Every software vendor has added "AI" to their product page. Every conference has an AI keynote. Every business owner has been told they need to "adopt AI or fall behind."
Most of it is noise.
This post is a practical breakdown — based on what I've built and what I've seen working for New Zealand businesses — of which AI applications are genuinely delivering results and which ones are still more demo than deliverable.
The Honest Starting Point
Most AI tools marketed to small and medium businesses in New Zealand right now are one of three things:
- Generic SaaS tools with an AI feature bolted on. Your CRM now has an "AI assistant" that summarises notes. Useful, but not transformative.
- Consumer AI tools used informally. Staff using ChatGPT to draft emails, summarise documents, or generate content. High value per dollar, but informal and unintegrated.
- Custom AI systems built for the specific business. The highest ROI, the most effort, and the only category that produces defensible competitive advantage.
Most of the hype is about category 3. Most of the budget gets spent on category 1. Most of the actual current value sits in category 2. Understanding this gap is the starting point for any sensible AI strategy.
What's Genuinely Working Right Now
1. Document Processing and Data Extraction
The problem it solves
New Zealand businesses in agriculture, construction, logistics, and professional services process enormous volumes of documents — consignment notes, compliance certificates, invoices, contracts, inspection reports. Extracting structured data from these has historically required staff time or expensive OCR systems.
How AI changes it
Large language models are remarkably good at extracting specific fields from unstructured documents — even poorly scanned ones, even handwritten ones, even ones with inconsistent formats. A system built on AWS Bedrock can ingest a PDF, extract the relevant fields, validate them against business rules, and push the data into your ERP or spreadsheet automatically.
Real NZ Use Case
A Bay of Plenty logistics company processing 200+ consignment notes per day. Each note required 3–4 minutes of manual data entry. An AI extraction pipeline reduced this to near-zero human input for 85% of documents, with exceptions flagged for review. The system paid for itself in under three months.
What to look for
High-volume, repetitive document types with consistent enough structure that a human could extract the data in 30 seconds. If a human can do it quickly and repeatedly, an AI system can likely do it faster and at volume.
2. Customer Query Automation
The problem it solves
Staff time spent answering the same questions repeatedly — product availability, pricing, order status, policy questions, technical specifications. In industries with complex product catalogues or technical requirements, this is a significant drain.
How AI changes it
A well-built AI chatbot trained on your product catalogue, policies, and FAQ content can handle 60–80% of inbound queries without human involvement. Unlike the rule-based chatbots of five years ago, modern LLM-based systems handle natural language, understand context, and manage follow-up questions correctly.
Real NZ Use Case
This is exactly what I built for Ballance Agri-Nutrients — a production AI system on AWS Bedrock handling complex, domain-specific queries from farmers and agronomists. The queries involve technical product knowledge, application rates, regional considerations, and regulatory requirements. A rules-based system couldn't have handled the variation; an LLM-based system could.
What to look for
Businesses receiving high volumes of inbound queries where the answers are consistent and documentable. The key constraint is data quality — the AI system is only as good as the knowledge base it's trained on.
3. Internal Knowledge Search
The problem it solves
Large organisations accumulate years of documentation, policies, procedures, and institutional knowledge that's practically inaccessible. New staff take months to become productive because the knowledge isn't surfaced efficiently.
How AI changes it
Retrieval-Augmented Generation (RAG) systems index your internal documentation and allow staff to ask questions in plain language and get accurate, cited answers. "What's our returns policy for commercial customers in Auckland?" gets a direct answer from the actual policy document, with a link to the source.
What to look for
Organisations with more than ~50 internal documents that staff need to reference regularly, especially where inconsistent application of policy or procedure has been a problem.
4. Reporting and Analysis Automation
The problem it solves
Many NZ businesses produce the same reports weekly or monthly — pulling data from multiple sources, formatting it consistently, distributing it to the right people. This is often done manually by someone who could be doing more valuable work.
How AI changes it
AI can generate narrative summaries of data, identify anomalies, flag metrics outside expected ranges, and format reports for different audiences — all automatically, on schedule, without manual intervention.
What to look for
Any business running regular reports that require consistent data aggregation across systems — especially where the report includes a written summary alongside the numbers.
What Isn't Worth It Yet (For Most NZ SMEs)
AI-generated marketing content at volume
It's possible. It works. But the content it generates is generic and indistinguishable from every other business using the same prompts. Content that converts for a NZ audience — particularly in agricultural and regional markets — requires specific local knowledge and voice that generic AI content doesn't have.
Autonomous AI agents for complex decisions
The vision of an AI agent that autonomously handles complex decisions — approving credit, managing procurement, triaging complex complaints — is real but not yet reliable enough for most NZ businesses. AI works best when it's assisting human decisions, not replacing them entirely.
AI for compliance and legal work
Tempting, but dangerous. NZ-specific regulatory requirements, case law, and compliance obligations are underrepresented in the training data of most AI models. AI can help draft and summarise, but anything with legal or compliance consequences needs human review.
The Right Way to Evaluate an AI Investment
Before committing budget to any AI system, run through this:
What specific problem does it solve?
If the answer is vague ("it'll make us more efficient"), keep asking until you get to a specific workflow with a specific time or error cost attached.
What does the manual process currently cost?
Estimate it in staff hours per week, multiplied by the hourly rate. An AI system that saves 10 hours per week at $50/hour is worth $26,000/year — that's your maximum justifiable investment on a one-year payback.
What data does it need, and do you have it?
AI systems are only as good as the data they run on. Audit your data before committing.
Who owns it after it's built?
A custom AI system is an asset. Make sure you understand how it will be maintained, updated as your data changes, and extended as your needs evolve.
The NZ-Specific Advantage
New Zealand businesses that invest in custom AI now have a real window of advantage. Most NZ SME competitors aren't doing this yet. The businesses that build AI into their operations now — in document processing, customer queries, internal knowledge — will compound that advantage over time as the systems learn more about their specific business context.
The window won't stay open indefinitely. But right now, for the right applications, the return on a well-scoped AI investment for a NZ business is as good as any technology investment I've seen in the last decade.