AI Training: The 90-Day Plan

Key takeaways

• ✓Ninety days is enough time to move from a vague mandate to a delivered first cohort, but only if the first thirty days are spent on scoping and stakeholder alignment rather than content development.

• ✓The biggest implementation risk is not budget or logistics. It is starting to build training before you understand what behaviour change you actually need.

• ✓A phased approach lets you validate your design assumptions early, so you are not discovering problems on delivery day.

• ✓Baseline your organisation's AI literacy before you set content scope. Skipping this step is the most common reason programs land flat.

• ✓Ninety days is a sprint, not a shortcut. The plan works because it enforces decisions at each phase gate, not because it compresses the thinking.

Why 90 days is the right horizon

Most AI training initiatives stall before the first session is booked. Not because the idea is wrong, but because the gap between "we should do something" and "here is a structured program" feels too wide to cross without a clear planning frame.

Ninety days is short enough to maintain momentum and long enough to do the job properly. You can complete a needs assessment, secure internal sign-off, design or procure a program, and run a first cohort inside that window. Try to compress it to four weeks and you skip the diagnostic work that determines whether training actually lands. Stretch it past six months and the initiative loses organisational air cover, budgets shift, and the urgency that sparked the conversation quietly disappears.

There's also a practical rhythm to it. Ninety days maps cleanly onto a single quarter, which is how most Australian enterprises plan, budget, and report. That alignment matters more than it might seem. A proposal that fits within a single quarter's L&D budget and shows measurable output before quarter-end is a much easier conversation with a CFO or People and Culture executive than one that spans two planning cycles.

The 90-day frame is a discipline, not a deadline

The goal is not to rush training out the door. It is to force the decisions that L&D teams typically defer: who the audience is, what behaviour change looks like, and how success will be measured. Without a timeline, those questions stay open indefinitely.

One important caveat: 90 days to first cohort is not 90 days to finished program. If your organisation needs AI fluency across several business units, you are looking at a rolling series of cohorts, each building on what the last one taught you. The first cohort is a test as much as it is a delivery. Getting there quickly means you start learning from real participants sooner, rather than spending six months designing something in isolation and discovering the gaps on day one of delivery.

If your organisation hasn't yet mapped where AI capability gaps actually sit, the AI readiness assessment work described elsewhere in this series belongs in the first two weeks of the plan.

What does Phase 1 (Days 1-30) actually cover?

The first 30 days are entirely about scoping. No content is built, no sessions are booked. Before any of that makes sense, you need to know who you are training, why, and what success actually looks like.

Get stakeholder alignment first

This sounds obvious and consistently gets skipped. If the L&D lead, the relevant business unit heads, and whoever controls the budget are not aligned on what the training is meant to achieve, you will spend months building something that satisfies nobody.

Schedule a short working session, not a steering committee, with the people who own the outcome. A finance transformation lead who wants their team using Copilot to cut invoice approval time has a very different brief from a CIO who wants to lift general AI literacy across 400 staff. Both are legitimate goals. They are not the same program.

Come out of that conversation with a written one-paragraph statement of intent: what changes in behaviour or capability after training, for whom, and by when. Everything else flows from that.

Run a baseline assessment

Before you can define what good looks like after training, you need to know where people are starting from. A baseline assessment does not need to be exhaustive. A short survey covering current tool usage, confidence with AI concepts, and day-to-day tasks where AI could plausibly help will give you enough signal to make good decisions.

Building that baseline well is a distinct exercise, and it is worth doing properly. The output tells you whether you have a fragmented picture (some teams are already using AI daily, others have never opened Copilot) or a more uniform starting point. That distinction shapes everything from cohort design to content depth.

Define the first cohort

Trying to train everyone at once is one of the most reliable ways to underdeliver. The first cohort should be small enough to manage and specific enough to measure. Somewhere between 20 and 60 people is a workable range for an initial rollout, though this depends on your organisation's size and the delivery model you are considering.

Pick a cohort that represents a real business unit with a real use case, not a mix of volunteers from across the organisation. A focused group gives you a proof of concept you can take to the rest of the business. It also gives you a contained environment to learn from before you scale.

The temptation to go broad early is understandable

Starting with one well-defined cohort produces better outcomes than a wide rollout with vague objectives. A tight first cohort gives you measurable results you can actually use to build the case for what comes next.

Set learning objectives that are specific enough to measure

Vague objectives produce vague training. "Staff will understand AI" is not a learning objective. "Finance team members will be able to use Copilot to draft and summarise internal reports without IT assistance" is.

For each role or team in your first cohort, write two or three objectives in that format: a specific action, a specific context, and a measurable outcome. If you find yourself unable to write them that concretely, that is a sign you need another conversation with the business unit before you move forward.

This is also the moment to decide on delivery format. Will this be a live workshop, a custom program built for your environment, or some blend? The answer depends on your cohort size, geographic spread, and how tightly the content needs to reflect your actual tools and workflows. In-person delivery has advantages that are easy to underestimate, particularly for teams new to AI who benefit from being in the same room to ask questions in real time.

