A ChatGPT Team plan costs $25 per user per month. That number is on the invoice. But how many hours does your team spend writing prompts, adjusting AI output, and fact-checking whether the numbers hold up? Most organizations know exactly what their AI license costs. Almost none know what AI actually costs them. Four calculations help you find the answer, and the result is often less comfortable than the sales deck suggested.
What are you actually spending on AI tools?
The first step is surprisingly difficult: a complete inventory of every AI tool your team uses. Not just the official licenses, but also the free accounts where employees are putting company data.
The known subscriptions are easy to add up. Here's what the most widely used tools cost per user per month, in USD:
| Tool | Plan | Price per user/month (USD) |
|---|---|---|
| ChatGPT Plus | Individual | $20 |
| ChatGPT Team | Team (min. 2) | $25 |
| Claude Pro | Individual | $20 |
| Claude Team | Team (min. 5) | $25 |
| Microsoft Copilot | Business | $30 |
| GitHub Copilot | Business | $19 |
| Cursor Pro | Individual | $20 |
For a team of ten running a mix of ChatGPT Team and GitHub Copilot, you're looking at $440 to $550 per month before anyone opens a single tool. Those are the visible costs.
The real problem sits below that line. TheAIDaily's shadow AI research found that 44% of SMB employees used a free AI tool with company data at least once in the previous month, averaging 3.1 times per week. Only 28% knew that many free tools can use submitted data for model training. Research from Awareways puts it in starker terms: just 11% of organizations have meaningful visibility into how employees actually use AI day to day.
Think of it this way: your team has a company car, but half the trips happen in personal vehicles without insurance. You never see the fuel receipts, but you carry the liability.
How much time is your team putting into AI?
This is the cost category almost nobody tracks. AI generates text, analysis, or code in seconds. But the time a team member spends writing the prompt, adjusting the output, and verifying the result? That never shows up anywhere.
Salesforce research finds that marketers save an average of 6.1 hours per week with AI tools. That's the headline. But 49% of those same marketers cite time savings as their primary reason for continuing to use AI, which means the productivity story has become self-reinforcing. The savings get measured. The investment in time rarely does.
For context, how much time does AI actually take?
A simple prompt takes thirty seconds. A complex task ("analyze this customer satisfaction dataset and write a report with recommendations") can easily consume fifteen to thirty minutes of iteration. The first version is never quite right. You refine the prompt, redirect the output, verify the facts. That's not a failure of the tool; that's how AI works. But it's working time you need to count.
A calculation that helps: ask three people on your team to track for one week how many minutes per day they spend on AI interaction. Not just typing prompts, but also reading, checking, and rewriting output. Most teams land between 45 and 90 minutes per person per day. At an average hourly rate of $50, that's $188 to $375 per person per month, just in time.
What happens when AI gets it wrong?
AI models regularly produce output that looks confident but isn't accurate. In the industry, this is called a hallucination. A fabricated statistic in a report. A legal citation that doesn't exist. A piece of code that compiles but does the wrong thing.
The cost here is hard to quantify but impossible to ignore. A proposal with a calculation error loses you a deal. A blog post with a wrong figure costs you credibility. A subtle bug in generated code costs your development team a full day of debugging.
This is why someone needs to check every AI output. That review time is the line item organizations consistently forget. If a team member saves an hour by having AI write a first draft, but then spends twenty minutes checking and correcting it, the net saving is forty minutes, not an hour.
There's a compounding risk here. The longer a team works with AI, the more confident they become. After three months, almost nobody checks as carefully as they did at the start. That's precisely when errors start making it through review.
When does AI pay for itself?
According to data from TheAIDaily's AI Workforce Statistics, AI tools generate an average productivity value of around $11,600 per knowledge worker per year. That's the ceiling: what's possible when the tool is well-configured and the person uses it every day. In practice, most teams capture 30 to 60% of that potential, depending on how well AI is embedded in actual workflows.
