As working with AI agents becomes the norm, a new task and a new job title is quietly appearing inside businesses: managing token expenditure.
Who says AI doesn’t create jobs!
But what actually is a “token”, and how much does it cost? If you are running AI agents in your business, or thinking about it, this is worth ten minutes of your time.
What is a token?
Every AI model, whether it’s OpenAI’s ChatGPT, Anthropic’s Claude, or any other platform, breaks language down into small chunks called tokens. Roughly speaking, a token is a few characters of text. Every request you send, and every response you get back, is measured and billed in tokens.
How many tokens a task uses depends on several signals:
- The size of the prompt you send
- Previous conversation and project context sent again with each request
- Any files and tool results included
- The length of the answer
- Hidden “reasoning” tokens the model uses to think before answering
- Agent retries, loops and subagents working in the background
For example, one model might use 10,000 tokens to perform a task, while another model might use 20,000 tokens for the same job (these exact figures are illustrative, not real numbers). The results from each model are often similar, but how they got there, and what it cost to get there, can be very different.
The token counts themselves are managed by the platforms, and I wouldn’t worry too much about the internal mechanics. What does need your attention is how much those tokens cost to buy.
Tokens cost real money, and it adds up faster than you think
As of this morning, a leading model costs around $5 per million tokens sent to the platform, and $30 per million tokens sent back (these figures move regularly and vary by model, but they are a realistic ballpark).
A million tokens sounds like a lot. “I’ll never use that much,” you might think.
Oh yes, you will. Maybe not this week. But it is amazing how quickly a long conversation, a document review, or an agent working through a multi-step task can chew through tokens.
Keeping an eye on usage is going to matter a lot more soon, especially if:
- flat-rate subscriptions get scaled back or disappear, which I expect to happen at some point in the near future, or
- you are already paying directly through an API, where every token is billed in real time.
How do you actually monitor this?
Every AI platform has a usage page that shows you overall spend: how many tokens and how many pounds or dollars you have used against an API key.
The problem is that this is just a top-level number. It doesn’t tell you what you spent it on, which project it belongs to, which task it supported, or which client should be paying for it.
That’s where your CRM needs to step in. It needs to help you track:
- how many tokens you’ve used
- how much that has cost, per project or per task
- the running cost of a long conversation with your AI agent
Without that visibility, AI spend becomes a black box, and black boxes are how businesses end up with unexpected invoices.
Why this matters for managers and business owners
This is the new normal. As AI agents take on more real work inside a business, the cost of running them becomes a line item that managers need to understand, forecast and control, in the same way they would manage staff time or software licences.
Left unmanaged, token spend can spiral quietly in the background until a large invoice lands on your desk.
How WI1 AI CRM handles this
This is exactly the problem WI1 AI CRM (WI1.com) was built to solve. AI is baked into the platform from the ground up, and every bit of usage and cost is presented back to you clearly.
That means:
- no more ugly, unexpected AI invoices
- a clear view of what AI spend belongs to which project or task
- the option to pass AI expenditure on to a client, with a breakdown of exactly what those tokens were used for
If you’re using AI agents in your business, or planning to, the question isn’t just “what can AI do for us?” It’s also “who is watching what it costs?”
If you’d like to see how WI1 AI CRM keeps AI costs visible and under control, get in touch and I’ll walk you through it.
