Your Team May Be Paying Too Much for AI Seats

Buying everyone an AI subscription is a simple way to give your company access, and a great way to get started if you haven’t already taken that step. It’s also an easy way to pay more than you need to.

Some employees will use AI throughout the day. Others will open it for an occasional draft, summary, question, or task. A flat subscription charges the same amount for both.

If you buy seats broadly and adoption is uneven, you’re probably paying more than you need to. Usage-based billing deserves a look, whether you’re reviewing subscriptions you already have or deciding how to get started, especially if your employees are still warming up to AI or their needs are occasional.

Buying access does not guarantee regular use

A six-month field experiment covering 7,137 knowledge workers across 66 firms found substantial differences in Microsoft 365 Copilot adoption. During the final three months, 20% of workers given access never used it, while only 5% used it every week. The median worker used it in roughly one-third of those weeks.

Company-level differences were even larger: average weekly use ranged from 6.3% at one firm to 70% at another. These were early deployments, and the researchers caution against treating the participants as representative of every workplace. Still, the finding matters for purchasing: the same license can see very different levels of use across employees and companies. Read the study.

Other deployments achieved stronger adoption. A UK government trial involving 20,000 employees maintained approximately 80% monthly active use among the 14,500 users with available usage data. “Active” meant at least one interaction in the previous 30 days. Read the government report.

There is no universal percentage of unused seats. It will vary from company to company, team to team, person to person, and use case to use case.

Consider what adoption and usage look like today, how you expect them to develop, and whether the pricing model you’ve committed to fits where you’re likely to land.

Light use changes the cost comparison

A subscription buys access to a product for a fixed price. API pricing varies by provider and service. Charges may be based on the amount of information processed and generated, individual requests, or bundles of usage. The cost comparison depends on the rates and billing structure of the services you use.

An application built on those APIs can charge customers based on their consumption, alongside any platform or service fees.

For someone making occasional, short requests, metered model charges can fall well below a monthly subscription. For someone processing lengthy documents or running intensive automated work, they can rise quickly.

That creates an opportunity to match spending to demand. But comparing a subscription with raw API charges is only part of the calculation.

The subscription buys more than model access

If someone uses AI once a month, a recurring subscription deserves scrutiny. But the cost of their model usage alone does not tell you whether that subscription is worth keeping.

An AI application provides more than access to an underlying model. It can include an easy-to-use interface, connections to your business tools, file handling, and workflows that make the technology useful without requiring employees to build anything themselves.

Claude Cowork is one example. Its application brings together capabilities for working with files, tools, and connected services. Accessing the same underlying models through an API does not automatically give you that complete experience. Those capabilities have real value. Learn about Cowork.

The fair comparison is what it costs to give your employees the capabilities they need. Another application may provide them through a different pricing model. A custom solution may also make sense, provided you account for building, maintaining, and supporting it.

A simple hypothetical example shows the decision. One hundred seats at $20 per month cost $2,000. If an alternative delivers the required capabilities for $1,200 in total monthly platform and usage charges, the saving is $800. If it costs $2,200, there is no saving.

The useful number is the full bill for equivalent work.

The cost of getting through the door

For occasional users, a per-seat subscription creates a minimum recurring cost before they do any work. Multiply that across a team still finding its uses for AI, and the cost of providing access can become significant.

You may not have a usage-only option for the product you want. Cowork, for example, is currently available through paid Claude plans. Choosing that experience means accepting its pricing structure. Plan availability.

We would like to see more providers offer useful, complete applications with a lower fixed cost of access and charges that grow with use. Usage is not a perfect measure of value, but it gives occasional users a way to participate without the same recurring commitment as someone who relies on AI all day.

Until then, consider both the capabilities and the pricing. Keep subscriptions where the complete product earns its cost, and explore alternatives where lighter use makes a different model more economical.

Keep subscriptions where they earn their place

Frequent users may get excellent value from a subscription. So may an occasional user who depends on a particular integration or feature.

Someone who uses AI once a month to complete a valuable task can still justify the price. Low frequency is a reason to investigate the cost, rather than automatic evidence of waste.

A practical purchasing approach can combine several options: subscriptions for employees who benefit from them, metered access for lighter workloads, and reclaimed licenses where access is no longer needed.

The goal is to give people useful tools at a cost that makes sense for their work.

Measure before the next renewal

Start with a representative month or two. Account for seasonal work and give new users time to learn the tools.

Review:

• Paid access: seats purchased, seats assigned, and overlapping subscriptions.
• Actual use: who uses each tool, how often, and for which tasks.
• Required capabilities: integrations, document handling, security, and administration.
• Alternative costs: representative workloads priced with all platform and usage fees included.

Where your current tool does not expose enough information to estimate API costs, test representative tasks in the alternative. A login count cannot tell you how expensive a workload will be.

You may find that your current subscriptions are good value. You may find a group of light users who would cost much less on metered access. You may find licenses that can simply be reassigned or removed.

If you gave everyone the same AI subscription before understanding their needs, revisit that decision. Keep broad access to useful AI, and let the evidence from your own team guide how you pay for it.