If you've never written a line of code, 'tokens' can sound like a deliberately confusing way to bill you. It isn't complicated once you see it plainly: a token is roughly three-quarters of a word, and every AI response you read, and every document you feed in, gets counted and charged in these small chunks. Understanding this is the difference between an AI bill that makes sense and one that feels like it appeared from nowhere.
Where the money actually goes
Two things drive your AI bill: how much text goes IN (your prompt, plus any documents or context you attach) and how much comes OUT (the model's response). Long documents pasted in for summarising cost more than a short question, even if the response is brief. A 10-page contract you ask Claude to review might cost the equivalent of a few cents in tokens -- small on its own, but it adds up fast if that's happening 200 times a month across a team.
This is why the same task can cost wildly different amounts depending on how it's set up. Asking a model to re-read an entire policy document on every single question, instead of only the relevant section, can be five to ten times more expensive for the identical outcome. Most cost blowouts we see in Australian small businesses trace back to this, not to the model itself being expensive.
A plain-English budget sense-check
A short customer email draft: a fraction of a cent per response.
Summarising a 20-page report: a few cents per document.
A chatbot answering from a large knowledge base, done well: cents per conversation, not dollars.
The same chatbot, done badly (re-sending the whole knowledge base every time): can be ten times higher for no better answers.
A typical Australian small business running everyday AI use -- drafting, summarising, some customer replies -- lands somewhere between $50 and $400 a month once it's set up sensibly. If your bill is materially above that without a clear reason (high call volume, large documents, heavy automation), it's worth a technical person checking whether context is being sent efficiently before assuming the tool itself is just expensive.
The one question worth asking your provider
If you take one thing from this: ask whoever set up your AI tools 'is this sending the full document every time, or only the relevant part?' It's a fair, plain-English question and any competent setup should have an equally plain answer. If nobody can answer it, that's the first place to look before assuming your bill reflects genuine usage rather than an inefficient setup quietly re-sending the same context over and over.
It's also worth checking whether you're on a per-seat subscription or a usage-based API bill, because the economics flip depending on how heavily your team actually uses the tool. A five-person team using AI constantly might do better on usage-based pricing; a fifteen-person team using it lightly might do better on a flat per-seat plan. Neither is automatically cheaper, and an Australian business paying $3,000 a year for a subscription five staff barely touch is leaving money on the table just as surely as one with an unexplained usage spike.
A sanity check you can run yourself
Once a quarter, pull up your AI billing dashboard and look at the trend line, not just the latest total. A steady, gradual increase as your team uses the tool more is healthy and expected. A sudden jump in a single month, with no matching jump in how much work the tool is doing, is worth a five-minute conversation with whoever manages the setup before it becomes a habit baked into next year's budget.
The goal isn't to become a token-pricing expert. It's to have enough of a mental model that a bill either makes sense or clearly doesn't, and you know which question to ask when it's the latter. That's the whole of what this article is trying to hand you.
For what it's worth, most Australian small businesses we've reviewed have never once looked at this breakdown before asking us to. It usually takes fifteen minutes, and it usually changes at least one decision about how the tool gets used day to day.
You don't need to understand the maths to manage the budget. You need one number each month, a rough sense of what's driving it, and someone who'll flag it if it moves without explanation. If your current AI spend doesn't make sense to you and nobody's been able to explain it plainly, that's worth fixing before it's worth cutting. Get in touch through /contact and we'll translate your bill into plain English.
