AI Workshop Formats That Actually Work (and the Ones That Don't)

A lot of "AI workshops" are really conference talks with a Q&A. They get great feedback in the room and change almost nothing back at the desk. After running these for Melbourne organisations across professional services, ops-heavy businesses, and the public sector, a few formats consistently deliver and a few consistently do not. This is the working set.

What an AI workshop is supposed to do

Workshops are an expensive format. Three to ten thousand dollars of time leaves the building for a single session once you count participants. They have to earn it by delivering one of three things:

• A measurable lift in capability — participants can do something on Monday they could not do on Friday.
• A built artefact — a prompt library, a working draft of an automation, a documented process change.
• A team-level decision — a prioritised list of use cases, an agreed acceptable-use position, a workflow redesign.

If your workshop is not aiming for at least one of these, it is a briefing wearing workshop clothes. That is fine — briefings have their place — but price and resource them accordingly. The wider context lives in the AI education for organisations pillar.

Formats that work

The role-specific capability workshop (1 day, 8–14 people)

The workhorse of any decent program. One job family in the room, half a day on technique, half a day applying it to their real work. Output: each participant leaves with at least one prompt or workflow they will use that week, and the team has a shared library.

Works because participants self-reinforce — a sales rep watching another sales rep build a discovery-call summariser learns faster than from any facilitator. Requires a facilitator who has done the job, or near enough to it.

The build clinic (2–3 hours, 4–8 people)

Smaller, more advanced. Participants bring a specific task they want to automate or augment. Facilitator works through one or two of them live, the rest get individual time. Output: working solutions for the ones tackled, a queue and pattern for the rest.

Works as the follow-up to a capability workshop, two and six weeks after. Without the clinic, capability decays fast.

The use-case prioritisation workshop (half day, exec or function leadership)

A leadership-team format. Walk through the function's current workflow map, identify candidate AI augmentations, score against effort/value/risk, walk out with a prioritised quarter's roadmap. Output: a sequenced list, named owners, and the next four use cases to pilot.

Works because the room is the decision-making group — no escalation needed. Fails when run with people who cannot commit time and budget.

The verification lab (90 minutes, any audience)

A short, sharp drill on spotting bad AI outputs. Participants get a stack of outputs containing planted errors — wrong figures, hallucinated citations, subtle bias, misread documents — and have to flag and correct them. Output: a shared library of failure modes and the instinct to apply the verify-first habit.

Works as a literacy companion or a refresher. Worth running annually for any team that uses AI in customer-facing or regulated work.

The board / exec briefing (90 minutes)

Not technically a workshop, but listed here because it is often mis-scoped. Covered in detail in executive AI briefing curriculum.

Formats that look good but waste money

The 200-person all-hands "AI day"

Looks like commitment, behaves like theatre. Big-room formats cannot deliver capability lift because nobody touches their own work. Useful as a kickoff for a real program, useless as the program itself.

The unfacilitated e-learning module

Has its place as a primer or a refresher, but cannot do the verification drill or the applied-task component. If your "training program" is a SCORM module and a quiz, you have a compliance artefact, not capability.

The "innovation day" with no owner

A day off normal work where people experiment with AI, with no curriculum, no facilitation depth, and no follow-through. Produces enthusiasm, a few demos, and zero durable change. If you want hackathons, run them — but call them what they are, fund them properly, and have a path for the good outputs.

The vendor workshop standalone

A vendor running training on their own tool to your team. Useful if it is the third workshop in a sequence and they are teaching deep product features. Damaging if it is the first, because it skips the framework that decides whether you should be using that tool for that task in the first place.

The 30-minute "lunch and learn"

You cannot teach anything load-bearing about AI in thirty minutes minus a lunch line. Use the slot for sharing internal wins, not for teaching.

Facilitation patterns that move the needle

Three patterns we lean on heavily:

Show, don't slide. No more than 15 minutes of slides per hour. The rest is live tool use, participant work, and group debrief. If the facilitator is presenting more than a third of the time, the format is wrong.

Plant real failures. In any capability workshop, include at least one exercise where the AI gets it wrong in a load-bearing way. The participants finding the error is the lesson.

Force an artefact. Every workshop ends with each participant having produced something — a prompt, a draft policy, a workflow diagram — they take back to the team. Without an artefact, the workshop has no anchor.

Schedule the follow-up before the workshop. The two- and six-week clinics need to be in calendars before participants walk in. If you schedule clinics after, attendance halves.

Sample agenda: 1-day role-specific capability workshop

A defensible structure for a function-specific day with 10–12 people:

• 08:30 — arrivals, coffee, a one-page primer.
• 09:00 — framing: what changes after today, what we will build.
• 09:20 — capability tour: live demos of approved tools on tasks from this function.
• 10:30 — break.
• 10:45 — verification drill: spotting failures.
• 12:00 — lunch.
• 12:45 — applied lab block 1: participants tackle their own real task with facilitator support.
• 14:15 — share-back: three or four participants walk the room through what they built.
• 14:45 — break.
• 15:00 — applied lab block 2: pair work on a second task.
• 16:00 — team artefact: consolidate the prompt library and agree on next steps.
• 16:30 — close, with the two-week and six-week clinic dates confirmed.

You can compress this to half a day, but you lose the second applied lab and most of the artefact value.

Where workshops sit in the curriculum

Workshops are the high-cost, high-bandwidth slice of a program. They do not replace literacy, policy, or governance — they sit on top. Most organisations get the most leverage by running fewer, better workshops at the role-specific level, supported by always-on literacy, a clinic cadence, and an internal community of practice. The building an internal AI curriculum post covers how to sequence these.

What to do next

If you are about to commission a workshop, pressure-test it against three questions: who is in the room, what artefact will they walk out with, and what is the follow-up cadence. If any of those is fuzzy, fix it before booking the room. If all three are clear, you will be ahead of most programs we audit.