Most small businesses roll out AI the same way: someone buys a few licenses, sends a link, and a month later three people use it and everyone else forgot. The tools are not the problem. The rollout is.
Here is the four-week plan we use with clients. It works with ChatGPT, Claude, Microsoft Copilot or Google Gemini, because the hard part is the same in all of them.
Week 1: Pick the Work, Not the Tool
Start with a short list of repetitive writing and reading tasks that eat real time. Good first candidates:
- Drafting job posts, offer letters and routine emails
- Summarizing long documents, meeting notes or survey results
- Turning rough notes into a policy, SOP or training outline
- Answering the same internal questions over and over
Ask each team lead for the three tasks they would happily never do again. Estimate the hours each one takes in a typical week. That number becomes your baseline.
Week 2: Set the Rules and the Tools
Before anyone pastes in company information, decide:
- Which tool and plan. Use a business or team plan with data controls, not personal accounts. Business plans typically let you keep your data out of model training and manage who has access.
- What never goes in. Social Security numbers, medical information, payroll files, client confidential data, unless the tool is approved for it.
- Who reviews output. AI drafts, people decide. Anything sent to a customer, employee or regulator gets a human review.
Put this in a one-page AI use policy. Our AI use policy guide covers what to include.
Week 3: Train on Real Work
Generic AI training does not stick. Run short, hands-on sessions where each person brings a real task from their own job and leaves with a working prompt or a custom assistant built for it.
Three habits make the biggest difference:
- Give context. Who the audience is, what good looks like, and an example.
- Iterate. Treat the first draft as a starting point and ask for specific changes.
- Save what works. Build a shared library of prompts and custom assistants for repeat tasks.
Week 4: Measure and Expand
Go back to the baseline from week one. For each task, ask how long it takes now. Collect a few concrete examples, like the job post that took 10 minutes instead of an hour.
Then decide what is next: more tasks, more people, or connecting AI to the systems your team already uses.
What Usually Goes Wrong
- No owner. Someone needs to own the rollout, the prompt library and the policy.
- No rules. Without a policy, cautious people avoid the tools and careless people overshare.
- Training on features instead of tasks. People adopt what saves them time on Monday morning.
- Trusting output blindly. AI makes confident mistakes. Review is part of the process.
Where King & Co. Helps
We run hands-on AI training built around your team's actual work, on site or online, and help set up the policy and shared prompt library that keeps it going. Workshops start at $750.
Book a free 15-minute call and tell us which tasks eat the most time.
Josh runs HR, payroll and PBJ support for employers across New England and New York, from five-person offices to multi-building nursing home groups. He works in every major payroll system and answers the phone personally.
Book 15 minutes with Josh