AI Academy · Certificate course
From Enthusiasm to Evidence.
The AI Leaders course.
You've been made responsible for AI. Usually that lands on the person who was already good at everything else.
- Authority through a managed and measured approach
- A board and leadership team that backs the plan
- Value you can prove, and risk you can show is closed
You finish able to walk into your executive team with a mandate, a register, a costed plan and a risk position, and defend every one of them.
Who this is for
You did not apply for this job.
You were handed it. Sometimes formally, in a leadership meeting, with a title that arrived without a budget. More often by accident, because you read the article, or asked the good question, or already had a reputation for sorting things out. Either way, you are now the person other people bring AI questions to, and you are expected to have answers.
Operations, finance and people leaders.
GMs, practice leads, operations managers, finance and people directors, and EAs who became the de facto AI person. You run things for a living. AI is now one of the things you run.
Organisations of 20 to 150 people.
Big enough that scattered adoption is a real risk, too small for a transformation office and a change team. The course is built around a firm of 94, and carries the scale-down for firms under 50, where the mechanism survives and the ceremony does not.
People who have used ChatGPT or Copilot.
That is the whole prerequisite. No code, no prompt engineering curriculum, no model architecture. This course teaches judgement, which is the part nobody has written down for you.
Anyone who has to face a board.
Or an executive team, a partnership, a council panel or a professional indemnity insurer. If someone can ask you to explain and defend how your organisation uses AI, this gets you ready for the question.
We built it around four fears.
Because they are the four we hear every week, from capable people who are doing this on top of a full job.
- Looking naive in front of the board
- Approving something that causes a privacy breach
- Spending money you cannot justify
- Being the person who slowed everything down, or the person who let it run wild
The job is not to be the AI expert. It is to be the person who makes good AI use happen safely, and can show their working.
Who it is not for.
If what you want is to get better at using the tools yourself, this is the wrong course. The Claude at Work course or the Microsoft Copilot course is the right one, and either pairs well with this in either order. If you want model architecture, agent frameworks or prompt engineering, we do not teach any of them here.
Why this course exists
Widespread, and shallow.
Almost every New Zealand organisation has started. Almost none has finished. Individuals are getting real value from these tools and the organisation is getting almost none of it, while carrying all of the risk. That is stage three of four, scattered tool adoption, and it is where most firms of this size are sitting right now.
Source: 2026 NZ AI Roadshow Attendee Report.
Nobody wrote down who decides.
Your MD said "we need to be all over this" at a leadership meeting and meant it. Nothing was written down and nobody was given authority, so four months later you are still asking permission for the things you were supposedly put in charge of.
It is already happening without you.
People are using AI at work today and not telling you, because in most organisations disclosing that you used AI costs you something. Your data is not leaking. It is being handed over, by good people, in the absence of anything better provisioned.
You cannot cost it or prove it.
Hours saved with no destination is not a benefit, and any chair worth having will say so. No baseline, no approval. And without a measured column sitting next to the estimated one, what you have is a slide deck.
How it works
You are not studying a case. You have the job.
From the first lesson you are the AI Lead at Calder Rowe, a New Zealand engineering, planning and environmental consultancy. Newmarket head office, offices in Tauranga and Christchurch, 94 people, trading since 1998. Eleven working shareholder-directors and an independent Chair.
They are not a technology company. Neither, probably, are you.
Their work is document-heavy and it goes to statutory decision-makers. They hold personal information about submitters and landowners, commercially sensitive client material under confidentiality deeds, and data provided by two iwi-owned entities under relationships that took years to build. There is no AI policy, only a paragraph in the 2024 IT policy that nobody has ever complied with.
Every task is done for these people, in order, and the work compounds. The register you build in Section 4 is the register you cost in Section 6 and defend in October.
Swap Calder Rowe for your own organisation as you go, and you finish with a body of work you can take to your actual board.
Who you answer to.
- Diane Verrall, independent ChairShe marks your work. Former listed-company CFO, reads every paper, arrives with three questions in the margin. Her most-used sentence is "says who?"
- Stuart Ngan, Managing DirectorGenuinely excited, slightly embarrassed by how little he understands it, will say yes to anything in a meeting and forget he said it. Your biggest structural risk.