By day 30, you should have a clear brief: who the cohort is, what they need to be able to do, and a shortlist of delivery approaches to evaluate in Phase 2.

What happens in Phase 2 (Days 31-60)?

Phase 2 is where your AI training implementation plan moves from insight to decision. You have a skills baseline, you know which teams are the highest-priority, and now you need to agree on format, choose a delivery approach, and get a program outline onto paper before the clock runs out.

Agree on format before you talk to anyone

The format decision shapes every conversation that follows, so make it internally before you brief a provider. The main variables are:

For most enterprise teams, a blend of in-person and structured practice outperforms either extreme. Live sessions build shared vocabulary and momentum. Structured follow-up, whether that is a short practice task or a team challenge, converts that momentum into habit. A self-paced module on its own rarely does either.

If your pilot cohort is non-technical, the design choices look different again. Role-relevant scenarios, familiar tools, and low jargon matter far more than technical depth.

Brief a provider or build internally?

This is also the phase where you decide whether to buy, build, or both. Building internally using a train-the-trainer model can work well at scale, but it takes longer to stand up and puts real demands on the people you nominate as facilitators. For a first cohort on a 90-day timeline, most organisations find external delivery faster to execute and easier to quality-control.

If you are going to market, do it in the first two weeks of Phase 2. Allow time for provider scoping calls, a proposal, and at least one round of content review. Rushing this step is one of the most reliable ways to end up with a generic program dressed in your logo.

What a good provider brief actually contains

State your audience's current skill level, the tools they use day-to-day, the specific outcomes you want, and any constraints (time zones, languages, compliance requirements). The more specific the brief, the more useful the response. A brief that says "AI training for 200 staff" produces a generic proposal.

When you evaluate proposals, look past the credential list and ask how the provider would adapt their content to your workflows. A concrete example of that adaptation, even a hypothetical one, tells you more than a client roster. For more on what to ask, see the sibling article on choosing an AI training provider.

Build the program outline

Whether you are working with an external provider or building in-house, the output of Phase 2 should be a documented program outline covering:

• Target audience and cohort size for the pilot

• Learning objectives stated as observable behaviours, not vague goals

• Session structure including timing, topics, and any pre-work

• Tools and platforms the training will reference

• Assessment or measurement approach (even a simple one, such as a before-and-after skills check)

Getting this onto a shared document, reviewed by at least one business stakeholder and signed off, protects you from scope creep later. It also gives you something concrete to show senior sponsors who want to know where the program is heading.

By Day 60, you should have a confirmed format, a provider briefed or an internal facilitator assigned, a program outline approved, and a date locked for the first session. If you do not have those four things, Phase 3 will be reactive rather than deliberate.

How do you execute Phase 3 (Days 61-90)?

Phase 3 is where the planning becomes visible. The first cohort runs, real people sit through real sessions, and you find out quickly what you got right and what needs adjusting.

Confirm logistics before day 61

The two weeks leading into Phase 3 are mostly about removing friction. Confirm room bookings or virtual session links, check that licences are active for any tools participants will use during training, and send calendar invites with enough lead time that people can actually protect the time. If you are running in-person sessions, brief line managers directly: attendance drops sharply when a manager treats training as optional.

Send participants a short pre-work note. Not a reading list, just context: what the session covers, what they should bring (a real task or workflow they want to improve works well), and what they will be able to do differently by the end. Two paragraphs is enough.

Run the first cohort

Keep the first cohort deliberately small. Twelve to twenty people is a workable range. You want enough participants to generate useful discussion, but few enough that facilitators can respond to the room rather than broadcasting at it.

The first session should prioritise application over explanation. Participants who spend two hours using AI on their actual work will retain far more than those who sit through a capability overview. If you are using a structured workshop format, work with your provider to ensure at least half the session time is hands-on. This is especially important for non-technical teams, where abstract demonstrations of AI capability rarely translate into changed behaviour on Monday morning.

Assign one internal person, ideally from L&D or a relevant team lead, to be present throughout. They are not there to co-facilitate. They are there to observe what lands, what confuses people, and what questions come up that the content did not anticipate.

Capture feedback while it is still fresh

Run a short debrief within 24 hours of the session, not a week later when the detail has blurred. Five questions is enough: what was most useful, what was unclear, what they wished had been covered, whether they feel confident applying what they learned, and what support they need next.

Quantitative ratings have their place, but the open responses are where the usable signal lives. A comment like "I still don't know what I'm allowed to use AI for at work" tells you something a satisfaction score cannot.

The first cohort is a calibration, not a verdict

If feedback reveals gaps in the content or pacing, that is the plan working as intended. Build in time after the first cohort to adjust before you scale to the next group.

Combine participant feedback with the observer's notes and any usage data your tools surface. If you are rolling out Microsoft Copilot, for example, Copilot adoption metrics from the days following training can tell you whether behaviour actually shifted. Compare that picture against your baseline from Phase 1.

