Are your AI copilot seats paying for themselves? A renewal review
In brief
What Microsoft, Google, GitHub, OpenAI and Anthropic admin reports show on AI seat use, which metrics matter, who keeps a seat, and the per-seat price math.
Steve Elliott · · 7 min read
On this page
The admin reports can't tell you whether a copilot seat is paying for itself, but they can tell you which seats are idle, and in our view the renewal should turn on which seats show sustained use. The price bar is low enough that almost any regular use clears it: at Microsoft's own default rate of $72 an hour, a $30-a-month Microsoft 365 Copilot seat pays back, by our arithmetic, if it saves 25 minutes a month.
Microsoft and Anthropic seats generally can't be cut mid-term, so the review has to happen before the renewal date. This post compares what five vendors charge and report as of November 2026.
What does a copilot seat cost as of November 2026?
List prices from each vendor's own pages, per seat per month, in US dollars:
| Product | Annual billing | Monthly billing | What the seat includes |
|---|---|---|---|
| Microsoft 365 Copilot (enterprise add-on) | $30.00 | $31.50, annual commitment | Requires a qualifying Microsoft 365 license |
| Microsoft 365 Copilot Business (up to 300 seats) | $21.00; $18.00 promotion to December 31, 2026, first year only | $25.20 | Add-on to a Microsoft 365 Business plan |
| Gemini in Google Workspace | No separate seat; part of Business Starter ($7), Standard ($14) and Plus ($22) | $8.40, $16.80, $26.40 | Gemini in Docs, Sheets and other apps from Standard up; AI Expanded Access add-on extra |
| GitHub Copilot Business / Enterprise | Not listed separately | $19 / $39 | 1,900 / 3,900 AI credits a month, pooled; overage $0.01 a credit |
| ChatGPT Business, Standard / Premium seat | $20 / $100 | $25 / $125 | Seat types can be mixed; Enterprise priced by sales |
| Claude Team, Standard / Premium seat | $20 / $100 | $25 / $125 | Per-member usage limits; Enterprise seat fee covers access only |
If your business case dates from 2024 or 2025, note that Google stopped charging for its Gemini add-on after January 31, 2025 and added the features to its Workspace Business editions. GitHub moved Copilot Business and Enterprise to usage-based billing on June 1, 2026, and Anthropic's usage-based Enterprise plan bills usage separately at API rates, so on those plans a per-seat license is no longer the whole per-person cost.
What usage data does each admin console expose?
Every vendor reports active people against licensed seats; all but Anthropic give a last-activity date per person (Google's via audit logs). Definitions and delays differ:
| Vendor | Where | "Active" means | Per-person data | Windows and delay |
|---|---|---|---|---|
| Microsoft | Microsoft 365 Copilot usage report (admin center) | An intentional action, such as a prompt; opening the pane doesn't count | Last activity date per app; names hidden by default | 7, 28, 90 or 180 days; usually within 48 hours |
| Microsoft | Copilot Dashboard (Viva Insights) | At least one Copilot activity in the previous 28 days | Groups above a minimum size only | Previous 28 days; up to six days' delay |
| Gemini reports (Admin console) | Not defined; usage counted as feature uses, not unique documents | Active days, usage level, days at limit; last use via audit logs | 7, 28 or 90 days; 2 to 3 days' delay | |
| GitHub | Usage metrics dashboard and API | Interacted with Copilot on a given day | Per-person reports, AI credits used per day, last activity per seat | 28-day dashboard; data within two full UTC days |
| OpenAI | ChatGPT Enterprise and Edu workspace analytics | Sent at least one message in the period | First and last active day, message counts; CSV export | Custom range up to 12 months; target within 48 hours |
| Anthropic | Claude Team and Enterprise analytics | Not defined in the help article | Active members (aggregate); per-member spend and tokens on usage-based Enterprise only | Daily refresh; spend one day behind |
Two details trip up reviews. Google warns that the "Gemini Last usage" column in the Admin console may be inaccurate or incomplete, and points to audit logs instead. GitHub's data depends largely on IDE telemetry, so check telemetry settings before calling a developer idle.
Which adoption metrics mean something?
Of the AI adoption metrics these consoles report, we think three are worth putting in front of a budget owner. The rest help rollout teams more than renewals.
- Sustained use. Microsoft's Habitual level requires Copilot use in at least 9 of the past 12 weeks, so one curious week doesn't qualify.
- Days since last activity. Comparable across vendors. GitHub's reminder workflow example flags 30 days without use, or none since assignment.
