AI for New Zealand logistics.

Where AI earns its keep in New Zealand logistics

AI earns its keep in logistics on the documents and messages that surround each consignment: proof of delivery dockets, tracking exceptions, damage reports, freight invoices and the stream of customers asking where their goods are. Every operator already has this data. The work is reading it, matching it and explaining it, which is exactly what language models do well.

Most organisations are not yet using AI this way. Datacom’s 2026 State of AI Index, a survey of 207 senior leaders at New Zealand organisations with 100 or more staff, run by Curia between 28 July and 3 August 2026, found general-purpose assistants such as ChatGPT, Gemini and Claude were the most common AI technology at 68%, while only 13% used agentic or autonomous AI (Datacom, August 2026). For freight, that is no bad thing. Most gains come from well-bounded automations with approvals, not agents acting on their own; our lesson on when a workflow does not need an agent explains the difference.

Freight and warehouse workflows

The detailed guides cover proof of delivery extraction, exception summaries and delivery enquiry handling.

Carrier liability, driver hours and other rules AI must not misstate

Freight law is specific, and a model answering from general knowledge will often reach for overseas rules. Quote New Zealand sources and get legal advice on any contested claim.

• Carrier liability. Under s 259 of the Contract and Commercial Law Act 2017, a carrier’s liability for contracts at limited carrier’s risk is capped at $2,000 for each unit of goods lost or damaged, or the declared value for contracts at declared value risk. An AI reply that implies full replacement value creates expectations your terms do not support.
• Claim deadlines. Under s 274, no proceeding may be brought against a contracting carrier for damage or partial loss unless written notice with reasonable particulars is given within 30 days after the carrier’s responsibility ends. Section 275 sets a 10-day notice period for a contracting carrier’s claim against an actual carrier. Claim assembly tools should calculate and display these dates from the delivery record.
• Driver work time. Under s 30ZC of the Land Transport Act 1998, a driver may not exceed 13 hours of work time in a cumulative work day, must have at least 10 hours’ continuous rest, and may not exceed 70 hours in a cumulative work period. Section 30ZD requires time and payment records to be kept for 12 months. AI can summarise these records; it should not approve extra hours.
• Safety. The Health and Safety at Work Act 2015 primary duty (s 36) covers fatigue, forklift traffic and loading. Tools that change how drivers or warehouse staff report hazards need worker engagement (s 60).
• Privacy. Consignee details, delivery photos and telematics are personal information. If data goes to an offshore AI provider that uses it for its own purposes, the Privacy Commissioner’s guidance on disclosure outside New Zealand applies.

Tracking, WMS and accounting systems

New Zealand operators and shippers commonly use Starshipit or GoSweetSpot for shipping labels and tracking, CartonCloud or similar warehouse systems, Cin7 for inventory, Shopify for online orders, carrier tracking APIs from NZ Post and the major courier networks, and Xero for invoicing. AI work in logistics is mostly integration: reading events from these systems, writing structured results back, and keeping the customer conversation consistent with the tracking record. Before any build, decide which system is the source of truth for each status, because a carrier’s API, your warehouse system and the online store can disagree for hours after a scan, and an assistant that reads the wrong one will confidently tell a customer the wrong thing. Our Shopify and Cin7 guides cover the API side for shippers.

A first project: proof of delivery extraction

Inputs

Scanned and photographed POD dockets, driver app images, consignment records.

Systems

Document store, transport management or tracking system, exception queue.

Steps

Classify the image, extract consignment number, date, time, receiver name, signature present and any notes, then match to the consignment.

Human review

Operators clear low-confidence extractions, missing signatures and any damage notes.

Outputs

POD fields written to the consignment, damage notes raised as exceptions, a searchable POD index.

Measures

Field accuracy on a labelled set of 300 dockets, share auto-matched, time to answer POD queries, claims lacking a POD.

Illustrative example: a regional carrier handles 8,000 consignments a month and receives 400 POD queries from customers (assumptions). At 7 minutes each to find and send the docket, that is about 47 hours. If extracted, indexed PODs cut that to 2 minutes, the work falls to about 13 hours, saving roughly 34 hours a month. Our document processing service covers this pattern, and the engineering note on OCR versus vision models explains the extraction choices.

Where AI fails in freight

If customers will talk to the system directly, our AI customer support service sets up the hand-off to dispatch and the logs you need when a delivery is disputed.

Workflows we map in this sector

• Shipment exception summariesTurn carrier events, broker emails and driver notes into one ranked list of exceptions with an owner and a deadline.
• Proof of delivery extractionRead signed dockets, driver photos and ePOD exports into structured records, and catch damage notes before invoicing.
• Customer delivery enquiry handlingAnswer where-is-my-order questions from live tracking, check who is asking, and hand lost or damaged goods to a person.

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