Plain-English definitions of the AI, automation and software terms New Zealand business owners run into — written by the Avtrix team so you can read a proposal, a vendor pitch or a news story and know exactly what is being promised. Each definition is short enough to quote and includes what the term means in practice for a Kiwi SME.
Jump to a Section
AI fundamentals
Artificial intelligence (AI)
Software that performs tasks which normally need human judgement — understanding language, recognising patterns, making decisions. In business today “AI” almost always means systems built on large language models that can read, write, classify and decide. For an NZ SME the practical question is not “should we use AI” but “which repetitive process should it take over first”.
Large language model (LLM)
The type of AI model behind ChatGPT, Claude and Gemini: trained on enormous amounts of text so it can understand and generate language. LLMs are the reasoning engine inside most modern AI agents and assistants. They are excellent at language tasks and unreliable at arithmetic and facts unless connected to your real data.
Generative AI
AI that creates new content — text, images, code, audio — rather than only analysing existing data. Most NZ businesses first meet it through ChatGPT for drafting emails and marketing copy. The bigger value comes when generative AI is embedded in a process, such as drafting quotes from a site visit note or replies to customer enquiries.
Machine learning (ML)
A branch of AI where software learns patterns from data instead of following hand-written rules — for example predicting which invoices will be paid late from past payment history. LLMs are one product of machine learning; classic ML is still the right tool for forecasting, scoring and anomaly detection on structured business data.
Natural language processing (NLP)
The ability of software to read and understand human language — emails, chat messages, phone calls, documents. NLP is what lets an AI agent work out that “can someone come Thursday arvo?” is a booking request for Thursday afternoon.
Prompt
The instruction given to an AI model. In a business system the prompt is engineered once — defining the agent’s role, rules, tone and limits — and reused thousands of times. Good prompt design is a large part of why one AI agent answers accurately and another makes things up.
Hallucination
When an AI model states something false with confidence — an invented price, policy or fact. It happens because LLMs generate plausible language rather than look things up. The fix is grounding the model in your real data (see RAG) and setting it to say “I don’t know” and hand off rather than guess.
Token
The unit AI models read and are billed in — roughly three-quarters of a word. Platform costs for an SME agent are usually tens to a few hundred dollars a month, driven by how many tokens are processed. Long documents and long conversations cost more.
AI agents and automation
AI agent
Software that can understand a request, decide what to do, and take action across your systems — look up an order, book a job, update the CRM, send an invoice — without a person handling each step. The difference from a chatbot is action: a chatbot explains your returns policy; an agent processes the return. See AI agent development.
Agentic AI
AI systems built to act autonomously toward a goal over multiple steps — planning, using tools, checking results and trying again. Datacom’s 2026 State of AI Index found only 13% of New Zealand organisations use agentic systems, making it the least-adopted and highest-return layer of AI in NZ business.
Chatbot
A conversational interface that answers questions, on a website, in Messenger or WhatsApp. Older chatbots followed scripted decision trees; modern ones use LLMs and can hold a natural conversation. On its own a chatbot informs; connected to your systems it becomes an AI agent that can act.
AI voice receptionist
An AI agent that answers phone calls in natural speech, understands the caller, answers questions, books or reschedules appointments and routes urgent calls to a person — 24 hours a day. Common in clinics, trades businesses and property firms that miss calls after hours. See AI voice receptionist.
Business process automation (BPA)
Using software to run a whole business process — invoice approval, customer onboarding, job scheduling — from trigger to completion without manual steps. BPA usually combines integrations between your systems, rules, and increasingly AI for the steps that need judgement. See business automation.
Robotic process automation (RPA)
Software “bots” that mimic a person clicking through applications to move data between systems that have no integration. RPA is brittle — it breaks when a screen changes — and is being replaced by API integrations and AI agents wherever those are possible.
