Your team already uses AI. The question is whether they use it well, safely and consistently, or whether three enthusiasts run ahead while everyone else quietly pastes client data into free chatbots.
That second scenario is what we see most often when we start working with a business. Someone bought licences, sent a Slack message, and hoped. Six months later usage has collapsed to a handful of people, nobody can say what the tools have changed, and shadow AI is everywhere.
Training fixes this. But most AI training doesn't work, so it's worth being clear about why before you spend money on it.
Why Most AI Training Fails
Three failure patterns come up again and again.
Tool rollout without training. The business buys Copilot or ChatGPT licences and treats access as adoption. It isn't. People try the tool twice, get a mediocre result because nobody taught them how to prompt, and go back to how they worked before. The licences keep billing.
Generic courses. Online courses teach AI concepts using someone else's examples. Your quantity surveyor doesn't need to summarise a Wikipedia article. They need to turn site notes into a client-ready report. When the exercises don't match the work, the skills don't transfer, and by Friday the course is forgotten.
No follow-through. A one-off lunch-and-learn creates a spike of interest and no lasting change. Capability builds through repetition on real tasks over weeks, not a single session.
The common thread: training fails when it's disconnected from the actual work. That's the thing to fix.
What Works: Six Principles
We've trained teams across recruitment, deep tech R&D, civil contracting and professional services. The programmes that stick share the same structure.
1. Train on your own workflows. Every exercise should use a real task from the participant's week. A recruitment consultant practises on a real job brief. A field engineer practises on real site data. Skills learned on your own work get used on Monday because they were built on Monday's work.
2. Start with fundamentals, not features. People need a working mental model of what AI can and can't do before tool-specific training makes sense. Teams that skip this stay stuck at surface level and can't adapt when the tools change. Which they will.
3. Make prompt engineering practical. Prompting is the core skill and it's learnable in hours, not weeks. The gap between a vague prompt and a structured one is the gap between a 30-minute rewrite and a usable first draft.
4. Deal with governance early. If you don't give people a sanctioned way to use AI, they'll use an unsanctioned one. Cover data handling, what can and can't go into which tools, and responsible-use practices as part of the training itself. One recruitment firm we trained eliminated shadow AI use entirely this way, because the sanctioned path was also the easier path.
5. Measure against a baseline. Assess capability before training starts, then again at the end. Without a baseline you're guessing about whether anything changed. With one, you can show measurable competency improvement and identify who needs more support.
6. Build internal champions. Someone on the team should leave the training equipped to support everyone else. Peer support keeps capability growing after the trainer leaves.
A Programme Structure That Works
For a business of 20 to 200 people, a workable structure looks like this:
- Baseline assessment. Where is each participant now? What do they use, what do they avoid, what worries them?
- Fundamentals session. What AI is, what it isn't, where it fits your industry. Live demonstrations on your own use cases.
- Hands-on prompt engineering. Participants work on their own tasks with structured techniques and feedback.
- Governance and responsible use. Your rules, your tools, your data boundaries. Made practical, not preachy.
- Applied sprints. Two to four weeks of applying the skills to real work, with support available when people get stuck.
- Capability handover. Final assessment against the baseline, a reference playbook the team keeps and champions identified for ongoing peer support.
Total time investment: somewhere between 6 and 15 hours per participant depending on depth. That's the difference between a team that uses AI and a team that has licences.
You can run a version of this yourself if someone in the business has the AI depth and the time to design it. Many businesses don't, which is where structured programmes come in.
What Trained Teams Actually Do Differently
The test of training is what the team does independently after it ends. Some results from teams we've trained:
A 22-person recruitment team in Auckland now completes meeting notes in 2 to 3 minutes, down from anywhere up to 30. Pre-meeting research is automated. Every consultant uses AI daily, and shadow AI use is gone.
An IT recruitment firm cut job ad drafting from 30 minutes to about 2 after we trained the team on an AI tool built around their brand voice. Seek NZ independently called the output the best AI-generated job ads they'd seen.
A civil contractor's field team processes site data 80% faster, and new team members are productive from day one instead of shadowing someone for weeks. Projected annual saving: $75,000.
A deep tech R&D team increased publication review productivity more than fivefold and found a 25% patent renewal cost reduction using AI-assisted analysis.
None of that came from licences alone. It came from structured training applied to each team's own work.
Where to Start
Start by understanding where your team actually is. If you're not sure, a free AI readiness assessment will show you where the gaps are across people, process, knowledge and technology in a few minutes.
If the people dimension is your weak point, that's your answer: train before you buy more tools. Our AI training programmes run 6 to 15 hours, in person or online across New Zealand and Australia, and they're built on the principles above. Your team keeps the skills, the playbook and the independence.
The goal isn't to replace anyone. It's one person producing the output of three, and a team that manages its own AI tools without waiting on external support. That's a training outcome, not a purchasing one.
