Interpreting usage and adoption metrics for GitHub Copilot

Use this article to interpret usage and adoption metrics across the Copilot usage metrics and impact dashboards, and to identify opportunities to increase adoption and engagement.

Reviewing overall usage trends

Use the main usage charts in the dashboard to understand overall adoption and engagement patterns. These charts help you identify where usage is growing, leveling off, or declining, so you can take action to maintain engagement.

Metric	What it shows	How to interpret it
IDE daily active users (DAU)	Unique users who interacted with Copilot each day.	Sustained DAU growth signals consistent engagement; sharp declines may indicate configuration issues or reduced interest.
IDE weekly active users (WAU)	Unique users active over a 7-day rolling window.	A healthy WAU-to-license ratio (>60%) indicates strong ongoing usage.
Code completions acceptance rate	Percentage of suggestions accepted.	A rising rate suggests increasing trust and usefulness; a drop may point to mismatched suggestions or workflow friction.

Reviewing feature adoption

The "Requests per chat mode" and "Agent adoption" charts show how developers are using Copilot Chat and Copilot agent.

Signal	What it tells you	What to look for
Requests per chat mode	Breakdown of chat interactions by mode—Ask, Edit, Plan, or Agent.	A balanced distribution suggests users are exploring multiple capabilities. Heavy use of one mode can highlight where enablement should focus.
Agent adoption	Percentage of active users who used Copilot agent.	Growth over time shows that developers are progressing from basic completions to more advanced Copilot features.

Reviewing customization adoption with the API

To assess how developers adopt customizations across Copilot CLI, VS Code, and other supported clients, use the usage metrics API. It reports skills, custom agents, Model Context Protocol (MCP) server connection activity, slash commands, and plugins for enterprises and organizations. The fields appear in per-user and aggregated 1-day reports, per-user 28-day reports, and aggregated 28-day day_totals. They are not available in the dashboard. For field definitions, supported clients, and minimum versions for reporting names, see Data available in Copilot usage metrics.

interaction_count represents a different event for each category: skill invocations, custom agent starts, MCP connection or reconnection attempts, slash command invocations, or plugin-associated skill invocations. These values count activity events, not users. MCP counts include both successful and failed connection attempts, but not tool calls through an already connected server or ongoing server use. Every plugin interaction is already included in skill interaction counts, while skill interactions that are not associated with a plugin appear only in skill totals. Do not add plugin and skill interaction counts together.

To identify strong adoption signals, look for configured customization names that appear in many users' arrays. Custom names appear in both per-user and aggregated enterprise and organization reports. Use aggregated reports to see overall adoption and per-user reports to see who uses each customization. Each customization array retains up to 25 named entries per record, plus a trailing other entry for remaining activity.

The data comes from client telemetry and can undercount activity. An absent name does not mean zero usage: the customization may be outside the retained entries, its name may be unavailable, or client telemetry may be missing. Use these reports to understand trends, not to calculate exact adoption rates or verify that only approved customizations are in use.

Distinct counts show how many different item identifiers had activity. At enterprise and organization scope, an identifier counts once even if many users interacted with it, so do not add per-user distinct counts to reproduce an aggregate value. Compare distinct counts across records or day_totals to see whether the variety of customizations with activity is changing.

Signal	What it suggests	Suggested action
A named skill, custom agent, slash command, or plugin has high interaction counts across many users	The customization is broadly useful and may be a good candidate for standardization.	Promote it through internal examples, onboarding materials, or a shared customization library.
A named customization appears in the retained entries for only a few users but has high interaction counts for those users	Observed activity is concentrated among a small group. Other users may have activity outside their retained named entries.	Ask those users to share example workflows. Then use developer feedback or other internal data to assess broader adoption before expanding enablement.
Distinct counts are high, but most interactions are grouped under other or custom	Custom names may be unavailable, filtered because they look like file paths, or outside the retained entries.	Check client versions and telemetry coverage before drawing conclusions about which customizations are in use.
Interaction counts remain low for a skill, custom agent, slash command, or plugin you expect developers to use	Developers may not know that the customization exists or when it is useful.	Improve discoverability with task-focused examples and include the customization in relevant onboarding or enablement sessions.
MCP connection activity is higher or lower than expected	Developers may be reconnecting repeatedly, encountering failed connection attempts, or the planned rollout may not have reached the intended audience. Connection counts alone do not show whether developers are actively using MCP tools.	Check your MCP configuration and rollout process, then use other internal data or developer feedback to assess ongoing server use.
Adoption grows after an enablement activity	The activity may have helped developers discover useful customizations.	Compare multiple reporting periods and gather developer feedback before deciding whether to expand the activity.

