes | SOPs, work instructions | Medium: version control | The three detailed guides cover [supplier document extraction](/industries/manufacturing/supplier-document-extraction/), [quality report triage](/industries/manufacturing/quality-inspection-report-triage/) and [maintenance knowledge search](/industries/manufacturing/maintenance-knowledge-search/). ## Getting data out of Cin7, SAP Business One and MYOB Acumatica New Zealand manufacturers run a wide spread of systems: Cin7 and similar inventory systems for stock and production, MYOB Acumatica, SAP Business One or Microsoft Dynamics as ERP, and SharePoint or network drives for drawings and procedures. As listed on 19 October 2026, [Cin7](https://www.cin7.com/features/ai/) offers document recognition for invoices, purchase orders and supplier documents, and [MYOB Acumatica](https://www.myob.com/nz/erp-software/solutions/ai-erp) includes an assistant that answers plain-language questions about revenue, customers, inventory and financial performance. Use those features where they cover the job. Maintenance and SOP questions suit a [retrieval system with citations](/services/rag-development/) built over your controlled documents. Custom work is justified when the task spans systems, for example a certificate arriving by email that must be checked against a PO in [Cin7](/integrations/cin7/), filed in SharePoint and flagged in [MYOB](/integrations/myob/). That is the core of our [document processing](/services/ai-document-processing/) work. Before any build, list what data you hold and where; our [AI data inventory template](/resources/ai-data-inventory-template/) is a quick way to do it. ## Health and safety and product claims on the factory floor The [Health and Safety at Work Act 2015](https://www.legislation.govt.nz/act/public/2015/70/en/latest/) puts a primary duty on the business to ensure worker health and safety so far as is reasonably practicable (s 36), and requires engagement with workers when proposing changes that may affect their health or safety (s 60). An AI assistant that answers lock-out or chemical handling questions, or a camera that monitors a line, is such a change. WorkSafe's [psychosocial risk guidance](https://www.worksafe.govt.nz/topic-and-industry/work-related-health/mental-health/managing-psychosocial-risks-at-work/) lists over-reliance on technology and the use of digital surveillance to constantly monitor workers as psychosocial hazards. Plan for both. Product and origin claims written with AI help, on spec sheets, labels or websites, are still claims under the [Fair Trading Act 1986](https://www.legislation.govt.nz/act/public/1986/121/en/latest/), which requires reasonable grounds when a claim is made (s 12A). A technical specification drafted by a model needs the same engineering sign-off as one typed by hand. Get regulatory advice for any claim about safety standards or certification. ## A first project: supplier document extraction - **Inputs**: Supplier invoices, packing lists, certificates of conformity and mill certificates arriving by email or scanned at goods-in. - **Systems**: Shared inbox, ERP or inventory system purchase orders, document store. - **Steps**: Extract fields, match to the PO line by part number, quantity and unit, check certificate values against the specification, route exceptions. - **Human review**: Goods-in or quality staff clear every mismatch; a weekly sample of auto-matched documents is checked. - **Outputs**: ERP receipt drafts, certificate records linked to batch numbers, an exception queue. - **Measures**: Field accuracy on a labelled test set, auto-match rate, exception clearance time, receipts entered per hour. Illustrative example: a fabrication shop receives 600 supplier documents a month (an assumption). At 4 minutes each to key and check, that is 40 hours. If 70% match automatically and the remaining 30% take 3 minutes to clear, plus a 5-hour monthly sample check, the total is about 14 hours, saving roughly 26 hours a month. The figure that matters as much is how many certificate problems were caught before material reached the floor. ## Where AI fails in manufacturing - **Units and tolerances.** A model reads 0.05 mm as 0.5 mm or converts inches without saying so. Control: extracted values keep their original units, and validation rules reject out-of-range numbers. - **Drawing revisions.** Answers come from Rev C when Rev D is current. Control: index only the controlled revision, and show the revision in every answer. - **Tacit knowledge.