Roll AI out to a team of 50 in phases, not on a single launch day. Write short ground rules first, start with a small pilot group, train people on their own real tasks, appoint a few champions in each function, and keep a simple feedback loop running. Measure whether people use AI in their daily work and whether the work improves, rather than counting how many logged in once. This article is general guidance; adapt it to your own tools, data and sector.
Why a phased rollout beats a big launch
Fifty people is large enough that informal habits will form quickly and small enough that you can still manage the change deliberately. A single announcement followed by a tool licence usually produces two groups: a few enthusiasts who use AI constantly, and a majority who tried it once and stopped. A phased rollout lets you learn what works in your business before everyone is involved, and fix problems while they are still cheap.
Phase 0: set the ground rules before anyone starts
Before opening access, decide the basics:
- Which tools are approved and which account type staff should use.
- What data may be entered, including customer and employee information. This touches Malaysia's Personal Data Protection Act, so read our guide to AI governance and the PDPA and confirm the details with your legal or compliance team. This is not legal advice.
- Who is accountable for output. A person remains responsible for anything they send or decide using AI.
- Where to ask questions and how to report a problem.
One to three plain pages is enough. You can refine it as you learn, but people should never be guessing on day one.
Phase 1: run a pilot with a small group
Pick roughly five to ten people from different functions who are curious and respected by their peers. Give them a clear, narrow set of tasks to try, such as drafting, summarising, analysing or preparing reports. Ask them to note what saved time, what produced poor results and what they wished they had known. Our guide on how to run an AI pilot covers the steps in more detail.
The pilot has two jobs: to prove value on real work, and to produce the examples, prompts and warnings that make the wider rollout easier.
Phase 2: appoint champions in each function
Champions are not a new department. They are people in sales, finance, operations, HR or any other team who are a little ahead of their colleagues and willing to help. Ask pilot participants to take the role. Their job is to answer everyday questions, share useful prompts and workflows, and pass problems back to whoever owns the rollout. For 50 people, a handful of champions spread across teams is usually enough. Give them some protected time, otherwise the role becomes a favour on top of a full workload.
Phase 3: train in stages, on real work
Generic training fades fast. Train people on tasks from their own jobs and let them practise during the session. A sensible structure is:
- Everyone: the ground rules, how to write a clear request, and how to check output before using it. See how to write a good prompt for Claude and how to check if AI output is accurate.
- By role: workflows specific to each function, built from the pilot's examples.
- Champions and power users: deeper skills, such as building reusable workflows. Some may want a recognised credential; Agmo Studio, a Select Partner of the Anthropic Claude Partner Network, runs exam-preparation workshops for the Claude Certified Associate and Claude Certified Architect certifications, which are issued by Anthropic.
If you would rather have a facilitator run sessions tailored to your team, Agmo Academy offers in-house training quoted per scope, and our AI training programs page lists the public options. HRD Corp claimability is subject to eligibility and HRD Corp approval.
Bring managers along
Teams follow what their manager does. If managers never use AI, or quietly discourage it, adoption stalls however good the training. Brief managers first, show them what the pilot found, and agree what good use looks like in their team. Senior leaders setting direction may benefit from a more strategic session, such as the AI Forward CXO Program.
Build a feedback loop
Create one place, such as a shared channel or a short monthly check-in, where people post what worked, what failed and what they need. Review it regularly. Look for repeated questions (a sign the training missed something), repeated errors (a sign a rule or prompt needs fixing) and tasks that clearly benefit (candidates for a shared template). Update the ground rules as tools and use cases change.
Measure adoption and results
Choose a few measures before you start, and keep them simple:
- Use: how many people use AI on real work each week, not just who has access.
- Quality: whether output is accepted with fewer corrections.
- Time or cost: estimated time saved on specific, named tasks.
- Confidence: a short staff survey on whether people know what they may and may not do.
Record a baseline from the pilot so that you have something to compare against. Our guide to measuring AI training ROI goes deeper on this.
Common mistakes to avoid
- Buying licences for everyone before knowing which roles will benefit.
- Skipping the ground rules and hoping people will be careful.
- Training once and never following up.
- Treating enthusiasm as a metric. Usage on real tasks is what counts.
- Ignoring the sceptics. Their objections often point to real risks.
A simple timeline to adapt
Treat this as a template, not a promise of how long your organisation will take: ground rules first, then a pilot, then champions and role-based training, then wider access with a feedback loop, then a review of results before you decide what to extend. Give each step enough time to learn from it before moving on.