The Short Answer
To train your team on Claude, do not start with a course. Start with one real workflow per team, train people in groups of three on that work, and do not call anything finished until it runs a second time on new data with a named owner and a date for the next run. Four weeks is enough to get a team from licensed to using it every week, if the four weeks are spent on the right things.
This is the plan we run. It comes from delivering Claude training for teams across finance, sales, operations and engineering, most recently a two-day onsite for a 35-person finance department last week. The tools change every quarter. What decides whether the training sticks has not.
Nobody measures whether a team learned Claude. What gets measured is whether the team is still using it in week six, and almost none of what decides that happens in the training room.
What We Learned Preparing Last Week's Onsite
The finance team we trained last week asked for an agenda. What decided the outcome sat outside it. Three things from the prep changed how we run every engagement:
- The biggest risk was not technical skill. It was logins. The Excel and Outlook add-ins had to be installed and signed in for every attendee before day one, and Microsoft says a centrally deployed add-in can take up to 24 hours to appear for every user. A person who spends the first hour fighting a login is someone you have lost for the day.
- Real work, fake data. We built 16 workbooks for a fictional company whose numbers actually tie out: a real chart of accounts, a real close checklist, and planted errors for people to find. The team practised on their own workflows without a single client record in the room.
- Self-commitment does not survive. Our first draft named eight owners, each of them the person who built the skill. An expert review flagged that as the pattern that decays, so we added a manager sign-off and a dated calendar invite for the next real run, created in the room before anyone left.
Every week of the plan below is built around those three lessons.
Before Day One: Three Decisions That Set the Ceiling
Most of the outcome gets decided here, and most teams skip it.
- Name the approved surface. Decide whether the team is on Claude Team or Enterprise, which connectors are on, and what data is allowed in. If you do not say it out loud, people use whatever is open in a personal browser tab. Our shadow AI breakdown covers why that happens by default.
- Get every account working before the session. Seats assigned, SSO tested, add-ins pushed, one test prompt run per person. It is the most boring item on the list and the one most likely to sink day one.
- Pick one workflow per team. Choose the recurring task with the most hours and a check figure people already trust: a reconciliation, a weekly pipeline summary, a support macro review. Workflows that repeat are the ones where adoption sticks.
Week 1: Train on the Work, in Groups of Three
Skip the feature tour. Put people in groups of three and give each group a written task built on their chosen workflow. Each person has a role:
- The driver is the one at the keyboard, prompting Claude.
- The checker verifies the output against the source: totals tie, dates match, nothing is invented.
- The reader never touches the keyboard. They read the result cold, the way their manager would, and say whether they would sign off on it.
The checker role is where good habits form. Teams that learn to verify in week one keep verifying later. Teams that learn to trust the output in week one end up with a costly mistake in month three. Rotate the roles every block so everyone drives once.
By Friday each group should have one prompt that produces a result they would actually use, saved in a shared Claude Project with the reference files it needs. If you are unsure which surface a workflow belongs on, our guide to Chat, Cowork and Code covers where each one fits.
Week 2: Turn the Best Prompt Into a Skill
A prompt only lives in one person's chat history. A skill belongs to the whole team. According to Anthropic, skills are folders of instructions, scripts and resources that Claude loads when it needs them, and on Team and Enterprise plans an organization Owner can provision a skill so it shows up in every team member's list automatically. That is the step that turns one person's trick into the team's standard way of working.
In week two, each group turns its week-one prompt into a skill and runs it against a checklist. A skill counts as shipped only when it passes every line:
- It runs on two different files, not just the one it was built on.
- The output ties to a number the team already trusts.
- No restricted data went in to make it work.
- The trigger is written down, so a colleague knows when to use it.
- It has a named owner and a manager who has signed off.
- A calendar invite exists for its next real run.
The first line is the one teams resist and the one that matters most. A skill that has only worked on the file it was written against is still a demo. Our open-source skills library has more than 200 examples to copy the structure from.
Week 3: Put It on a Schedule and Get Managers Involved
Recurring work should run on a schedule, not wait for someone to remember it. Anthropic says scheduled tasks in Claude Cowork are available on every paid plan. Each one runs as its own Cowork session, hourly, daily or weekly, and keeps running while your laptop is closed. In week three, the most reliable skill from each team goes on a schedule, and its output lands where the team already works.
This is also the week managers stop sponsoring and start using. Gallup's May 2026 data found that employees whose manager actively supports AI use are 1.7 times as likely to use it a few times a week or more, and only 25 percent of U.S. employees say their organization has communicated a clear plan for it. The cheapest intervention in the whole program is a manager opening the scheduled output in a team meeting and asking a question about it.
Week 4: Measure Use, Not Satisfaction
Post-training surveys measure whether the trainer was pleasant. What you want to know is whether the skills are running. At the end of week four, measure three things: weekly active users against seats, how many shipped skills ran on schedule, and how many fixes owners logged. A skill with zero logged fixes is usually a skill nobody runs.
The baseline makes the gap obvious. Correlation One's July 2026 enablement report found enterprises typically hold two to three times more AI licenses than trained, active users. Decay is the other half. In a study of 150 organizations, Saks and Belcourt found that 62 percent of trainees applied what they learned right away, 44 percent after six months, and 34 percent after a year. They also found that what happens in the workplace before and after training predicted transfer more strongly than what happens during it.
So week four ends with the next 60 days on the calendar: a ranked build-next list, an owner for each item, and check-ins at day 30, 60 and 90. The 60 days after training are where most of the decay happens, so they need an owner, not just good intentions.
What We Deliberately Leave Out
- Agents in week one. A team that has not learned to check Claude's output cannot supervise an agent. We covered the full argument in why Claude training should come before Claude agents.
- Real client data in the room. Synthetic data with a realistic structure teaches the same skills and removes the most common reason legal says no.
- Company-wide kickoffs. We usually start with a leadership cohort of about ten, because a team whose leaders use Claude adopts faster than a team whose leaders only announced it.
Do It Yourself or Bring In a Partner?
You can run this plan in-house. Anthropic's Claude Academy has free courses on Claude, Cowork and agent skills, and they are good. Pair them with the four weeks above and a motivated internal owner, and a small team can get there alone.
The courses cannot build the synthetic data for your workflows, sit with your checkers in week one, or build the automations your team asks for in week five. That is the gap a partner fills. OneWave is an official Anthropic Partner Network member, and we train and build: the build-next list from week four is usually where our engineering work starts. If you are comparing firms, our checklist on how to choose a Claude partner covers which questions to ask. When you are ready to run it, the Claude bootcamp is where it starts.
Sources
- What are Skills? - Claude Help Center
- Schedule recurring tasks in Claude Cowork - Claude Help Center
- Global Indicator: Artificial Intelligence - Gallup (data through May 2026)
- The State of AI Enablement 2026 - Correlation One (July 6, 2026)
- Saks and Belcourt, An Investigation of Training Activities and Transfer of Training in Organizations - Human Resource Management (2006)
- Centralized Deployment of add-ins - Microsoft Learn
- Claude Academy - Anthropic