Onboarding Users at Scale
The onboarding gap
Giving someone access to Claude is not the same as onboarding them. The evidence for this is in every enterprise's usage analytics: organisations that send an invitation email with no supporting material typically see 30–40% of users never activate their accounts, and of those who do, 40–50% send fewer than 10 messages in their first month before going dormant.
The gap is not disinterest — it is uncertainty. Employees do not know where to start, what Claude is for in the context of their work, or what the rules are. Good onboarding closes that gap in the first 14 days.
The five components of effective onboarding
1. A purposeful welcome message
The invitation email from Anthropic is functional but generic. Follow it with an internal email or Slack message that:
- States the purpose: "We have deployed Claude Enterprise to help you [specific use cases]"
- Names the ground rules in plain language: "You may not enter [X]. You should not rely on Claude for [Y]."
- Points to resources: your getting-started guide, the Project(s) they should use first, and who to contact with questions
- Sets an expectation: "Try it for the first time this week — use it for [specific task]"
Template:
Subject: Your Claude Enterprise access is live
Hi [name],
You now have access to Claude Enterprise at claude.ai. We are using Claude to help with [use cases].
To get started: log in at claude.ai using your company email. On your first visit, look for the "[Project name]" Project on the left — that is the best place to start.
A few important rules:
- Do not enter client names, personal data or confidential financials into any conversation.
- Claude can make mistakes — always review outputs before using them.
- If you are unsure whether something is appropriate to enter, ask [contact].
Resources: [link to getting-started guide] | [link to acceptable-use policy]
Questions? Message [contact] or post in #claude-help on Slack.
2. A getting-started guide
A one-page or two-page document (or a brief video) that covers:
- What Claude is (and is not)
- How to access it and log in
- Which Projects to use for which tasks
- 3–5 example prompts relevant to the recipient's role
- The top 3 things not to do
- Where to get help
Tailor this by department if possible. A finance analyst's getting-started guide should have finance examples, not generic ones.
3. First-week activation nudge
If a user has not sent their first message within 7 days of receiving access, send a personal nudge. This does not need to be elaborate:
"Hi [name], just checking in — did you get a chance to try Claude? If you want a hand getting started, [I'm/we're] available for a 15-minute call this week. Here's a quick example to try: [paste a specific, relevant prompt]."
The nudge should come from a human (the AI champion for their team, or the admin), not an automated system. Personal outreach converts far better.
4. Live or recorded training
A 45–60 minute session covering:
- Demo: a live walkthrough of a real use case relevant to the audience's work
- Hands-on time: give participants 15 minutes to try Claude themselves during the session
- Q&A: answer questions about capabilities, limitations and policy
For large organisations, record this session and add it to your internal L&D library. Refresh it every 6 months as Claude's capabilities evolve.
Topics to cover in training:
- Logging in and navigating the interface
- Starting a conversation and refining outputs
- Using and creating Projects
- Prompt basics: how to give clear instructions
- What Claude is good at vs. what it gets wrong
- Data rules: what not to enter
- How to give feedback when something goes wrong
5. An ongoing support channel
Create a Slack channel (e.g., #claude-help or #ai-tools) where employees can ask questions, share useful prompts and report problems. Seed it with content: post a "prompt of the week" or share an interesting use case someone shared with you. Active channels build community and surface problems early.
Measuring onboarding effectiveness
Track these metrics for each onboarding cohort:
| Metric | Target | When to measure |
|---|---|---|
| Invitation acceptance rate | >90% within 14 days | 14 days after invite |
| First-message rate | >80% of accepted invites | 14 days after acceptance |
| Week-4 active rate | >65% | 28 days after first access |
| Avg. messages per active user/week | Set baseline in pilot | Ongoing |
| Self-reported satisfaction (survey) | >3.5/5 | 30 days after first access |
If invitation acceptance is below 90%, the invitation email or spam filtering is the problem. If first-message rate is low, the getting-started guide or welcome message is not compelling. If week-4 active rate drops, the training or use cases are not relevant enough.
Scenario: an L&D team rolling out to 60 finance staff
Rachel manages L&D at a financial services company. She is responsible for onboarding the Finance department (60 people) to Claude Enterprise. She:
- Tailors the welcome email with examples specific to finance: "Use Claude to summarise board reports, check financial models for errors, and draft commentary sections."
- Writes a 2-page finance-specific getting-started guide with 5 example prompts.
- Runs a 50-minute live session: 20 minutes of demo (summarising a publicly available annual report), 20 minutes of hands-on practice with their own documents, 10 minutes Q&A.
- Creates a #finance-claude Slack channel with weekly prompts.
- After 7 days, 8 people have not activated. She sends a personal Slack message to each, offering a 10-minute call.
Four weeks later: 91% activation rate, 78% week-4 active, average 12 messages per active user per week. The CFO asks Rachel to replicate the rollout for Operations.
Key takeaway
Onboarding is the difference between a licence cost and a productivity tool. The investment is small (a welcome email, a guide, a training session, a Slack channel) relative to the return. Do not skip it because it is not technical.
📖 Official Documentation See this in practice in Anthropic’s live support docs: