Review the Preview Bill First
When Copilot usage is metered by tokens, governance shifts from "Can we use this?" to "Can we control spend without slowing teams down?" Before configuring any controls, build a baseline of what current usage would look like under the new model.
A baseline keeps the rollout grounded. Budgets set without observed data tend to either block legitimate work or fail to catch real overspending.
Set Budgets at the Right Scope
GitHub provides four budget scopes. Each one serves a different governance need, and admins should pick scopes deliberately rather than stacking all four.
- Enterprise budget — global control and centralized oversight.
- Organization budget — business-unit or tenant-level ownership.
- Cost center budget — the direct answer for per-team accountability.
- User budget — individual limits and exception control.
Avoid overlapping budgets that could unexpectedly block users. When a user-level budget is exhausted, that user's Copilot access is halted even if the organization credit pool still has credits available.
Why cost centers are the right unit for per-team budgets
- Cost centers map cleanly to engineering team ownership.
- Costs become attributable without manual per-user credit tracking.
- A user belongs to one cost center at a time, which simplifies reporting.
- GitHub teams do not automatically sync to cost centers — plan identity mapping (ideally via SCIM) as part of onboarding and offboarding.
Decide When Budgets Should Act as Caps
A budget can either monitor or enforce. Both roles are valid; admins should be explicit about which is which.
Configure Alerts That Drive Action
Alerts are useful only if the recipients can act on them. Default to a small set of thresholds tied to clear owners.
- Enable budget threshold alerts at 75%, 90%, and 100%.
- Enable included usage alerts at 90% and 100%.
- Assign recipients who can act quickly: billing managers, platform owners, and engineering managers.
Treat 75%, 90%, and 100% alerts as action points, not just informational emails. Document the expected response for each threshold so alerts do not become noise.
Define the Additional Usage Policy
When included AI Credits are exhausted, admins choose what happens next. The policy should be explicit, written, and shared with engineering leadership.
- Allow additional usage — usage continues at published rates after the included pool is exhausted.
- Block additional usage — usage is blocked until the next billing cycle.
- Mixed policy — allow additional usage for approved cost centers, block it elsewhere, and require explicit exceptions for the rest.
Map Ownership Before You Need It
Spend controls fail when nobody owns the response. Assign owners during configuration, not during the first incident.
- Budget configuration owner.
- Alert response owner.
- User communication owner.
- Cost center review owner.
- Exception approval owner.
Caveats to Plan For
A few operational details matter during rollout. Bake them into the runbook before June 1.
- Included credits are pooled at the billing entity for Business and Enterprise — individual heavy users can consume more while lighter usage offsets total spend.
- There is no automatic fallback to cheaper models when a budget is exhausted. Usage either continues at higher cost or stops, based on policy.
- Seat additions increase the credit pool immediately.
- Seat removals take effect in the next billing cycle.
- Copilot code review consumes both AI Credits and GitHub Actions minutes — account for it in both budgets.
Suggested Operating Model
| Area | Recommendation |
|---|---|
| Default model behavior | Use auto model selection where available |
| Budget strategy | Enterprise/org monitoring budgets plus targeted cost-center or user caps |
| Alerting | 75%, 90%, 100% budget alerts plus included usage alerts |
| Exceptions | Allow additional usage only for approved users, teams, or cost centers |
| Review cadence | Weekly during preview, then after each billing cycle |
| Communication | Share guidance before June 1, refresh after first month of usage data |
Summary
Usage-based billing makes Copilot spend a platform-engineering concern, not just a procurement line item, and admins need controls that match team accountability.
Multi-scope budgets, hard caps, threshold alerts, an explicit additional usage policy, and clear ownership turn the new billing model into layered, predictable governance.
Admins enter June 1, 2026 with budgets configured, alerts wired to the right responders, and an exception process that does not slow engineering down.
References
- Usage-based billing for organizations and enterprises
- Set up budgets
- GitHub announcement: GitHub Copilot is moving to usage-based billing