AI tool pricing traps: how $20 a month becomes a $500 bill
The short answer
An AI bill grows in exactly three ways: seats, usage units, and overage. A $20 seat plan is already a $500 a month bill at 25 people, before anyone runs a single task above the allowance.
The three levers
| Lever | How it grows | What it looks like |
|---|---|---|
| Seats | Multiplies with headcount | $20-$25per seat, per month |
| Usage units | Multiplies with volume | $0.99per AI resolution |
| Overage | Starts after the included allowance | variesper minute, call or task |
Most buyers negotiate the first lever and ignore the other two. That is how a plan that reads as $20 a month turns into a bill that reads as $500.
The worked example: 3, 10 and 50 people
Take a published seat price, $20 per user a month on annual billing and $25 monthly. Add admin features by moving from individual subscriptions to a team workspace. The only variable is how many people need it.
| People | Individual plans | Team plan, annual | Team plan, monthly |
|---|---|---|---|
| 3 | $603 x $20 | $603 x $20, plus admin | $753 x $25 |
| 10 | $20010 x $20 | $20010 x $20 | $25010 x $25 |
| 50 | $1,00050 x $20 | $1,00050 x $20 | $1,25050 x $25 |
Two things fall out of that table. First, the annual against monthly choice is worth 20 percent, which at 50 people is $250 a month. Second, at three people the team plan costs the same as three separate subscriptions while adding admin controls and single sign on. That is the easiest upgrade on the page, and the one most small teams miss because the individual and business plans sit on different tabs.
The full seat comparison, including the premium seat type and the minimum seat count, is in our guide: ChatGPT for Business vs Plus.
Where usage units bite
Seats are predictable. Usage is where the surprises live, because every vendor counts a different thing.
- Automation tools count steps. A five step workflow run 10,000 times is 50,000 tasks, not 10,000. One vendor bills per task and counts every step in the chain; another bills per operation and starts at $9 for 10,000 operations. Same workflow, different bill.
- Support AI counts resolutions. At $0.99 per resolved conversation, 10,000 resolutions a month is $9,900, and the vendor decides what counts as resolved. A human handled ticket in the same month runs $6.00 to $13.50 each.
- Voice counts minutes. Per minute billing punishes long calls, and a transferred call is often billed twice.
The comparison of how each automation platform bills is here: Zapier vs Make vs n8n, and the per resolution maths is here: AI customer support cost per resolution.
The five questions to ask before you scale
- What counts as one billable unit, in the vendor's own words?
- What is the overage rate once the included allowance runs out?
- Is there a seat minimum, and what happens if we shrink?
- What is the annual price against the monthly price, exactly?
- Which features cost extra, such as admin controls, single sign on or API access?
When per seat wins, and when usage wins
Per seat is cheaper when most people use the tool lightly, because the price is predictable and you can forecast it from the payroll. Usage pricing wins when a few power users generate all the volume, since you are no longer paying for people who open it twice a week.
The crossover is not company size, it is active users against total volume. Measure both for a month before you sign a year.
Questions people ask
How does an AI tool go from $20 to $500 a month?
Seats, usage and overage. A $20 seat plan is $500 a month at 25 people, and usage charges such as $0.99 per AI resolution sit on top.
Which is cheaper, per seat or usage based pricing?
Per seat for light users, usage pricing when a few people drive most of the volume. The crossover is active users against volume, not headcount.
What should I ask before scaling?
What a billable unit is, the overage rate, the seat minimum, the annual against monthly price, and which features cost extra.