Copilot Metrics

Load data

Drop the file(s) here, or click to browse

Specific Metrics
Metrics Summary
Functionality
IDE / Language

Total Active Users Per Day

i

Total Engaged Users Per Day

i

Total Chats and Chat Events Per Day

i

Total Code Completions Per Day

i

Lines Not Accepted vs. Lines Accepted

i

Percentage of Lines Accepted

i

Total Code Suggestions Per Day

i

Total Code Acceptances Per Day

i

Percentage of Code Suggestions Accepted

i

Summary Charts

Total Active vs Engaged Users Per Day

i

Percentage Engaged Users Per Day

i

Copilot Functionality Metrics

Total Engaged Users for Copilot IDE Chat

i

Total PRs Reviewed Per Day

i
The number of pull requests that were reviewed with the help of Copilot. This is a count of PR reviews, not unique users.

Total PR Summaries Created Per Day

i

IDE Charts

IDE's Used Per Day

i

Lines Suggested Per Language

i

Percentage Lines Accepted Per Language

i

User Analytics

Top 20 Users by Lines Accepted

i
The 20 users who accepted the most lines of code from Copilot completions. Bar colour indicates tier: purple = Power Users, green = Regular Users, amber = Light Users.

Lines Accepted Concentration (Pareto)

i
Shows what percentage of all accepted lines are contributed by the top N users (sorted by individual lines accepted). The 80% mark reveals how broadly — or narrowly — Copilot value is distributed.

Tier Distribution — Code Completions

i
Users grouped into Power / Regular / Light Users using k-means clustering on inline code-completion activity — this finds natural breaks in usage instead of forcing a fixed percentage into each tier.

Tier Distribution — Chat Usage

i
Users grouped into Power / Regular / Light Users using k-means clustering on chat activity (chat panel modes + inline chat interactions) — this finds natural breaks in usage instead of forcing a fixed percentage into each tier.

Tier Distribution — Combined

i
Users grouped into Power / Regular / Light Users using k-means clustering on total activity across all features (code completions + chat + agent edits combined) — this finds natural breaks in usage instead of forcing a fixed percentage into each tier. This combined tier is what drives the "Top 20 Users", "Lines Accepted Concentration" and "Feature Adoption by User Tier" charts.

Feature Adoption by User Tier

i
Percentage of users in each tier who used each feature at least once during the period. Shows which advanced features are adopted by power users versus lighter users. Tiers here use the combined activity classification, detected via k-means clustering on total activity across code completions, chat, and agent edits: Power Users are the highest-activity cluster, Light Users the lowest, and Regular Users the middle cluster. Tier sizes vary based on the actual usage distribution rather than a fixed percentage.

Features & Models

Daily Users by Feature Flag

i
Number of unique users per day who used Chat (any chat panel or inline chat), Agent mode (Copilot agent mode), CLI, or the Copilot App. Each point is a distinct user count for that day — a user is counted once per day regardless of how many interactions they had.

Feature Type Distribution

i
Overall split of all Copilot activity across feature types: Code Completion (inline suggestions), Chat (all chat panel modes + inline chat), Agent Edit (background file edits), Copilot CLI, and the Copilot App.

Chat Mode Breakdown

i
How chat usage is distributed across the different Copilot chat modes: Agent Mode, Ask Mode, Edit Mode, Plan Mode, Custom Mode, and Inline Chat.

Feature Usage Over Time

i
Daily breakdown of Copilot activity by feature type. All chat panel modes are grouped as "Chat". Shows how feature usage evolves over the reporting period.

Copilot App Usage

i
Daily sessions, requests, and prompts sent through the Copilot App, based on the totals_by_copilot_app data. Only populated once users start using the Copilot App.

AI Model Usage

i
Which AI models are being used for code generation activities, based on the totals_by_model_feature data. Shows the top models by number of code generation events.