The break-even calculation is straightforward:
(Hours saved × hourly rate) − (subscription costs + prompt time + review time + error costs) = return
A concrete example. A marketing team of five:
- Subscription costs: 5 × ChatGPT Team ($25) + 5 × a writing tool ($15) = $200/month
- Prompt time: 5 people × 60 min/day × 22 workdays × $50/hour = $4,583/month
- Review time: avg 20 min of review per hour of AI output = $1,528/month
- Total costs: approximately $6,311/month
- Savings: if each person nets 1.5 hours saved per day = 5 × 1.5 × 22 × $50 = $8,250/month
- Return: $8,250 − $6,311 = $1,939/month net
Positive, but lower than most vendors would lead you to believe. And this assumes 1.5 hours of genuine daily time savings, something many teams don't reach.
Worth noting: BCG reports that companies are planning to spend an average of 1.7% of their revenue on AI. For a business with $2 million in annual revenue, that's $34,000 per year. Add the hidden time costs and you're often looking at twice that in real spend.
McKinsey research adds a sobering caveat: fewer than one in three decision-makers can connect the value of AI to their organization's financial performance. The tools are being deployed at scale, but the measurement infrastructure to justify them isn't there yet. If you're running this calculation, you're already ahead of most of your competitors.
How do you run this calculation for your own team?
You don't need a spreadsheet model. Four steps are enough to know where you stand within a week.
Step 1: inventory every AI tool. Ask each team member which AI tools they use, including free accounts and browser extensions. Frame it as discovery, not surveillance. Shadow AI research consistently shows organizations miss at least two tools per team in their first pass.
Step 2: measure one week of AI time. Ask three to five people to track how many minutes per day they spend writing prompts, adjusting output, and reviewing results. A tally on paper works fine. The goal is order of magnitude, not precision accounting.
Step 3: estimate error costs. Ask your team: how often per week does AI deliver something you need to completely rewrite, or that contains an error you catch later? Once a week is normal. Five times a week suggests the tool isn't right for the task.
Step 4: compare total costs with total savings. Add subscriptions, measured time costs, and estimated error costs. Compare to the time your team says AI saves. Positive gap? AI is generating real return. Negative or marginal? Worth looking at which tasks you should stop delegating to AI.
Where is the most waste hiding?
In practice, there are three places where AI costs accumulate without proportionate return.
Automating the wrong tasks. AI is strong where tasks are repeatable and "good enough" is an acceptable outcome: first drafts of emails, summaries of long documents, data processing. AI is weak where precision is essential and errors carry real consequences: legal language, financial calculations, sensitive customer communication. Teams that use AI for the second category spend more time on oversight than they save on execution.
Running too many tools in parallel. Enterprise AI adoption is accelerating across Europe and beyond. But at companies that have adopted AI, the average team is already running three or more separate tools, each with its own interface, its own limitations, its own learning curve. The overlap is large. Most teams would achieve the same outcomes with two tools, not five.
No shared knowledge about what actually works. AI governance policies help teams get more from AI, but most organizations miss a step before that: shared knowledge about effective use. One team member has found a prompt that saves twenty minutes on a recurring task. Another team member is still struggling with the exact same problem. Without a place to share working workflows, everyone reinvents the wheel.
What can you do with this this week?
Start small. You don't need to audit the entire organization. Three actions you can take tomorrow:
- Send a short survey to your team: "Which AI tools do you use, how often, and for what?" Three questions, five minutes to complete. The results will surprise you.
- Track one week of AI time for three people. Not as surveillance, but as data. The output is the starting point for your cost-benefit analysis.
- Cancel the tools nobody uses. Every team has subscriptions that were bought with enthusiasm and are now gathering dust. That's the easiest savings available.
The organizations getting the most from AI aren't the ones with the most expensive tools. They're the ones who know what AI costs, what it returns, and which tasks they've consciously decided not to delegate. That clarity starts with a calculation, and that calculation takes a week.
