I wanted to know which AI skills my team actually uses. Not the ones we built, the ones we run.
So I counted mine. In Cowork, I'd called a skill eight times since July. That's a much shorter list than the one I'd built, and it's the kind of thing you only learn by looking.
If your company is on Claude Team or Enterprise, you don't need to build anything to start looking. Claude already has an analytics dashboard that shows who's using it, which products they use, which skills and connectors get run, and what it all costs. Most people have never opened it. This guide shows you how to get in and what's worth your time once you're there.
Who can open it
| Plan | Who can see the dashboard |
|---|---|
| Team | Owners and Primary Owners |
| Enterprise | Owners, Primary Owners and Admins (Admins see everything except spend) |
| Pro or Max | No org dashboard. These are individual plans, so there's no admin view |
If you're a regular member on Team or Enterprise, the page won't load for you. Ask whoever set up your Claude account (usually the person who pays for it) to either show you or give you the right role.
How to get in
- Go to claude.ai and sign in with your work account.
- Click your initials in the bottom-left corner.
- Click Analytics.
Or skip the menu and go straight to claude.ai/analytics/overview.
Each product has its own tab with a direct link, which is handy for bookmarking:
| Tab | Link |
|---|---|
| Overview | claude.ai/analytics/overview |
| Chat | claude.ai/analytics/usage |
| Claude Code | claude.ai/analytics/claude-code |
| Cowork | claude.ai/analytics/cowork |
| Spend export | claude.ai/analytics/activity |
What each section shows
The overview page is built around a few plain questions.
| Section | What it tells you |
|---|---|
| Who's using Claude? | Active members against the seats you're paying for, by product and by group |
| How are they using Claude? | Adoption and stickiness, which skills get run (uses and cost per use), which connectors get used (reads and writes), and a beta measure of how agentic the work is |
| What are the results? | Conversations, files worked on, pull requests, designs created, and an estimated time saved |
| How much is Claude costing? | Spend against limits, month and year to date, spend by model, and how concentrated spend is across people |
Then each product tab goes deeper:
- Chat: chats per day, the share of people chatting, project activity and artifacts created.
- Claude Code: productivity (pull requests, lines accepted, a leaderboard), usage (top skills, top MCP servers, tool usage) and value (cost per commit, pull request and session).
- Cowork: sessions per day and daily, weekly and monthly active users.
A note on skills. The skills and connectors panels are listed for Team and Enterprise in Anthropic's help docs, but the per-person and per-group breakdowns were launched for Enterprise. If you're on Team and don't see a skills panel, that's probably why.
The three things worth looking at first
You could spend an hour in here. Start with these three.
1. Active members against seats. If you're paying for 20 seats and 8 people are active each week, that's your first conversation. It's rarely a tool problem. Usually nobody showed those 12 people what to use it for.
2. Which skills actually get run. This is the honest list. The skills you spent a week building and nobody runs are worth fixing or deleting. The one everyone runs is worth making better, because it's already proven.
3. Spend concentration. If two people account for most of the spend, they're either your power users or your problem. Either way, you want to know which.
Getting the data out
The dashboard is good for a glance. For anything you want to sort or share, export it.
Spend report (CSV):
- Open claude.ai/analytics/activity.
- Scroll to How much is Claude costing?
- Click Export spend report.
- Pick month to date, last month, last 90 days, or a custom range (up to 90 days back).
- Download.
You get one row per person per model, with their email, the product (Chat, Claude Code, Cowork and so on), requests, tokens, and spend before and after discounts. Drop it into a spreadsheet and you can see exactly who's using what.
Claude Code leaderboard: on the Claude Code tab, use Export all users on the leaderboard. This needs contribution metrics turned on (see below).
What the numbers don't tell you
- It runs a day behind. Spend updates daily with a one-day delay, and Claude Code updates daily.
- Code metrics need GitHub. Pull requests and lines of code only show up after a GitHub admin installs the Claude GitHub app and an Owner switches on GitHub analytics in the Claude Code admin settings. Usage numbers work without it.
- The code numbers are conservative on purpose. Anthropic only counts lines and pull requests it's confident Claude helped with.
- Usage isn't value. The dashboard tells you someone ran a skill. It doesn't tell you whether the output was any good or the work got done faster. "Estimated time saved" is an estimate, so treat it like one.
Going further
If the dashboard isn't enough, there are two next steps. Both are more technical.
- Enterprise Analytics API. Enterprise only. Your Primary Owner creates an analytics key in Organization settings, and you can pull usage by skill, connector, plugin and person into your own reporting.
- Claude Code telemetry. Claude Code can send its own usage data, including which skills were run and by whom, to a monitoring tool. It works on any plan and is set up through managed settings. Anthropic's Claude Code monitoring docs walk through it.
Where the DIY version stops
The dashboard answers "who's using Claude?" That's a real step forward. Most companies paying for AI can't answer it.
The harder question is "is it working?" That means knowing which processes AI should be doing, building it into those processes properly, and measuring the result against how the work was done before. A dashboard can't tell you that, because it doesn't know what the work was supposed to be.
That's what we do at Whitehorse. We map how the business actually runs, build AI into the parts that matter, train the team, and measure the result. If you've opened the dashboard and want to know what the numbers should look like, reply to the welcome email with audit and I'll tell you how we'd start.
Start here
- Open claude.ai/analytics/overview.
- Check active members against seats.
- Look at which skills actually get run.
That's ten minutes, and you'll know more about how your team uses AI than most companies do.
Built by Alex Sidhu, Whitehorse AI.