Most business owners understand a software invoice. You pay per user. You pay per month.

You know roughly what is going out and what you get for that. AI can be a llittle different.

You may still pay a monthly amount for ChatGPT, Copilot, Claude, Gemini or another AI tool. But behind that monthly fee, there can be another meter running, one that's slightly opaque.

That meter might be called tokens, credits, usage, requests, searches, file analysis or something else. The name matters less than the point.

Your team may be using more AI capacity than you realise.

That does not mean AI is bad value. Far from it. Used well, AI can save hours, improve the quality of work and help people think more clearly. But like any business tool, it needs to be understood and managed.

If Uber can get caught out, so can anyone

A useful example came from Uber.

Reports in June 2026 said Uber had burn through its annual AI budget in just four months, so they brought in caps on staff use of AI coding tools.

Uber is a highly technical business. It has smart people, finance teams, engineers and procurement teams. Yet even there, AI use appears to have grown faster than the budget expected.

That is the lesson for all businesses.

If a company like Uber can get caught out by AI usage, your SME could too.

The issue is not that people are using AI. That is a good thing, because AI can bring real results. The issue is people using AI without knowing what is being measured, what is included in the subscription and what costs extra.

What is the hidden meter?

You do not need to become a technical expert. But you do need to understand the basic idea.

AI tools process information called tokens. If you ask AI to read a long document, write a long answer, search through files, compare information or keep rewriting the same piece of work, more chunks are being processed, and the more tokens you use. For context, a typical word uses about one and a half tokens.

So in plain English, the more work you ask AI to do, the more usage you will burn through.

This matters because on some plans, even though they may seem unlimited, they are not. You may know what your AI subscription costs.

But do you know what your AI usage costs?

This is already reaching normal business tools

Microsoft’s recent guidance for Copilot Cowork Credits says usage-based billing can sit alongside fixed subscriptions. Microsoft gives you reports, budgets, alerts and you can set hard caps to track consumption and prevent overspending.

That tells us something.

If Microsoft is giving businesses tools to set limits and track spend, business owners should be asking how those costs work. This, however, can sometimes be tricky to do. 

Why poor AI use wastes money

This is where your team's training makes a real difference. People who have not been trained often use AI in a messy way. They paste in too much or the wrong type of information. They ask vague questions. They keep asking for rewrites because the first answer was not right. They upload files the AI does not need.

They use the most powerful tool for a simple task, you wouldn't use a reasoning model just to rewrite an email or check the grammatical status of something. They treat AI like a search box, then wonder why the result is weak. That burns time. It can also burn usage and tokens.

Good training helps people ask better questions, give clearer instructions and use the right tool for the job. It also helps them know when not to use AI.

Responsible AI use is not about using AI for everything. It is about using it where it helps.

A trained team will usually get better answers with fewer attempts. That means less waste, better results and a clearer return on the money being spent.

There is an environmental cost too

The hidden meter is not just financial. AI runs through data centres. Those data centres need electricity. That does not mean we should stop using AI.

It does mean we should use it responsibly.

If a team is constantly asking AI to redo poor work, read huge documents it does not need, or produce long answers no one reads, that is wasteful in more than one way.

Good AI training helps with this too.

Clearer prompts.

Shorter instructions.

Better judgment.

Fewer repeated attempts.

Less unnecessary processing.

That is better for the budget and better for responsible business practice.

Before you add another AI tool, ask yourself these questions

1. Is this a fixed monthly cost, a usage cost, or both?
2. Can we set budgets and limits?
3. Who should have access to the more costly features?
4. What business result are we measuring?
5. Ask what is included and what costs extra.

Then you need to find out whether the tool lets you set caps, alerts or usage limits before the team starts using it widely.

Not everyone needs every feature. Some tools should be limited to the people who have been trained to use them well.

Do not measure success by how much AI your team uses.

Measure what it improves.

Are proposals being written faster?

Are sales follow-ups better?

Are reports taking less time?

Are customer replies clearer?

Are managers making better decisions?

Are staff getting hours back each week?

That is what matters.

The answer is not to use less AI

This is not an argument for slowing down.

It is an argument for getting clearer.

AI is one of the most useful tools a business can add. But it needs the same common sense you would apply to any other business cost.

Know what you are buying.

Know how it is charged.

Know who is using it.

Know what good use looks like.

Train people properly.

Review whether it is saving time, making money or improving quality.

The businesses that get the best from AI will not be the ones that press the most buttons.

They will be the ones who know where the meter is, teach their people how to read it and connect AI use to real business results. At ConkerAI, Jo and I help business leaders not get caught out with this issue.