- People at their limit. Google reports who hit AI feature limits and for how many days; Anthropic shows how many members are blocked or near their limit, but not who.
The weak ones are prompt counts, acceptance rates and the vendors' value estimates, which the vendors themselves caveat. Microsoft says Copilot is "not likely to be solely responsible" for metric differences in its dashboard. GitHub tells readers to treat its ROI comparisons as directional associations, not proof of cause. OpenAI says its impact surveys are not causal ROI measurements.
Take Microsoft's Copilot assisted value. It credits 6 minutes per search, summary or creation action at a default $72 an hour: $7.20 per action, by our arithmetic. Someone at the bottom of the Habitual band averages one action a week, about 4.3 a month. If each is a search, summary or creation action, that is about $31 of assisted value, more than the $30 seat. A metric almost every regular holder clears can't sort seats.
Self-reported time saved is weak too. The UK Government Digital Service's experiment with 20,000 civil servants reported an average of 26 minutes saved a day, but the report states that "the time savings presented throughout are self reported." The Department for Business and Trade's evaluation of 1,000 licenses found 72% of respondents satisfied, yet no strong evidence that time savings were improving productivity. METR's 2025 randomized trial found experienced developers took 19% longer with AI help while believing it had sped them up by 20%. METR now calls those results out of date and thinks developers are likely more sped up today, though on weak evidence.
None of this means the seats don't pay. The payback case has to come from the work, not the console; see AI initiative costs: tokens vs labor.
How do you decide who keeps a seat?
We suggest five steps, run 60 to 90 days before the renewal date:
- Export 90 days of per-person data. Microsoft's report hides names by default; unhide them for whoever runs the review.
- Sort into three groups. Sustained (most weeks active), occasional (some use, not most weeks) and idle (no activity in 60 to 90 days, or none since assignment).
- Warn the idle group, then reclaim. GitHub's documented pattern is to remind people first and say what happens if the license stays unused. Reassign reclaimed seats before buying more.
- Give the occasional group a cheaper fallback, not nothing. Microsoft 365 Copilot Chat is available at no additional cost with eligible Microsoft 365 plans. The equivalent move for anyone in the occasional group holding a Premium seat on ChatGPT Business or Claude Team is a Standard seat.
- Upgrade on evidence. Buy Google's AI Expanded Access only for people its reports show at their limit, and give Premium seats on ChatGPT Business or Claude Team only to people who report hitting Standard limits repeatedly, since Anthropic's analytics don't name them.
What does the price math look like?
An illustrative Microsoft renewal. The numbers are made up. A company holds 1,000 Microsoft 365 Copilot seats at $30 a month paid yearly: $360,000 a year. Its export shows 640 people active in the last 28 days and 200 with no activity in 90 days.
| Before | After cutting 200 idle seats | |
|---|---|---|
| Seats | 1,000 | 800 |
| Annual cost | $360,000 | $288,000 |
| Active in last 28 days | 640 | 640 |
| Share of seats active | 64% | 80% |
| Cost per active person a year | $562.50 | $450 |
The saving is $72,000 a year, and the active share rises 16 points without anyone using Copilot more. That is why, in our view, cost per active person belongs next to the active rate.
An illustrative GitHub case, where idle seats can cost nothing. A $19 Copilot Business seat adds 1,900 credits to a shared pool, and overage costs $0.01 a credit, so by our arithmetic the seat's credits are worth exactly $19. An organization that runs past its pool every month saves nothing net by removing an idle seat; one inside its pool wastes $19 per idle seat. How pooled credits work is covered in AI agent pricing models for buyers.
On Anthropic's usage-based Enterprise plan, the seat fee buys access only, so per-member spend is the number to review.
When can you cut seats?
Usually at renewal or the end of a billing cycle, which is why the review comes first.
- Microsoft. For accounts on a Microsoft Customer Agreement, licenses can be removed only within seven days of buying or renewing; later reductions show up on the first invoice after renewal.
- GitHub. Seats can be removed at any time, but charges continue until the end of the current billing cycle.
- Google. Removing an AI Expanded Access license mid-month may still incur the full month's charge.
- Anthropic. Removing a Team member brings no immediate refund or credit, though the seat can be reassigned. Self-serve Enterprise seats can't be removed mid-term.
Negotiated terms can differ, so confirm with procurement or your reseller. Who should own that renewal calendar is covered in who owns SaaS spend.
In our view, a seat review works best when each license is tied to the team, the work and the budget owner it serves, which is why seat decisions belong in the same operating model as those teams and budgets.
