Why Per-Seat AI Billing Overcharges Most Teams
You buy a seat for everyone, then a handful of people use it and the rest barely log in. Usage is a power law, so a flat per-seat price charges the quiet majority to subsidize a few power users. Here is the math, and the alternative.
When does usage cost less than a $20 seat?
A seat is a flat bill whether someone opens the tool or not. In this illustrative example, 87% of users cost less than $20 a month. Compare your team’s actual usage, equivalent features, and any platform fees before choosing a plan.
What one person’s AI use would cost per month · share of people
Think about how you pay for streaming. You hold a subscription to one service, watch it a few nights a month, and pay full price the whole time. Then you add a second service, and a third, and now you are paying four flat fees for content you mostly do not watch. The waste is not any single subscription. It is the pile of them, each priced as if you used it every day.
Company AI is heading the same way, and faster. Most teams now hold a per-seat subscription to at least one assistant, and often several. Each seat is priced as if the person on it uses AI all day. A few of them do. Most do not. And when you stack several tools, each with its own per-seat fee, the same waste compounds, one idle seat at a time.
This is not a knock on the tools. It is a problem with the pricing model. Here is what the data actually shows.
The seat you bought is mostly idle
Start with utilization. Across business software, the average organization uses only about 54% of the licenses it pays for, which leaves roughly 46% of applications under-used or sitting idle. AI seats are worse, not better. Independent tracking of Microsoft Copilot found that only about 36% of licensed employees actively use it, with sustained weekly use settling somewhere in the 30% to 55% range in the first year. Once you count the idle seats, the effective cost per active user runs closer to two or three times the sticker price.
The pattern shows up at the decision level too. Gartner reported that only 5% of organizations that ran an AI pilot moved it to a broad deployment. Companies keep paying for the seats regardless. The seat is easy to buy and easy to forget, which is exactly what makes flat per-seat pricing so quietly expensive.
Usage is a power law
The deeper issue is how AI use is distributed. It is not a bell curve where most people land near the average. It is a power law: a small group of heavy users does most of the work, and a long tail barely engages. In one analysis of enterprise AI activity, the top 5% of users generated about 144 conversations while the bottom half had 12 or fewer. Task usage concentrates the same way, with the bottom 80% of task types accounting for only about 13% of activity.
That distribution is the whole story. When you translate real usage into what it would actually cost at metered rates, most people would spend a few dollars a month. A flat $20 seat only pays off for the heavy few. Everyone else quietly pays full price for a fraction of the use, month after month.
You are not paying for AI. You are paying for seats, most of which sit idle.
What usage-based billing changes
The fix is to stop pricing by head and start pricing by use. Put the whole company on one account with a shared, prepaid balance, and charge only for the AI each person actually consumes. The heavy few draw down real spend, which is exactly what you want. The quiet majority cost cents, because that is what they use. Nobody gets cut off mid-task by a session limit, and a slow month simply costs less.
This also fixes the stacking problem. One account can run more than one model, so you are not buying a separate per-seat subscription for each vendor and eating the idle time on all of them. You add usage credit once and spend it across whatever model fits the task.
A single idle seat is a small leak. The real cost is the pile: several AI subscriptions, each billed per head, each mostly unused, multiplied across a whole company. Usage-based pricing collapses that pile into one bill that tracks reality. You stop paying a flat fee times the number of tools times the number of people, and start paying for the work that actually happened.
- Light users cost the same as heavy users
- Idle seats are pure waste, and easy to forget
- Locked to one vendor's models per subscription
- Stack several tools and the waste multiplies
- Little visibility into who actually uses what
- Light users cost cents; heavy users draw real spend
- A quiet month simply costs less, with no session limits
- Many models on one account, switch freely
- One balance instead of a stack of flat fees
- Spend and usage are visible, per person and per team
Keep your power users where they are
None of this means taking premium tools away from the people who live in them. If 5% of your company is heavy enough to justify a top-tier direct subscription, let them keep it. That is the right call for them. The point is that you should not be paying the same rate for the other 95%, who would cost a fraction of it on usage-based billing and often use AI more, not less, once it is easy and governed.
The move is simple: put the whole company on one usage-based account, give each person the right model and a sensible spending limit, and let your power users go premium where it pays off. You cover everyone, you see where the value actually is, and you stop paying flat fees for seats that sit idle.
That last part matters as much as the cost. When usage is visible per person and per team, you can give the right model and the right budget to the right people, instead of handing everyone the same expensive seat and hoping. The savings are the hook. The control is what keeps it working.