Seat based pricing

Because it decides who gets access, and that decision shapes adoption more than any training programme. When every additional person carries a fixed monthly cost, the rational move is to limit seats to the people most likely to use them, which means the people who might have discovered a use never get the chance. Organisations then conclude the tool has limited application, when what they measured was a decision they made at purchase. The pricing model quietly authored the adoption result and nobody recorded it as a cause.

The unit is a named person, not usage. A seat is normally tied to an identity rather than to a concurrent session, so it cannot be shared between two people who work different hours without breaching the terms. That distinction matters more than it sounds, because it is where the temptation to share credentials comes from and where account hygiene starts to erode.

Cost is predictable and that is the genuine selling point. Multiply people by rate and you have the year, which makes budgeting straightforward and approval easier. For a tool with roughly even usage across a team, this is a perfectly sensible arrangement and there is no cleverness required.

Waste concentrates in the people who never adopted. Usage in most organisations is heavily uneven: a minority of people account for most of the activity, and the tail costs the same as the head. Nobody notices because the bill does not itemise, which is why a seat audit before renewal is usually the fastest saving available.

The model sits awkwardly with what AI actually costs to run. A per-answer cost varies enormously with how much text is read and produced, so a flat per-person charge is a vendor absorbing that variance. They cover it by capping something, which is why seat-priced AI products so often carry message limits, model tiers or slowdowns after heavy use, and why those caps feel arbitrary until you see the reason.

Many vendors now pair it with something metered rather than abandoning it, so the predictable part covers access and a variable part covers heavy work. Whether a given plan does that is a question for the pricing page rather than an assumption: ask what the seat includes, and what is charged separately once it is exceeded. A hybrid is harder to forecast than either pure model, and knowing you are on one is most of the difficulty.

It punishes exactly the pattern AI adoption starts with. Early use is a few enthusiasts and a long tail of curious people trying something once, which is the worst possible shape for per-person charging and the best possible shape for the usage-priced alternative. Teams that pilot on seats often draw a pessimistic conclusion from an expensive experiment.

Adding seats is usually instant and removing them usually is not. Most agreements let you grow at any point, charged from the moment you add, while reductions take effect at renewal rather than when somebody leaves. That asymmetry is ordinary commercial practice rather than sharp dealing, and it means a seat count drifts upwards across a year unless somebody deliberately reconciles it. The reconciliation is worth diarising before the renewal conversation rather than during it, because that is when you still have the option of not renewing the surplus.

It makes the leavers process a billing question, which nobody expects. When a seat is tied to an identity, somebody has to reclaim it when a person changes role or leaves, and that step lives with whoever administers the tool rather than with the process that handles everything else. Where it is not wired in, an organisation quietly pays for the access of people who left, and the invoice reports a total that looks stable precisely because it is.

• Price a workspace assistant across a whole department and notice that the quiet half of the team costs the same as the active half.

  Notion AI

• Look at a seat-priced assistant's message caps and model tiers, which are how a flat charge survives a variable underlying cost.

  ChatGPT

• Compare that with paying per request through a routing layer, where the same light user costs almost nothing.

  OpenRouter

• Check whether a tool charges per person at all, since anything you run on your own hardware moves the cost to operating time instead.

  Ollama

How do we decide whether it suits us?

Look at how evenly the work is spread. If most of the team will use the tool regularly, per-seat is predictable and fair. If usage will be concentrated in a few people with a long curious tail, you are paying full price for the tail, and a usage-priced or hybrid arrangement will usually cost less for the same access.

What is the fastest saving available on a seat-priced tool?

A seat audit before renewal. Usage is almost always more uneven than anybody expects, and dormant seats do not announce themselves because the invoice reports a total rather than a distribution. Ask the vendor for per-user activity, which administered tiers generally expose, and reconcile it against what you are paying for.

Should we restrict seats during a pilot?

Only if the budget forces it. Restricting access to people already enthusiastic produces a flattering and uninformative result, because you learn that the tool suits the people you selected for suiting it. Wider access with modest expectations is a better experiment, and it is the one that surfaces uses nobody predicted.

Why do seat-priced AI tools have message caps?

Because the cost of answering varies enormously and the charge does not. The vendor is absorbing that variance, and caps, model tiers and slowdowns after heavy use are the mechanisms that make it survivable. They read as arbitrary restrictions until you notice they are all solving the same arithmetic problem.

How should we handle people who need occasional access?

Ask whether the vendor offers a lighter tier or a shared allowance before buying full seats for occasional users, because that is the case per-seat charging handles worst. Where no such tier exists, it is worth pricing a usage-metered product alongside for that group rather than accepting full price for people who will log in monthly.

Is the industry moving away from it?

It is drifting rather than abandoning. The common shape now pairs a seat charge for access with metered components for heavy work, which is harder to forecast than either pure model. Expect to be quoted a hybrid and to have to model both halves against the volume you actually expect.

• The bill follows how many people have access, not what they do.
• Predictability is the genuine advantage and the reason finance departments prefer it.
• Waste sits in the dormant tail, and the invoice will never show you where.
• Caps and model tiers exist because a flat charge sits on a variable cost.
• Restricting seats during a pilot buys budget certainty and costs you the finding.

• 

  Notion AI

  Per-person pricing across a department, where the quiet half costs the same.

• 

  ChatGPT

  Caps and model tiers as the mechanism that makes a flat charge survivable.

• 

  OpenRouter

  The contrast: paying per request, where a light user costs almost nothing.