Workflow automation
Automating a defined sequence of tasks: when a form is submitted, create the job, notify the team, send the confirmation. Tools like Make, Zapier and Power Automate cover simple cases; custom workflow engines handle the complex, high-volume or compliance-sensitive ones.
Intelligent document processing (IDP)
AI that reads documents — invoices, forms, contracts, delivery dockets — extracts the relevant fields, validates them and posts them into your systems. It replaces manual data entry and OCR-only tools that could read text but not understand it. See AI document processing.
Optical character recognition (OCR)
Technology that converts images of text — a scanned invoice, a photographed docket — into machine-readable text. OCR reads; it does not understand. Modern document agents combine OCR with an LLM so the system knows which number is the total and which is the GST.
Human-in-the-loop
A design where an AI system does the work but a person approves specific steps — a refund over $500, a medical question, a contract term. It is how well-built agents stay safe: the AI handles volume, humans handle judgement on the cases that matter.
Guardrails
The rules and technical limits placed on an AI agent: what it may say, which systems it may touch, which actions need approval, and when it must hand over to a person. Guardrails are what make the difference between an agent you can trust with customers and a demo.
Hand-off (escalation)
The moment an AI agent passes a conversation or task to a human, with the full context attached. Good hand-off design — confidence thresholds, clear triggers, no repeating yourself — is the single biggest driver of customer satisfaction with AI support.
Data and knowledge
Retrieval-augmented generation (RAG)
A technique where the AI first retrieves relevant passages from your own documents and data, then generates an answer grounded in them — so it answers from your price list, policies and past jobs rather than general knowledge. RAG is the standard way to make an AI agent accurate about your business without retraining a model. See our article on RAG vs fine-tuning.
Fine-tuning
Further training an AI model on your own examples so it adopts a specific style, format or specialised behaviour. Fine-tuning is more expensive and less flexible than RAG and is rarely the right first step for an SME; most business needs are met with a well-designed prompt plus RAG.
Vector database
A database that stores text as numerical “embeddings” so it can find passages by meaning rather than exact keywords — the search layer behind RAG. When a customer asks about “delivery to rural addresses”, a vector search finds your “regional freight policy” even though the words differ.
Embedding
A numerical representation of a piece of text that captures its meaning. Embeddings let AI systems compare, search and cluster documents by what they mean. They are generated once when your content is indexed and reused for every query.
Knowledge base
The organised set of documents, policies, product data and past cases an AI agent draws on. Building and maintaining the knowledge base — deciding what goes in, keeping it current — is usually more work than the AI itself, and it is where most of an agent’s accuracy comes from.
Business intelligence (BI)
Turning the data your business already collects — sales, jobs, stock, cash — into dashboards and reports people actually use to make decisions. In NZ SMEs the common tools are Power BI and Looker Studio connected to Xero, your CRM and your operational systems. See data & business intelligence.
Dashboard
A live, visual view of the numbers that matter — today’s jobs, overdue invoices, this month’s margin — updated automatically from your systems. A good dashboard replaces the Monday-morning spreadsheet and the questions it took to build it.
Data pipeline
The automated flow that moves data from where it is created (Xero, your job system, a form) to where it is used (a dashboard, a report, an AI agent’s knowledge base), cleaning and combining it on the way. Reliable pipelines are what make “single source of truth” real.
Software and integration
Custom software
An application built specifically for one business, around its exact processes, instead of an off-the-shelf product everyone uses. Right when the process is core to how you compete or no product fits without painful workarounds. See custom software development and our guide to off-the-shelf vs custom.
API (application programming interface)
The published way one piece of software lets another read or change its data — how an AI agent creates an invoice in Xero or a booking in your calendar. If a system has a good API, integrating it is straightforward; if it doesn’t, integration is slower and more fragile.
Integration
Connecting two or more systems so data flows between them automatically — orders into accounting, leads into the CRM, jobs into the calendar. Most manual double-entry in NZ businesses is a missing integration. See enterprise integration.
