Rolling out AI to a small team.

Introducing AI tools to a team of 5 to 25 people usually falls to one owner or manager, who handles the plan, the licences and the conversation. If that is you, the target of this lesson is a working habit by day 30: everyone has used AI on real work, two shared workflows are running, and you know what to keep, change or stop. Allow 45 minutes to plan. For the risk side of a rollout, including gates and stop conditions, see the companion lesson on planning a safe AI rollout; this one is about the people.

Why AI rollouts stall after the licences arrive

Most small-team rollouts fail quietly. Licences are bought, a message goes out saying “feel free to use it”, a few enthusiasts do, and everyone else carries on as before. Two things cause most of it: nobody knows which tasks to use it on, and some people are worried about what it means for their job but do not say so.

TRA’s 2025 research for MBIE describes SME attitudes to AI as a mix of fear and fascination, often in the same person, and advises against positioning AI as a replacement for people or as urgent and mandatory (TRA for MBIE). The plan below deals with both causes directly.

The 30-day plan

If AI will change what people actually do, the rules on workplace change apply. Employment New Zealand lists changing an employee’s duties as workplace change, which needs genuine business reasons, consultation and a fair process (Employment New Zealand). Consult before the kickoff, not after.

Exercise: plan your first 30 days

1. Choose two workflows. Use your shortlist from choosing your first AI workflow. Give each an owner and a saved prompt before day 1.
2. Draft the kickoff message. Use the prompt in the worked example. Edit it until every sentence is true for your business.
3. Book the calendar. Kickoff, four weekly practice slots and the day-30 review, all before you announce anything.
4. Make buddy pairs. Pair a confident user with a hesitant one. Rotate after two weeks.
5. Run the weekly practice. Fifteen minutes, standing up if you like. One win and one failure each. Failures go in a shared issues log.
6. Review at day 30. Run a short survey, count who used AI on a real task in the last week, and decide keep, change or stop for each workflow.

Worked example: a Napier commercial cleaning company

Illustrative example: a 14-person commercial cleaning company in Napier, with an owner, an operations manager, an office administrator and supervisors who manage cleaning crews from their phones. The two workflows are drafting roster change messages to cleaners and turning supervisors’ site issue notes into a weekly client summary. No roles are changing. The operations manager drafts the kickoff message:

Draft a short message from the operations manager of a 14-person commercial cleaning
company in Napier, introducing an AI tool to the office team and site supervisors.

Facts to include:
- We are using [approved tool] on our business account from Monday.
- Two uses only to start: drafting roster change messages to cleaners, and turning
  supervisors' site issue notes into the weekly client summary.
- No jobs, hours or roles are changing because of this.
- Everything AI drafts is read and sent by a person. You stay responsible for what you send.
- Do not put cleaners' or clients' personal details into the tool beyond first names.
- 15-minute catch-up every Thursday at 2pm for the next four weeks.

Tone: plain, warm, no hype. Under 180 words. Acknowledge that some people may be
unsure about AI and that questions and doubts are welcome.

Example output (illustrative and abridged):

Kia ora team,

From Monday we're trying an AI tool on two jobs that eat up time: roster change
messages to cleaners, and the weekly site summaries for clients.

To be clear: no jobs, hours or roles are changing because of this. The tool drafts;
you read, fix and send. If it gets something wrong, that's useful to know, so tell us.

Some of you will be keen and some won't be sure. Both are fine. Bring your questions
to our Thursday 2pm catch-up for the next four weeks.

At day 30, the company found every supervisor using the summary workflow, but only one person using the roster messages, because cleaners preferred the old text format. They kept the first workflow, stopped the second, and chose a new one from the issues log. Our guide to roster change communications covers that workflow in more depth.

Where small-team rollouts go wrong

• Licences without workflows. “Use it where it helps” produces little. Two named workflows give people somewhere to start.
• Promises you cannot keep. Only say no roles are changing if that is true and will stay true for the foreseeable future.
• Silencing sceptics. The person who keeps finding errors is doing quality control. Put their findings in the issues log.
• Measuring logins. A login count says nothing about value. Count real tasks and ask what changed, using a short AI adoption survey.
• Customers before staff. If AI use will affect customers, the Privacy Commissioner expects you to tell them how, when and why it is used (OPC). Get the internal habit working first.

After day 30, measuring adoption after training shows how to track use over the following months. Teams that want an outside facilitator for the kickoff and first practice sessions can use our AI training for business teams.

Check your plan

• Two workflows are named, each with an owner and a saved prompt.
• My kickoff message says honestly what is and is not changing.
• I consulted anyone whose duties will change before the kickoff.
• Weekly practice slots and the day-30 review are in calendars.
• Buddy pairs and a shared issues log are set up.
• I know how I will count real use at day 30.

Next lesson: Managing AI mistakes and escalation prepares your team for the first time AI output goes wrong.

Written by the Levy Street team in Wellington. Last updated 19 October 2026.