Training an OC Respite Agency's Team to Effectively Use Claude Cowork

Low AI adoption inside a business is usually mistaken for a tooling problem. It is almost always a training and workflow problem.

A staffing agency came to us with AI licenses already purchased and almost nobody using them. Recruiters were still hand-writing job descriptions, skimming resumes one at a time, and rebuilding the same client updates every week. The software was sitting there. The work had not changed.

Why the Seats Went Unused

Nothing had connected the tool to the actual tasks. A vendor demo shows what a product can do in the abstract; it does not tell a recruiter what to do differently on the requisition open in front of them at 9am. Without that translation step, people fall back to the method they already trust, which is the manual one.

There was a second, quieter blocker: nobody knew what was allowed. Candidate records are full of personal information, and in the absence of a written rule, a careful recruiter's safest move is to not use the tool at all. Ambiguity reads as risk.

What We Did

We trained the team on Claude Cowork against their own live workflows, not generic demos. Three things carried the engagement:

• Role-specific sessions for recruiters, account managers, and back office, because those three groups do genuinely different work and needed different examples.
• Shared, reusable prompts for intake notes, candidate summaries, and client updates, so a better approach discovered by one person became available to everyone.
• Clear rules on what candidate data may go into a chatbot, written down before rollout rather than after an incident.

Training on live work is the part that makes adoption stick. When someone practices on a real requisition they are already responsible for, they finish the session with work completed and a habit started, rather than notes they will not revisit.

The prompt library does the long-term work. It turns individual skill into a team asset and gives new hires a starting point on day one.

Writing the Rules First

Putting the data rules in place before rollout is not a compliance formality, it is what lets people move quickly. A recruiter who knows exactly which fields stay out of a chatbot does not have to stop and deliberate every time; they have an answer.

This is the same reasoning behind writing an AI usage policy for a small business, and it is why we treat policy and training as one engagement rather than two. If you are weighing how much of your data is safe to put into a general-purpose assistant, is ChatGPT safe for business data covers the underlying question.

The Result

The team uses AI on real requisitions every day, with PII handling agreed on up front. The licenses stopped being a line item and became part of how the agency runs.

What to Take From This

If you have paid for AI seats and cannot point to daily use, more software will not fix it. What closes the gap is role-specific training on your own workflows, a shared prompt library, and written data rules people can follow without asking permission.

That is the substance of our employee training and managed AI work. If you would rather start by finding out where the time is actually going, a SafeStart Audit maps your current tools and workflows first, and you keep the policy and roadmap either way. For the build-versus-hire math behind bringing this in-house, see AI consultant vs. hiring an in-house AI engineer.