CRM (customer relationship management)
The system that holds your customers, leads, conversations and deals — HubSpot, Salesforce, Pipedrive, or a custom build. AI sales agents typically read from and write to the CRM, so its data quality sets a ceiling on what the agent can do. See CRM & ERP solutions.
ERP (enterprise resource planning)
Integrated software that runs the core of a business — finance, inventory, purchasing, production, sometimes HR — in one system. Mid-sized NZ firms often run an ERP plus Xero or MYOB; AI document and workflow agents plug into it to remove the manual steps around it.
SaaS (software as a service)
Software you subscribe to and use in a browser rather than install and own — Xero, HubSpot, Microsoft 365. Most business tooling is SaaS; custom software is what you build when SaaS products don’t fit, and it usually integrates with several of them.
Cloud computing
Running software and storing data on providers’ servers (AWS, Microsoft Azure, Google Cloud) instead of your own hardware. For NZ businesses the practical questions are which region the data lives in (Auckland, Sydney), who can access it, and what it costs each month. See cloud solutions.
Data residency
Where your data is physically stored and processed. New Zealand’s Privacy Act 2020 does not ban offshore storage but requires comparable protection; many NZ businesses and public-sector buyers prefer NZ or Australian regions, and the major cloud and AI providers now offer them.
Digital transformation
Rebuilding how a business operates around modern software, data and increasingly AI — not one project but a sequence: integrate the systems, automate the processes, then add intelligence. Done well it is a series of measurable steps, each paying for the next. See digital transformation.
Managed services
An ongoing arrangement where a provider runs, monitors, maintains and improves your systems — including AI agents, which drift and need tuning as your business and the models change — for a fixed monthly fee. See managed services and managed AI services.
Compliance and governance
Privacy Act 2020
New Zealand’s privacy law, in force since December 2020. It sets 13 information privacy principles covering collection, use, storage, disclosure and access to personal information, requires notification of serious privacy breaches to the Privacy Commissioner, and restricts sending personal information offshore unless comparable protections apply. Any AI system that handles customer or staff data in NZ must be designed around it.
AI governance
The policies, roles and controls that decide how an organisation uses AI: what data can be sent to which models, who approves new uses, how outputs are checked, and how incidents are handled. Datacom’s 2026 index found 62% of NZ organisations have some AI strategy but only 16% a clearly defined one — governance is usually the gap.
Responsible AI
Building and using AI in ways that are fair, transparent, safe and accountable — telling customers when they are talking to an agent, avoiding biased decisions, keeping a human responsible for outcomes. In practice it is a checklist applied at design time, not a philosophy.
Audit trail
A complete, tamper-evident record of what a system — or an AI agent — did, when, with what data, and on whose authority. Audit trails are what let you answer “why did the agent approve that?” and are expected by auditors, regulators and increasingly customers.
Measuring value
Return on investment (ROI)
The financial return from a project relative to its cost. For AI and automation the honest way to measure it is the process metric agreed before the build — hours saved, response time, resolution rate, error rate — converted to dollars. Datacom’s 2026 index found 33% of NZ organisations already see AI returns exceeding costs.
Proof of concept (POC) / pilot
A small, time-boxed build to prove an AI or automation idea works on your real data before committing to the full project. A good pilot has a clear success metric and a fixed price; a bad one is an open-ended experiment. Most Avtrix engagements start with a scoped first project rather than a POC, because a fixed-price first project delivers value as well as proof.
Fixed-price delivery
A commercial model where the full scope, price and timeline are agreed before work begins, and the provider carries the risk of overruns. It is the opposite of time-and-materials (hourly billing). Every Avtrix project is delivered fixed price — see how we work.
Last updated 15 September 2026. Maintained by the Avtrix Software Solutions team, Taupō, New Zealand. Missing a term? Tell us and we’ll add it.
Not Sure Which of These You Need?
Book a free 30-minute discovery call and we’ll tell you in plain English where AI and automation will pay back first in your business.