How can companies effectively monitor and control AI spending with this new model?

Question

Accepted Answer

Companies can control spending by using Microsoft’s built-in caps and keeping the service off by default.

Microsoft says the new Copilot Cowork model bills each task separately, based on computing power used, so companies can manage costs by setting limits at the employee, team, or department level. The service is disabled by default, which is meant to prevent unexpected charges when the AI agent is not intentionally switched on.
Spending limits can be set at three levels

Companies are able to cap spending per employee, per team, or per department. That gives administrators a way to place a ceiling on usage before bills rise too far.
Billing is tied to each task run by the agent

Every task carried out by Copilot Cowork is charged separately, rather than through one fixed licence fee. The cost depends on how much computing power the task consumes, so heavier use will cost more than lighter use.
The service starts in an off state

Copilot Cowork is disabled by default. That means the AI agent does not begin running tasks automatically, which reduces the risk of charges appearing before a company has chosen its limits.
Users can also choose cheaper or more powerful models

Microsoft says customers will be able to choose which model is used, with more powerful models costing more and less powerful ones costing less. At general availability, Copilot Cowork runs on Anthropic models including Opus 4.8 and Sonnet 4.6, while Frontier-tier customers can use GPT 5.5, and a cheaper model called Cowork 1 is due soon for everyday tasks.

Reading history

Hide →

This history is cleared when the page is closed.

How can companies effectively monitor and control AI spending with this new model?

Loading

Was this article helpful?

Helpful

Not helpful

Report an issue

1

0

0

Share with friends