AI glossary for business, in plain English

Most AI jargon does one of two jobs. It describes something useful, or it makes something simple sound expensive. This is the short list of terms you'll meet when you look at automating part of your business. We've defined them the way we'd explain them on a call, not the way a vendor's brochure would.

The AI basics

Artificial intelligence (AI)

A broad label for software that does things we used to think needed a person: reading text, answering questions, spotting patterns, making a judgement call. In a business it almost never means a robot or a "brain". It means a tool that handles a fuzzy task a rule-based script couldn't.

Large language model (LLM)

The engine behind tools like ChatGPT. An LLM is trained on an enormous amount of text and is very good at predicting which words come next. That turns out to be enough to draft, summarise, classify and answer questions. Most "AI" business tools are built on top of one.

Prompt

The instruction you give an LLM. What you get back depends heavily on how you ask, which is why a good prompt library is worth keeping. A small set of tested prompts beats rewording the same request from scratch every time.

Hallucination

When an LLM states something false with total confidence. It isn't lying. It's predicting plausible text, and sometimes the plausible answer is wrong. That's the biggest reason a person reviews anything that matters before it goes out, and why a well-built system keeps AI on tasks where a mistake is cheap or gets caught.

How things get built

AI agent

A system that works towards a goal instead of answering once. It takes a task, decides the next step, uses tools, and keeps going until the job's done. A support agent that reads an email, looks up the order, drafts a reply and files it is doing far more than a chatbot. It's the idea behind our AI agents work.

Automation / workflow

A set sequence of steps that runs without anyone doing it by hand: "when a form comes in, create the record, send the confirmation, tell the team." Not every automation needs AI. Plenty of the most useful ones are just reliable plumbing between the tools you already have.

RAG (retrieval-augmented generation)

A way of making an LLM answer from your documents instead of its general training. It finds the relevant text first, then writes the answer from that. It's how you get an assistant that knows your policies, products or past matters rather than the whole internet.

Fine-tuning

Further training a model on your own examples so it behaves a particular way. It's powerful and often overkill. For most businesses a good prompt plus RAG gets there for a fraction of the cost and effort, and we'll tell you when fine-tuning is actually warranted.

Connecting your tools

API

The doorway one piece of software offers so other software can talk to it: read data, send data, trigger an action. When your CRM, store and accounting tool "integrate", they're using each other's APIs. It's the plumbing that lets a change in one place show up everywhere it should.

Integration

Connecting two or more systems so information moves between them automatically. Done properly, it ends double entry. Nobody has to copy an order from the store into the spreadsheet and then into the CRM. It's the core of our API integration work.

Webhook

A way for one system to tell another the moment something happens, so the second system doesn't have to keep checking. "A payment went through, do the next thing, now." It's what makes an automation feel immediate instead of running on a timer.

No-code / low-code

Tools that let you build automations by wiring boxes together instead of writing much code, such as Make, n8n and Zapier. They're good for getting started and for simple flows. At scale they can get expensive or brittle, and that trade-off is worth understanding before you put a critical process on one.

Cost and ownership

Per-seat pricing

Software priced per user per month. Fine when the tool fits. Painful when you're paying for logins people barely use, which is one of the main reasons a growing team eventually looks at building something it owns outright instead of renting forever.

Token

The unit LLMs are billed in, roughly a few characters of text. It matters because it's how the running cost of an AI feature gets measured. A well-scoped task costs cents, and knowing that keeps the "AI is expensive" myth in check.

Vendor lock-in

When leaving a tool is painful because your data, workflows or history are stuck inside it. The fix is owning the important parts (your accounts, your database, your code), so switching is a decision rather than a hostage negotiation.

Why we wrote this. Half the confusion in buying AI is vocabulary, and a vendor who fits ten of these terms into one sentence is usually describing something that could be built in a fortnight. Knowing what each word means is most of the defence. Ask what the thing does, not what it's called.