Why This Matters

Usage data drives cost control, capacity planning, and ROI measurement. The Metrics API (GA Feb 2025) gives reliable programmatic access.

Concept Explained

/copilot/usage endpoints return: per-user credit usage, per-feature usage (completions, chat, agent), per-model usage. Historical data available up to 1 year.

How It Works

GET /orgs/{org}/copilot/usage with date range. Returns per-user and aggregate metrics. Build dashboard; alert on over-usage.

Step-by-Step Tutorial

1. Pull usage

2. Per-user metrics

3. Aggregate

4. Build dashboard

5. Alert

Real-World Example

A team built a Copilot usage dashboard. Found one user burning 5x the average — using the coding agent for trivial tasks. Coached the user; saved $200/month. Dashboard became weekly leadership review.

Example Prompts / Commands / Code

import requests
from datetime import datetime, timedelta

ORG = 'your-org'
TOKEN = 'YOUR_GITHUB_TOKEN'

since = (datetime.now() - timedelta(days=30)).strftime('%Y-%m-%d')
until = datetime.now().strftime('%Y-%m-%d')

r = requests.get(
    f'https://api.github.com/orgs/{ORG}/copilot/usage',
    headers={'Authorization': f'Bearer {TOKEN}'},
    params={'since': since, 'until': until}
)
usage = r.json()

# Per-user breakdown:
for user in usage['users']:
    print(f'{user["login"]}:')
    print(f'  Total credits: {user["total_credits"]}')
    print(f'  Completions:   {user["completions_count"]}')
    print(f'  Chat:          {user["chat_count"]}')
    print(f'  Agent:         {user["agent_count"]}')

Track:
- Per user: total credits, feature breakdown, trend over time
- Per team: aggregate usage, top users, anomalies
- Per model: which models used, cost per model
- Per feature: completions vs chat vs agent
- Anomalies: user >2x avg, team >120% of budget

Alert thresholds:
- User >1.5x avg: notify manager
- Team >110% of budget: notify lead
- Org >90% of total credits: notify admin

Common Mistakes

• Not pulling regularly — data only available for 1 year.
• Not alerting on anomalies — surprise at month-end.
• Tracking but not acting — data without action is wasted.
• Forgetting historical data limit — 1 year max.

Best Practices

• Pull weekly; build dashboard.
• Alert on anomalies (>1.5x avg, >110% budget).
• Track per-user, per-team, per-model, per-feature.
• Act on data: coach over-users, right-size plan.
• Remember: historical data only 1 year.

Troubleshooting

Practical Exercise

Professional Challenge

Frequently Asked Questions

Further Reading

• Copilot usage metrics

Official References

• Copilot usage metrics

Related lessons: PR-07, PR-10