Reviewing model adoption

The "Model usage per day" and "Model usage per chat mode" charts help you understand which AI models are most frequently used.

Chart	Description	Insights to derive
Model usage per day	Shows which AI models power Copilot Chat activity.	Identify whether users are primarily engaging with default models or experimenting with newer ones.
Model usage per chat mode	Breaks down model usage by chat mode (Ask, Edit, Plan, Agent).	Monitor how model adoption evolves as new models are released.

Reviewing language usage

The "Language usage" and "Language usage per day" charts show which programming languages developers use most often with Copilot.

Chart	Description	How to use it
Language usage	Shows the share of Copilot Chat activity by programming language.	Identify where Copilot Chat provides the most value and where additional support or enablement might help.
Language usage per day	Tracks daily fluctuations in language activity.	Spot shifts in development focus or confirm whether new teams or projects are driving increased activity.

Reviewing adoption cohorts

The impact dashboard's adoption cohort distribution shows the share of users in each phase, and how that distribution shifts over time. For what each phase means and how classification works, see GitHub Copilot usage metrics.

Signal	What it suggests	Suggested action
Shrinking Phase 1 share, growing Phase 2 or Phase 3 share	Users are progressing from code completions toward agent-driven workflows.	Continue current enablement; consider sharing internal success stories to sustain momentum.
Large or growing passive users share	Many licensed users haven't yet reached the engagement threshold for a phase.	Verify license assignment and IDE setup; offer onboarding to re-engage inactive users. Also check whether affected users rely primarily on a surface that isn't part of any phase's classification criteria, such as Copilot Chat on GitHub.com.
Stalled progression from Phase 1 to Phase 2	Users are comfortable with completions and chat but haven't adopted agent surfaces.	Follow a recommendation card, such as configuring Copilot cloud agent, or run a demo of agent workflows.
Stalled progression from Phase 2 to Phase 3	Users have adopted one agent surface but haven't layered in additional agent surfaces.	Follow a recommendation card, such as enabling Copilot code review, to introduce a second agent surface.
A user's phase changes from one day to the next without a change in their habits	Phase assignment is recalculated daily from the trailing 28-day window, so a user's phase shifts as older days roll out of the window.	This is expected behavior, not a data error; no action needed.

When you see a recommendation card in the dashboard, such as configuring Copilot cloud agent, enabling Copilot code review, or sending an onboarding nudge, follow the linked how-to to complete the setup. Each recommendation is generated from your organization's current cohort distribution and targets the gap most likely to move users toward deeper adoption.

Reviewing the adoption multiplier

The adoption multiplier compares engaged users (Phase 1, 2, or 3) against passive users on code shipped (pull requests merged per user per month) and time to merge pull requests. This shows the relative impact of deeper adoption, independent of how many users fall into each phase.

Signal	What it suggests	Suggested action
Engaged users merge more pull requests per user per month than passive users	Deeper Copilot adoption is associated with higher individual output.	Use this comparison to support the case for continued enablement investment.
Engaged users merge fewer pull requests per user per month than passive users	Engaged users may be spending more time on larger or more complex work, or the comparison may reflect differences in team composition rather than adoption depth.	Cross-reference with adoption cohort distribution and team-level data before drawing conclusions; avoid treating the multiplier as a standalone measure of productivity.
Engaged users merge pull requests faster than passive users	Deeper adoption is associated with a shorter pull request lifecycle.	Use this comparison alongside pull request lifecycle metrics to evaluate overall delivery impact.
Multiplier stays flat despite growth in Phase 2 or Phase 3 share	Agent adoption hasn't yet translated into a measurable difference in output between engaged and passive users.	Check whether agent usage is concentrated in a few users or teams, and whether enough time has passed since rollout for the impact to show.

