Artificial intelligence (AI) helps businesses work faster than ever before. From writing emails and creating reports to analyzing data and building software, AI tools have become part of everyday work. But let’s peek behind the convenience fact lurking in the shadows: token usage.
Tokens are the units AI systems use to process information. Every prompt, question, response, and action uses tokens. To simplify this, picture yourself as the AI model and your three-year-old who never stops asking questions as the user. Your tokens are equivalent to your patience at times. If you could only place limits on the rapid-fire questions of the three-year-old (haha!). In AI models, the more tokens used, the more a company pays. While the costs may seem small at first, they can grow quickly when usage is not monitored.
The good
Let's start with the positive side.
AI can save companies thousands of hours of work. Teams can automate repetitive tasks, improve customer service, speed up research, and create content in minutes instead of hours. When managed properly, AI can increase productivity while reducing operational costs.
Many organizations are already seeing real value from AI because they have clear usage policies, spending limits, and reporting systems in place. They understand who is using the tools, how often they are being used, and whether the results justify the cost.
When businesses treat AI like any other business expense, they can enjoy the benefits without surprises. Raise your hand if you like surprises. My point exactly!
The bad
Role play time.
Tim the teller: I am really enjoying this new AI tool. I am practically using it all day.
Tim’s boss: That’s awesome Tim. I can see from your numbers it helps your overall daily productivity.
…meanwhile in accounting land.
John the accountant: Where the heck did this AI invoice come from for $20K? Don’t we already have a paid subscription for this?
Problems begin when companies assume AI is a flat-cost service.
Many AI vendors charge based on token consumption. That means every interaction has a cost attached to it. A simple request may use only a small number of tokens, while more advanced tasks such as document analysis, coding assistance, or AI agents performing multiple actions can consume thousands or even millions of tokens.
Recent reports show that many organizations are struggling to track and manage AI spending. Some companies have seen their AI budgets double or triple because employees were using powerful models without any oversight. Others discovered that different departments were paying for overlapping AI tools without realizing it.
Without visibility, AI spending can quickly become unpredictable.
The ugly
The most extreme example surfaced in May of this year.
According to reports shared by AI consultants and covered by multiple news outlets, an enterprise customer using Anthropic's Claude AI accumulated a staggering $500 million AI bill in a single month. The reason was surprisingly simple: no spending caps, no usage limits, and no monitoring system were in place. Employees had unrestricted access to advanced AI models and high-volume workflows, causing token usage to spiral out of control.
While this example represents an extreme case, it highlights a growing issue across the business world. Organizations are discovering that AI costs can escalate much faster than traditional software subscriptions because usage-based pricing is directly tied to consumption.
The lesson is clear: if you are not monitoring token usage, you are not managing your AI costs.
Why permissions and limits matter
Every company should establish permission levels and spending controls.
Not every employee needs access to the most powerful or expensive AI models. Different roles require different capabilities. A customer service representative may only need a basic model, while a software engineer may require a more advanced one.
Organizations should consider:
- Setting monthly token budgets;
- Limiting access to premium AI models;
- Creating approval processes for large-scale AI projects;
- Monitoring usage by team, department, and individual users; and
- Receiving alerts when spending exceeds predefined thresholds.
These simple controls help prevent budget surprises and ensure AI spending aligns with business goals.
Questions you should ask every AI vendor
Before signing a contract or adopting an AI platform, every company should ask these four questions:
1. How is token usage measured and billed?
Ask for a clear explanation of how costs are calculated. Understand the difference between input tokens, output tokens, and any hidden processing charges.
2. What monitoring and reporting tools are included?
A good vendor should provide dashboards that show usage, costs, trends, and spending by user or department. If you cannot see your usage, you cannot control it.
3. Can we set spending limits and user permissions?
The platform should allow administrators to create budgets, restrict access to expensive models, and prevent unexpected cost overruns.
4. What happens when usage exceeds our budget?
Some vendors simply continue processing requests and send a larger invoice later (i.e., Surprise). Others allow spending caps, alerts, or automatic shutdowns. Knowing the difference could save your organization thousands of dollars.
Final thoughts
AI is not just a tech area decision anymore; it is a financial decision. The companies that succeed with AI will not be the ones that use it the most. They will be the ones that use it strategically and responsibly.
The good news is that AI can deliver incredible value. The bad news is that costs can rise quickly without visibility. The ugly reality is that a lack of monitoring can lead to massive and unexpected bills.
Before adopting any AI solution, make sure your organization understands how tokens work, who can use them, and how spending will be controlled. A few simple questions today can prevent a very expensive lesson tomorrow.
Note: I am not saying we are a bunch of three-year-olds. Ha!