- Bridget Ualesi, Finance and Operations DirectorHolds the budget, the contracts and the credit cards. Unimpressed by imprecision, and your strongest ally the moment you show her a number with a method behind it.
- Tomasz Wieczorek, IT ManagerTwo and a half IT staff for 94 people across three offices. Not the blocker everyone assumes. The person who has been carrying the risk alone.
The curriculum
Seven sections. Seven pieces of work. One board paper.
Thirty seven lessons, each capped at 1,100 words. Seven quizzes that test judgement rather than recall. Seven practical tasks on real artefacts. Twelve sixty-second videos that never teach, they just tell you what has landed on your desk.
The Mandate.
The job you have actually been given. Committee decides, network diffuses. The four-stage maturity model and where you sit on it. The one-paragraph mandate with the words "is authorised to approve" in it, the Sponsor's four written questions, and the sunset clause. Then evidence discipline: what was the sample, when was the fieldwork, who paid for it.
Task 1. Board minutes and an all-leadership email arrive. Draft the mandate and answer the four questions.
The Honest Baseline.
Shadow AI as a provisioning failure. Tool discovery, a five-question anonymous pulse and the governance inventory nobody wants to write. The disclosure penalty, and why a better policy will not fix it. The three things to say out loud about data: not trained on is not not retained, residency is not inference residency, and who else touches it. Then the line between an IT ticket and a notifiable privacy breach.
Task 2. A network egress analysis, nine anonymous pulse comments and a fabricated citation that reached a regional council. Advise the Chair.
The People.
The case you have to make, and the order you build trust in. A shared mental model: confabulation rather than hallucination, fluency is not accuracy, automation bias. Then three frameworks built for this course and used throughout the rest of it.
- The Four GatesWhat data goes in. Is it ours to share, is there a person in it, where is it going, what if it came back out.
- The Delegation GridWhat the machine does. Undo it, or check it. If you can do neither, do it yourself.
- The Three CheckpointsWhere the human stays. Input, judgement, commitment. A signature is not a review.
Task 3. A leadership offsite pre-read and five real tasks. Make the case, run the frameworks, write the all-staff card.
The Register.
Why use cases are the wrong unit and workflows are the right one. The ten-field intake form, and why a third of submissions should die at "alternatives considered". Adapted RICE with a risk multiplier, and exactly where that score can be gamed. Five things the score will not catch. Now, Next, Later and Not doing. Then the Organisational Brain: four knowledge layers, and eight starter files in an hour.
Task 4. Fifteen raw submissions. Tier all fifteen, score three, build the roadmap and name the one that is really a findability problem.
The Guardrails.
"There is no AI law in New Zealand" is a dangerous half-truth: the Privacy Act, the Privacy Commissioner's expectations, the Fair Trading Act and the courts have already arrived. Risk tiers 0 to 3 plus Prohibited, with service levels. Tool tiers A to E, and the features your existing vendors quietly switched on. Guidelines first, policy second. The seven-step incident process, and the step everyone skips. What you tell a client, a panel questionnaire and your PI insurer. And who in your organisation may build an agent.
Task 5. A service provider proposal, a tool request and a council panel questionnaire. Score, recommend and answer.
The Money.
No baseline, no approval. Where the money actually goes, including the unbudgeted time cost that kills programmes. Tokens in plain English, for people who did not want to learn about tokens. Four cost controls that cost nothing: a dollar cap before every pilot, showback, one business unit metric and a weekly rolling number. Then the one page a board actually reads, and the discipline inside the line marked "value measured".
Task 6. A reseller quote, a token invoice and a cost picture with two deliberate errors in it.
Keeping It Alive.
Building the network that carries the mental models. Nominate, do not advertise, and include a sceptic. Recognition as a control rather than a perk. Measuring what actually changed, because licence counts are not adoption. Then the stall: month four to seven, the ceiling and the floor, the look-back and the sunset clause.
Task 7. February 2027 pulse data and a committee transcript. Diagnose the stall and write the recovery plan.
The final assessment.
The Calder Rowe 90 Day AI Leadership Plan. A board paper of 1,200 to 2,000 words in seven required sections, including a Not doing list that cannot be empty, every number labelled actual, estimated or unknown, and an answer to the Chair's last question: what would have to be true for you to recommend that the organisation stops.
