GitHub gives Copilot admins a lot to work with: seats with each person's last activity, usage metrics, a dashboard that links adoption to pull requests, and budgets down to one person. What your team adds is the link from each dollar to a team, and from each team to the work it shipped.
That link matters more since June 1, 2026, when GitHub replaced premium requests with GitHub AI Credits (announced April 27, 2026). Each Copilot Business or Enterprise seat now includes monthly AI credits, which chat and agents spend per token, at the model's price. So two people on the same seat can cost very different amounts, and the number of seats no longer tells you what Copilot costs. Prices change, so the ones here are from GitHub's pages in October 2026.
What GitHub already gives you
- Seats and last activity. The Copilot page in organization settings lists every seat, sorts people by when they last used Copilot, and exports an activity report as a CSV. GitHub's guide to user activity data notes that IDE activity appears only when the person has telemetry turned on.
- Usage metrics. The Copilot usage metrics API and dashboard show daily and weekly active users, chat and agent use, acceptance rates, lines of code, and pull request counts and time to merge. Reports go down to the repository and the user.
- An impact dashboard. It groups people by how deeply they use Copilot, compares the pull requests each group merges, and shows each group's monthly Copilot cost per developer from its actual AI credit use.
- Billing. The AI usage dashboard in enterprise billing settings shows credit use, the billing usage report shows it per user, and cost centers and budgets sit beside them.
What the reports leave out
Usage metrics measure activity: who used Copilot, how often, and what they accepted. That is useful during a rollout, but activity says little about what the work produced, and a thousand accepted lines can end in one merged change or in none. The impact dashboard comes closer, because it counts merged pull requests, but it compares groups of people, who may differ in many ways besides Copilot, so it cannot say whether a change you made last month, such as a new budget or a lighter model, did what you expected. GitHub itself calls its return figures "estimates rather than exact financial results".
Copilot is also often one of several coding tools a company pays for. Its metrics cover only Copilot, so a person who also has a seat on another coding tool looks like a light Copilot user, and the second bill does not show.
Valistry is built for that part: it puts every coding tool's seats and usage side by side, by team, and joins them to merged pull requests.
Give each dollar an owner
On Copilot, the owner is usually a team, and you can set it up without reading any code or prompt.
- Grant seats through teams. GitHub can give a seat to every member of a team, and the seat list then names the team that gave each seat.
- Put each team in a cost center. A cost center can hold an enterprise team, and its membership stays current as people join and leave. Copilot seats are charged to the users in the cost center, and so are AI credits, or to the organization that granted the seat when the person is in no cost center.
- Give each budget its own cost center. A budget applies to a whole cost center, so two teams that need separate budgets need separate cost centers.
GitHub says cost centers do not work with volume or subscription billing, so under a volume agreement, the team on each seat is the owner you can still use.
The levers GitHub offers
- Seats
- Unused seats$19 or $39 a seat each month
- The plan for each personBusiness $19, Enterprise $39
- Two coding tools for one personNo published figure
- Models
- The model for chat and agents$0.50 to $50 per million output tokens
- Long context tierInput costs twice as much past the threshold
- AI credits
- Included credits, pooled1,900 or 3,900 a seat each month
- Budgets by user, cost center or enterpriseNo published figure
- Paid usage past the pool$0.01 a credit, on by default
- Where requests run
- Data residency and FedRAMP10% higher model multiplier
GitHub publishes prices, not savings. Your bill shows what a change saved.
Seats.
- Unused seats. Copilot Business is $19 per user per month and Copilot Enterprise is $39, on GitHub's seats and billing page. A removed seat is billed until the cycle ends, so review the list a few days before it does. And because activity depends on telemetry, ask a quiet person before you take their seat.
- The plan for each person. Each plan includes AI credits worth its price: 1,900 credits on Business and 3,900 on Enterprise, at $0.01 each. On credits alone, an upgrade and paid usage cost the same: $20 buys 2,000 credits either way (our arithmetic). GitHub's plan guide says either can be the most cost-effective choice. We suggest choosing Enterprise for the features a team will use, since the upgrade is paid even in a month when the pool is not used up.
- Two coding tools for one person. Teams often try another coding tool beside Copilot, which is a good way to compare. GitHub publishes no figure here, and neither tool can see the other's seats, so decide per person when the trial ends.
Models. Code completions and next edit suggestions are not billed in AI credits and stay unlimited on paid plans. Chat, the CLI, cloud agent and code review spend credits at the model's token price. On GitHub's models and pricing page, output costs from $0.50 per million tokens on the lightest model to $50 on the most capable ones, at standard context length. The main lever is to use a light model for routine questions and a powerful one where the task needs it. On models with a long-context tier, input past the threshold of 200,000 or 272,000 tokens costs twice as much. Code review is the exception: GitHub picks its model, and each review also uses GitHub Actions minutes.
AI credits and budgets.
- The pool. Included credits are pooled across the billing account, so heavy users draw on what light users leave, and unused credits lapse at the start of each month.
- Paid usage past the pool costs $0.01 a credit and is on by default, says GitHub's usage-based billing page. An admin can turn off the AI credits paid usage policy to stop it.
- Budgets. A user-level budget caps what one person uses, from the pool and past it, and always stops at the limit. Cost center, organization and enterprise budgets cap only paid usage, and their setting to stop usage at the limit is off by default, so GitHub's budgets guide advises turning it on. The same guide notes that an enterprise budget is not a monthly total: 400 Business seats with a $5,000 budget can bill $12,600.
Where requests run. Data-resident and FedRAMP-compliant requests carry a 10% higher model multiplier, says the seats and billing page. Where your rules require them, they are the right choice.
If you knew Copilot by premium requests. Only existing annual Copilot Pro and Pro+ plans still use premium requests and model multipliers, until the plan ends, with extra requests at $0.04 each on GitHub's legacy billing page.
How to know a change worked
To see what a change did, measure the same work before and after. For Copilot, we suggest merged pull requests, because your code host already counts them and each one is a change your team chose to keep.
- Pick one team with its own cost center.
- Write down a baseline over four weeks. Its cost is its seats plus its paid usage past the pool, both on the bill, because credits used inside the pool are already paid for in the seat. Divide that cost by the pull requests the team merged to get the cost of one merged change.
- Change one thing, such as a lighter model for chat, a user-level budget, or the seats with no activity.
- Measure the same team over the same number of weeks, and compare cost per merged change rather than total cost, because the amount of work moves on its own.
Quality is part of the test. Check that the share of changes reverted and the time to merge held steady. Lines of code and accepted suggestions are useful signs of adoption, but on their own they say little about value, and the better change often adds fewer lines.
The test of a change is what one merged pull request costs the same team, with quality unchanged.
If your pool is rarely used up, a lighter model changes little on the bill, and the seat levers matter more.
What Valistry adds
Valistry reads Copilot seats, teams and activity through a read-only GitHub app, and stores each person as a code, never a login. It reports seat cost by team, shows seats with no activity in 28 days, and finds people with a seat on two coding tools at once. With merged pull requests from the same app, it shows what one merged change costs each team. The GitHub Copilot page lists what it reads.
With or without Valistry, we suggest the same order: give each seat a team, choose the lever that fits, and let merged changes show whether it helped.