Building an Internal AI Curriculum: A Step-by-Step Operating Guide

The shift from "we ran some AI training" to "we operate an AI curriculum" is the difference between a budget line that disappears and a capability that compounds. A curriculum is not a list of courses — it is an operating asset, with owners, refresh cycles, intake processes, and measurement. Here is how to build one that survives the next tool cycle and actually changes how your organisation works.

Start with the audience map, not the content

The most common failure mode is starting with content design — "let's build a great prompting module" — before agreeing who the curriculum is for. Build the audience map first.

A workable structure for most mid-market Australian organisations:

• All staff — literacy and acceptable use.
• People leaders — managing AI-augmented work, plus the all-staff layer.
• High-leverage role families — typically marketing, sales, customer support, operations, finance, legal/risk, HR. Each gets a role-specific capability path.
• Builders and power users — technical workshop track, evaluation, and internal community of practice.
• Executives and board — strategic and governance layer.
• New joiners — a 30-day onboarding sequence that pulls from the above.

For each group, agree the outcome: what should change in how they work after they complete this path. If you cannot answer that, the path is not ready to build. The cluster pillar on AI education for organisations walks through the coverage map in more detail.

Define the learning path per audience

A learning path is a sequenced set of modules with explicit prerequisites. For a customer support team member, a defensible path looks like:

1. All-staff literacy (90 min).
2. Acceptable use and data rules walkthrough (45 min).
3. Customer support capability workshop (1 day).
4. Two-week clinic.
5. Six-week clinic.
6. Quarterly refresh and incident review.

For a marketing director:

1. All-staff literacy.
2. People-leader module.
3. Marketing capability workshop.
4. A use-case prioritisation workshop with their team.
5. Quarterly leader briefing.

The path is the artefact. Write it down per role, publish it, and reference it in onboarding. When someone asks "what training do I need", you point at the path, not at a catalogue.

Pick your content sourcing model

Three viable sources for content:

Internally authored. Highest stickiness, highest cost. Right for anything tied to your data classifications, policies, tools, and use cases.

Externally authored, internally facilitated. A partner provides the curriculum and materials; your people deliver. Right for the general literacy and verification layers.

Externally authored and delivered. A partner runs the workshop end-to-end. Right for executive briefings, deep capability workshops, and where credibility matters more than ownership.

Most programs benefit from a deliberate split — roughly 60–70% internal, 30–40% external, refreshed annually. Pure-internal programs go stale fast; pure-external programs do not embed.

Stand up the operating rhythm

A curriculum needs an operating rhythm to stay alive. The minimum cadence:

• Weekly — facilitator office hours or a Slack/Teams channel staffed by a real human.
• Monthly — community of practice session, run by internal builders, with a rotating focus.
• Quarterly — curriculum review: what is being used, what is decaying, what needs adding. Also: incident review, where any AI-related incidents in the period become teaching material.
• Twice yearly — executive briefing refresh.
• Annually — full curriculum audit against the audience map and the regulatory landscape.

Most curricula that fail did not fail in design — they failed because nobody owned the quarterly review and the content quietly drifted from reality.

Tooling, tracking, and the boring bits

You do not need a fancy LMS. You do need:

• A single canonical location for the audience map and learning paths.
• A way to record completion against the path (your existing LMS or HRIS is usually fine).
• A simple capability assessment — short, role-relevant, repeatable.
• A use register linking who completed what to which AI systems they are now authorised to use.

The use register is the underrated artefact. It is what you produce when an auditor or a regulator asks how you ensure staff are trained appropriate to the systems they use — see AI safety and responsibility training for how this ties into the Voluntary AI Safety Standard.

Sample roadmap: zero to mature in 12 months

A realistic build sequence for a 300-person organisation starting from scratch:

Months 0–2. Audience map and learning paths drafted. Executive briefing delivered. All-staff literacy module built and piloted with one team. Acceptable use policy in place.

Months 2–4. Literacy rolled out organisation-wide. First role-specific capability workshop run for the highest-leverage function (often marketing or support). Clinic cadence established.

Months 4–7. Two more role-specific tracks stood up. People-leader module built. Community of practice launched with an internal owner. First quarterly review completed.

Months 7–10. New-joiner onboarding sequence integrated with HRIS. Use register populated. First capability assessments run.

Months 10–12. Annual audit. Curriculum 2.0 plan agreed. External partner footprint resized based on what is now internally sustainable.

By month twelve, the curriculum is no longer a project — it is a function. That is the goal.

What to budget

For a 300-person organisation, a realistic year-one investment is AUD 60k–150k, depending on how much you build internally versus buy in. The split is roughly: 40% curriculum design, 30% delivery, 20% executive layer, 10% tooling and admin. See AI training program ROI for how to think about return on that spend.

In year two, costs drop materially as you stop building from scratch and shift to refresh and delivery. Most organisations land at 50–60% of year-one spend in steady state.

Governance and accountability

A curriculum needs a named owner — usually a head of L&D or capability — and a steering group that includes the AI risk owner, a technology lead, and at least one operating function head. Meet monthly for the first six months, then quarterly. Without this, content drifts and gaps reopen.

What to do next

If you do not yet have a written audience map and a single role-specific learning path, that is the next thing to build. Everything else compounds from there. If you have those but no operating rhythm, fix the rhythm before adding content — you have enough material already; you are losing the value of it through neglect.