Marked against a published rubric. A human assessor makes every final decision, and no result is released without human sign-off.
Plus a reference pack of about twenty thousand words. Templates, checklists, tier tables and verified price tables, moved out of the lessons to hold the word cap. Optional, and not counted in the course time.
Stop asking permission for the thing you were put in charge of.
Start the courseWhat you leave with
Six things that exist when you finish.
Not notes, and not a certificate on its own. Seven tasks, each producing a piece of the same body of work.
A mandate paragraphThe sentence that says what you are authorised to approve, who sponsors it, and when it expires.
An honest baselineWhat is actually happening in your organisation, including the parts nobody told you, with the gaps written down as gaps.
A use case registerTiered, scored and sequenced into Now, Next, Later and Not doing, with a published reason against every rejection.
A tier model and guardrailsRisk tiers with service levels, tool tiers, guidelines people will actually read, and an incident process.
A twelve month cost positionIn New Zealand dollars, with measured and estimated value on separate labelled lines, and the controls that stop it leaking.
A ninety day planNamed owners, real dates, three to five live use cases, and a clear statement of what you are choosing not to do.
Certified on assessed work, not attendance.
Finish every section, pass the seven quizzes and submit the board paper, and AI New Zealand issues the AI Leaders Certificate. The paper is marked against a published rubric with a human assessor making the final call, so the certificate says you can do the work rather than that you watched the videos.
Before you start
What it costs, and what it takes.
What it costs
The group rate is the one worth taking. This works best when a leadership team goes through it together and comes out with the same vocabulary, the same register and the same view of what counts.
No paid AI subscription is required. There is no software to buy to complete this course.
AI Leaders Certificate issued by AI New Zealand after human assessment of a submitted board paper. Vendor-neutral: not affiliated with, or endorsed by, any AI vendor.
How long it takes
Self-paced. Start when you like, stop between sections and come back. Most people run it across three or four weeks.
Good to know
Frequently asked questions.
Do I need to be technical?
No. There is no code in this course and no prompt engineering. If you can read a board paper and form a view, you can do all of it.
Is this the same as the Claude at Work or Microsoft Copilot courses?
No, and they solve different problems. Those two teach you to use a tool well. This one teaches you to lead the organisation's use of all of them, and it is deliberately vendor-neutral. Doing a tool course as well is a good idea, in either order. See the Claude at Work course and the Microsoft Copilot course.
We are 30 people. Is this too much ceremony for us?
The mechanism holds and the ceremony does not, and the course says so explicitly. Section 1 carries the scale-down for firms under 50: what you still need in writing, what collapses into one person, and what you can drop entirely.
Is this an accredited qualification?
No. It is a certificate issued by AI New Zealand, backed by an assessed piece of work marked against a published rubric by a human. We are explicit about what that is and is not worth.
Who marks the final assessment?
A human assessor at AI New Zealand, against the published rubric. We use our own tooling to produce a first draft of the marking report, and a person makes every final decision. No result is released without human sign-off.
What happens if I do not pass?
Quizzes have unlimited retries, because the rationale after each question is the actual teaching. If the board paper does not meet the rubric you get the marked rubric back with the specific gaps named, and you can resubmit.
Do you offer group rates?
Yes. Groups of five or more are $295 per person plus GST. Contact us to arrange it, and ask about running the course as a facilitated programme for your leadership team on your own material rather than the case company's.
Can I share my access with a colleague?
No. It is an individual enrolment. If several people need it, take the group rate, which starts at five.
How current is it?
Every statistic in the course carries its source, its sample and its date, and there is an evidence register behind it. The pricing tables in the reference pack are re-verified before each cohort, because vendor pricing moves faster than course content does.
Weekly newsletter
Keep up with AI, without keeping up with all of it.
Receive our weekly newsletter, keeping 8,500+ business & community leaders connected to the latest Amazing, Interesting and Useful applications of AI for business and society.
- One short email, every Tuesday
- Free to join, no membership needed
- Unsubscribe in one click
The newsletter is sent from the AI Academy. See our privacy policy for how we handle your details.