Reviewing industry comparisons

The Industry comparison widget compares your enterprise's agentic adoption and leverage with aggregated results from participating peers. Review the percentile ranking, peer median, top decile, and peer distribution together. Do not treat any single value as a target.

Signal	What it suggests	Suggested action
Agentic adoption is above the peer median	A larger share of your users have adopted agent-first or multi-agent workflows than in the comparison cohort.	Continue the enablement practices that support agentic workflows, and review which teams still have lower adoption.
Agentic adoption is below the peer median	Your enterprise has room to expand adoption of agent-first and multi-agent workflows.	Review the adoption cohort distribution, then prioritize onboarding or enablement for teams that remain passive or code-first.
Leverage is above the peer median	The ratio of merge rates between agentic and non-agentic users is higher in your enterprise than in the peer cohort.	Review the practices of teams with strong agentic adoption and consider whether those practices can be applied more broadly.
Leverage is below the peer median	The ratio of merge rates between agentic and non-agentic users is lower than for the comparison cohort.	Check team composition, project complexity, and how recently agentic workflows were introduced before changing your rollout plan.
Your ranking improves across reporting periods	Your enterprise's position has improved relative to the participating peer cohort.	Confirm that the improvement appears alongside sustained adoption and pull request trends before expanding enablement investment.
Results are unavailable	There are not enough eligible participating peers of the same company size to provide the comparison, even after the dashboard expands the peer cohort to all industries. Results are also unavailable when your own score can't be calculated, such as when there are no active users or all active users are already agent-first or multi-agent users, leaving no non-agentic users to compare leverage with.	Review your enterprise's or organization's own adoption and leverage trends where available until a peer comparison is available.
Your ranking changes but your score doesn't	The participating peer cohort changed, or the dashboard compared you with a different cohort than in the previous period.	Check which peer cohort the ranking line shows before comparing results across periods.

Industry comparisons show associations, not causation. Differences in team composition, project type, engineering practices, and adoption maturity can affect both measures. The participating peer cohort can also change between reporting periods.

Reviewing potential return on investment

The Potential return on investment section compares Copilot cost, estimated percentage of payroll, and pull requests per developer per month across earlier and agent-first adoption phases. Treat these comparisons as directional associations, not proof that a change in adoption caused a change in output.

Signal	What it suggests	Suggested action
Agent-first users merge more pull requests with a similar or lower cost as a percentage of payroll	Agent-first usage is associated with more pull request output relative to the estimated investment.	Compare this signal with the adoption multiplier and trends over multiple periods before using it to inform enablement investment.
Agent-first users merge fewer pull requests or have a higher cost as a percentage of payroll	Differences in work type, team composition, or adoption maturity may be affecting the comparison.	Review cohort distribution and team-level context before drawing conclusions about return on investment.
The payroll percentage changes after you select a different developer cost	The dashboard is recalculating the estimate using a different compensation assumption, not showing a change in usage.	Use the same compensation band when comparing different periods or organizations.

Acting on your insights

Use trends in usage, feature adoption, and language activity to guide enablement and rollout planning.

Observation	Possible cause	Suggested action
High adoption in some teams but low in others	Some teams may not have Copilot Chat enabled or configured correctly.	Verify license assignment and IDE setup; offer team-level onboarding.
Steady usage but low agent adoption	Developers may not be aware of Copilot agent features.	Share internal demos or success stories.
Drop in DAU or acceptance rate	Configuration issues or reduced relevance of suggestions.	Encourage feedback and verify IDE and extension versions.