Document what you adjust and why. When you brief your provider on changes before the next cohort, a short written summary is more useful than a phone call. It creates a record that will matter when you are onboarding a second or third delivery team later in the year.

What trips up most AI training implementation plans?

Most AI training programs don't fail because the content was bad. They fail earlier, and for more mundane reasons.

Skipping the baseline. If you start designing training before you know what your people already understand, you will pitch the content at the wrong level. Senior professionals who feel talked down to disengage quickly. Beginners thrown into advanced prompting frameworks get lost just as fast. A short AI readiness assessment at the start of Phase 1 prevents both.

Treating it as a one-time event. A half-day workshop is a start, not a program. The organisations that see lasting behaviour change follow up with reinforcement: short practice activities, team check-ins a few weeks later, or a channel where people share what is working. Without that, capability built in the room evaporates within a month.

Choosing tools before choosing outcomes. There is a recurring pattern where the tool decision (Microsoft Copilot, Google Gemini, a custom-built agent) gets made in the technology team and handed to L&D as a fait accompli. Training then gets designed around product features rather than around what people actually need to do differently. The better sequence is: define the workflow changes first, then align the training to the tool that supports them.

The most common Phase 1 mistake

Rushing the discovery work to get to design faster. A scoped, well-briefed training program takes less time to build and runs more smoothly than one that gets revised mid-delivery.

Underestimating resistance from middle management. Frontline staff are often more willing to try new tools than their managers are. Managers worry about accountability, about their teams wasting time experimenting, or quietly about their own relevance. If you do not brief and involve managers early, they become passive blockers. They do not stop the training; they just make it harder for people to apply anything afterwards.

Picking a delivery format that does not fit the audience. A finance team scattered across four states has different constraints to a contact centre team working from the same floor. In-person training builds engagement and hands-on confidence faster for most audiences, but it is not always practical. The failure point is assuming one format fits all roles, all locations, all learning styles. The right answer is usually a deliberate mix, decided during Phase 1 scoping rather than defaulting to whatever is cheapest or easiest to schedule.

None of these are exotic problems. They are predictable, and most of them are avoidable with a structured approach to the first 30 days.

Frequently asked questions

Can we run this plan in less than 90 days?

You can compress the timeline, but only if two conditions are already in place: a clear AI readiness baseline and an executive sponsor who can move budget and approvals quickly. Without those, rushing Phase 1 typically produces a weak needs analysis, which means the training design in Phase 2 is built on assumptions rather than evidence. Many organisations find they save time overall by spending the first 30 days properly, even if the pressure to launch is real.

What does this typically cost for a mid-sized Australian organisation?

The honest answer is that it varies considerably depending on cohort size, delivery format, and how much customisation your content requires. A broad benchmark for a well-scoped first rollout (single cohort, hybrid delivery, some content adaptation) generally sits somewhere between $20,000 and $60,000 AUD. Our article on what enterprise AI training should cost in Australia breaks down the main cost drivers and where organisations commonly over- or under-invest.

How do we get senior stakeholders on board before we have results to show?

Frame the conversation around risk, not enthusiasm. Most executives respond better to "our teams are using consumer AI tools without guardrails right now" than to "AI is transforming everything." If you can show the gap between current AI literacy and what your tools (a Microsoft Copilot licence, for instance) actually require to be used safely and effectively, that tends to sharpen focus quickly. A short readiness assessment, completed in Phase 1, gives you the concrete data to make that case.

What if our workforce is spread across multiple states or works remotely?

Distributed delivery is manageable, but it does change the design decisions you make in Phase 2. In-person training tends to produce stronger early engagement and richer peer discussion, particularly for teams new to AI. If in-person isn't practical for all sites, a hybrid model, where one cohort runs face-to-face and the format is then adapted for remote groups, often works well. The key is not to default to asynchronous e-learning simply because it is logistically convenient.

How do we know the first cohort has actually worked?

Define success criteria before training starts, not after. The most useful measures are behavioural: are participants using the target tools in their daily workflow two weeks after training? Are they prompting differently? You can capture this through short follow-up surveys, manager check-ins, or usage data from the platform itself if your tools expose it. For a more structured approach to this question, the article on how to measure ROI on enterprise AI training covers the evaluation frameworks that actually work in practice.

Ready to turn your plan into a program?

A 90-day plan is only useful if someone owns it. If your organisation has reached the point of knowing it needs AI training but hasn't yet turned that into a structured program, the next step is a conversation, not a document.

Better People works with Australian enterprises to move from intent to delivery: scoping the cohort, setting the learning objectives, and running the first session. Whether that means a single workshop or a multi-phase rollout, the engagement starts with understanding what your teams actually need.

What would a 90-day AI training plan look like for your organisation?

We'll map your cohort, tools, and timelines in a single conversation and give you a clear starting point for your AI training implementation plan.

Book a 30-minute discovery call →