Sources
- Microsoft, Microsoft 365 Copilot enterprise pricing, accessed November 22, 2026
- Microsoft, Flexible Copilot plans for every organization (Copilot Business and Copilot Chat), accessed November 22, 2026
- Microsoft Learn, Microsoft 365 Copilot usage report, last updated August 18, 2026
- Microsoft Learn, Connect to the Microsoft Copilot Dashboard, updated October 7, 2026
- Microsoft Learn, Buy or remove licenses for a Microsoft business subscription, last updated July 8, 2025
- Google Workspace Admin Help, Workspace editions with AI and pricing changes, accessed November 22, 2026
- Google Workspace Admin Help, Review Gemini usage in your organization, updated November 10, 2026
- Google Workspace Admin Help, AI Expanded Access, updated November 10, 2026
- GitHub Docs, Copilot billing for organizations and enterprises, accessed November 22, 2026
- GitHub Blog, GitHub Copilot is moving to usage-based billing, April 27, 2026
- GitHub Docs, About Copilot usage metrics, accessed November 22, 2026
- GitHub Docs, Interpreting Copilot metrics, accessed November 22, 2026
- GitHub Changelog, AI credits consumed per user now in the Copilot usage metrics API, June 19, 2026
- GitHub Docs, Reminding inactive users and Reviewing user activity data, accessed November 22, 2026
- OpenAI, ChatGPT business pricing, accessed November 22, 2026
- OpenAI Help Center, Workspace analytics for ChatGPT Enterprise and Edu, accessed November 22, 2026
- Anthropic, What is the Team plan?, How is my Team plan bill calculated? and View usage analytics for Team and Enterprise plans, accessed November 22, 2026
- Anthropic, What is the Enterprise plan?, September 1, 2026, and How am I billed for my Enterprise plan?, October 13, 2026
- Government Digital Service, Microsoft 365 Copilot Experiment: Cross-Government Findings Report, June 2, 2025
- Department for Business and Trade, Microsoft 365 Copilot pilot: DBT evaluation report, published August 28, 2025 (PDF)
- METR, Measuring the impact of early-2025 AI on experienced open-source developer productivity, July 10, 2025, and uplift update, February 24, 2026
Frequently asked questions
What active rate should an AI copilot seat estate reach?
There is no standard. GitHub's guidance treats weekly active developers above 60% of licenses as strong ongoing use. We think the more useful test is sustained use, such as Microsoft's Habitual level of activity in at least 9 of 12 weeks, combined with cost per active person. An active rate can rise simply because idle seats were removed.
Can I trust Microsoft's Copilot assisted value figure?
Treat it as an estimate. Microsoft calculates it from assisted hours, using 6 minutes per search, summary or creation action, multiplied by a default rate of $72 an hour that admins can change. Microsoft calls these broad approximations and says Copilot is unlikely to be solely responsible for metric changes. By our arithmetic, one search, summary or creation action a week roughly equals a $30 seat.
Do I still need to review Gemini seats if Gemini is included in Workspace?
Not as a separate seat. Google now includes Gemini features in Workspace editions and stopped charging for the separate add-on in January 2025, so the decisions are which edition each group needs and who should get the AI Expanded Access add-on. Google's Gemini reports show active people, usage levels and who hit AI feature limits, which is the evidence for those choices.
How soon can I stop paying for an unused Microsoft 365 Copilot license?
Usually at renewal. For Microsoft Customer Agreement accounts, licenses can be removed only within seven days of buying or renewing, and later reductions show up on the first invoice after the renewal date. Between renewals, unassign idle licenses and give them to people waiting for one. Confirm the terms in your own agreement or with your reseller.
Steve Elliott
Published November 22, 2026
More on AI Economics
Tokens, inference cost, cost per task.
Keep reading
Seats, credits or outcomes: how to evaluate AI agent pricing as a buyer
Seat, credit, per-conversation and per-outcome prices each push a different risk onto buyers. This guide covers what each exposes you to and what to ask for.
Self-hosting an open-weight model vs paying per token: where the break-even sits
At Google Cloud prices as of November 2026, self-hosting gpt-oss-120b struggles to beat the same model sold per token. The break-even math, and when it wins.
Improving Call Center Renewal Rates: Strategies for Success
Call centers raise renewal rates with proactive outreach, well-trained agents, personalized offers and fast issue resolution. Strategies that work.
Why AI business cases fail the CFO's test, and what a fundable one contains
Most AI business cases count hours, skip the baseline and leave out run costs. What finance looks for, a one-page template and a worked full-cost example.