** The fix that works lives in a senior fitter's head, not the manual. Control: capture it as notes and job cards before expecting search to find it. - **Operational technology.** An agent with write access to a PLC or SCADA system can create a physical hazard. Control: read-only exports, network separation and controls engineer review. - **Batch and lot errors.** Transposed digits link a certificate to the wrong batch. Control: check digits and exact match against the PO and goods-in record. Our engineering note on [structured outputs and validation](/engineering/structured-outputs-and-validation/) explains how extraction systems reject bad values before they reach the ERP. ## Questions people ask ### Do we need a modern ERP before using AI in manufacturing? Not for document work. Extracting supplier certificates, packing lists and invoices, or searching maintenance manuals, works with files you already have. Forecasting, scheduling and margin analysis do depend on clean, connected data, and many mid-sized manufacturers will get more from tidying their ERP than from adding AI on top of spreadsheets. Start with the documents while the systems work continues. ### Can AI do visual quality inspection on our line? Camera-based defect detection is a different project from the language-model work on this page. It needs labelled images of your own defects, controlled lighting and careful testing against your acceptance criteria, and it usually starts by assisting an inspector rather than replacing one. Using AI to triage written inspection reports and non-conformances is a faster, lower-risk place to begin. ### Is it safe to connect AI to our machines or SCADA system? Treat operational technology as off limits for AI agents with write access. Reading exported data such as downtime logs, sensor summaries or maintenance history is reasonable. Letting a model change setpoints, start equipment or alter interlocks is not, because errors become physical hazards. Keep a clear network boundary and have your controls engineer review any integration. ### How do we capture knowledge from experienced staff before they retire? Record it in the form AI can use: short voice or video explanations of fault-finding, annotated job cards and corrections to the manual. A maintenance search tool can then answer questions from that material with citations, so new technicians learn from it. Ask staff to review answers in their area; they will spot gaps in the source material faster than anyone else. ### What does AI cost to run for a mid-sized manufacturer? For document extraction and search, model usage is usually a small part of the cost. Integration with your ERP, exception handling and the time staff spend reviewing outputs dominate. Work out the cost per document processed and compare it with current handling time, using stated assumptions. Our [AI project cost guide](/services/ai-project-costs/) walks through the arithmetic, and prices should always be checked against current vendor lists. ## Sources - [New Zealand mid-sized firms lag on AI readiness, MYOB finds](https://itbrief.com.au/story/new-zealand-mid-sized-firms-lag-on-ai-readiness-myob-finds) - [Small business and manufacturing fact sheets 2026](https://www.mbie.govt.nz/dmsdocument/32448-small-business-and-manufacturing-fact-sheets-2026-pdf) - [Regional gross domestic product: Year ended March 2025 (data tables)](https://www.stats.govt.nz/assets/Uploads/Regional-gross-domestic-product/Regional-gross-domestic-product-Year-ended-March-2025/Download-data/regional-gross-domestic-product-year-ended-march-2025.xlsx) - [Geographic units by industry and statistical area, 2000–2025 (CSV)](https://www.stats.govt.nz/assets/Uploads/New-Zealand-business-demography-statistics/New-Zealand-business-demography-statistics-At-February-2025/Download-data/geographic-units-by-industry-and-statistical-area-2000-2025-descending-order.zip) - [Health and Safety at Work Act 2015](https://www.legislation.govt.nz/act/public/2015/70/en/latest/) - [Managing psychosocial risks at work](https://www.worksafe.govt.nz/topic-and-industry/work-related-health/mental-health/managing-psychosocial-risks-at-work/) - [Fair Trading Act 1986](https://www.legislation.govt.nz/act/public/1986/121/en/latest/) - [MYOB Acumatica AI ERP](https://www.myob.com/nz/erp-software/solutions/ai-erp) - [AI inventory management features](https://www.cin7.com/features/ai/) Published by Levy Street (https://levystreet.com), Wellington, New Zealand.