The six-week Microsoft 365 Copilot pilot playbook: a week-by-week plan for Australian businesses

Run a Microsoft 365 Copilot pilot over six weeks with 8 to 15 people in document-heavy or meeting-heavy roles. Use a prep week to baseline the metrics you will judge against and fix SharePoint permissions. Weeks one to four build habits and depth. Weeks five and six turn usage data into a per-seat decision.

Most Copilot pilots that fail do not fail on the technology. They fail because nobody wrote down what success looked like before the licences were assigned, so the review meeting turns into a show of hands. This is the plan Frontrow runs with Australian clients.

A short pilot is commercially practical. Copilot is an add-on to a qualifying base plan such as Business Standard, Business Premium, E3 or E5, and Microsoft's licensing documentation imposes no minimum seat count. Australian pricing is published: the Copilot Business add-on lists at AU$31.40 per user per month on an annual commitment, or AU$37.68 monthly, both excluding GST, checked July 2026. The monthly option avoids a twelve-month term for a decision you have not made yet.

What has to be true before week one starts?

Week 0 is preparation and sits outside the six weeks on purpose. Skipping it is why most pilots produce an argument instead of an answer.

Pick the cohort deliberately

Eight to fifteen is the right size. Fewer, and one person's absence swings the result. More, and the coaching load gets too thin to be useful. Choose people whose output is measurable and whose work is heavy on documents, email or meetings: proposal writers, account managers, finance analysts, HR and operations coordinators, and a senior manager who lives in Teams. Include one or two sceptics, because an all-volunteer cohort flatters the result.

Baseline the metrics before anyone gets a licence

You cannot measure improvement against a number you never captured. Before licences are assigned, ask each participant to record honest baselines for the tasks Copilot is expected to touch. Rough self-reported figures are fine, because the same person re-rates the same task in week six.

• Time to produce one representative document, such as a first-draft proposal or a monthly report
• Time spent writing up meeting notes and actions in a typical week
• Time spent searching for information across SharePoint, Teams and email
• Volume measures where they exist, such as proposals sent per month
• A one-to-five rating on how easily they find internal information today

Fix the permissions problem first

This is the single most common reason Copilot pilots go wrong. Copilot answers using content the signed-in user can already open, and it grants no new access. What it does is make existing access instantly searchable in conversational language, so a tenant with untidy SharePoint permissions will have that untidiness surfaced within days.

The usual culprits are familiar: a remuneration file on a site shared with everyone, an old HR folder inheriting permissions from its parent, sharing links set to anyone in the organisation and forwarded on, departed-employee OneDrive content nobody reassigned. That material was always reachable, just invisible because nobody searched for it.

Microsoft treats this as a first-class deployment step. Its secure and governed data foundation blueprint is organised around three pillars: remediate oversharing, set up guardrails, and meet regulations. SharePoint Advanced Management is included with a Copilot licence and covers the first pillar. In practice, Week 0 means running the data access governance reports in the SharePoint admin centre and correcting permissions on the sites your cohort touches.

Confirm the technical prerequisites

Check Microsoft's minimum requirements against the actual cohort. Each participant needs a Microsoft Entra ID account and a primary mailbox in Exchange Online, because mailbox grounding is not supported for on-premises or hybrid mailboxes. Users need a supported operating system and modern browser, and the network must not block the Copilot service endpoints.

One timing detail shapes the schedule. Copilot Dashboard processing in Viva Insights starts once licences are assigned and takes up to seven days, and the dashboard reports a rolling 28-day window with up to a six-day delay. Assign licences at the end of Week 0 so reporting has caught up when you need it.

Weeks 1 and 2: enablement and habit formation

Objective: every participant uses Copilot for one real task per working day. Involved: the cohort and a nominated internal champion. Exit test: the usage report shows nearly all participants active in the last seven days, and each can name one task they now do differently.

1. 1Day 1: one 60-minute kick-off covering what Copilot can see, what it cannot, and the expectation that outputs are checked before they leave the building.
2. 2Day 1: hand each participant a one-page prompt sheet written for their role. A bid writer's sheet and a finance analyst's sheet should share almost nothing.
3. 3Days 2 to 5: one task per day, logged in a shared Teams channel with a sentence on what worked.
4. 4End of week 1: a 30-minute stand-up where the champion collects blockers and shares the prompts that worked best.
5. 5Week 2: move into the apps people already live in, starting with Outlook summarisation and drafting, then Word drafting from an existing brief.
6. 6End of week 2: anyone inactive for seven days gets a fifteen-minute one-to-one, not a reminder email.

Adoption is a habit problem rather than a technology problem. Nobody struggles to operate Copilot. They struggle to remember it exists when they open a blank document, because years of muscle memory say to start typing. The daily task and the shared channel break that reflex, and dropping either is why pilots go quiet in week three.

Weeks 3 and 4: depth and worked examples

Objective: move past summarisation into work that carries real time value, and capture examples with numbers attached. Involved: the cohort, the champion, and the managers who will sign off on seats. Exit test: one documented worked example per role, with a before-and-after time on the same task.

1. 1Week 3, meetings: turn on Copilot in the cohort's recurring internal meetings. Focus on recap, action extraction, and catching up on a missed meeting.
2. 2Week 3, documents: move from drafting single emails to drafting from source material, such as a first-cut proposal from a scoping document.
3. 3Week 4, data: work through Excel tasks with the finance and operations participants, including formula explanation, data cleaning, and turning a table into commentary.
4. 4Week 4, search: run retrieval tasks across SharePoint and Teams. This is where the Week 0 permissions work either pays off or shows up as a gap.
5. 5Week 4, agents: prototype one Copilot Studio agent only if a genuine repeating process exists. If nobody can name it in a sentence, record agents as out of scope.
6. 6Throughout: each participant writes one worked example on a page. Task, time before, time after, what a human still had to do.

The worked examples matter more than the telemetry. Usage data tells you people are using Copilot. An example showing a monthly board pack summary going from four hours to ninety minutes tells a finance director something actionable. Collect one per role, including the unflattering ones.

Weeks 5 and 6: measurement and the decision

Objective: convert six weeks of activity into a per-seat business case and make a documented decision. Involved: the champion, the IT lead, and the budget holder. Exit test: a written recommendation, signed off or rejected on the evidence.

What data to gather

• Usage telemetry from the Microsoft 365 Copilot usage report in the admin centre: enabled users, active users, active user rate, prompts submitted, adoption by app, and per-user last activity dates across 7, 28, 90 or 180-day windows, exportable to CSV. Data is typically available within 48 hours of the end of the day in UTC.
• Self-reported time saved, captured by re-asking the Week 0 baseline questions of the same people in the same wording.
• At least one worked example per role, with the before-and-after figures attached.

Two definitions stop people misreading the numbers. A user counts as active only on an intentional action such as submitting a prompt, so opening the Copilot pane in Word and closing it does not register. And the Copilot Dashboard makes benchmarks, week-level trendlines and sentiment metrics available at 50 or more assigned licences, which a pilot will not have. A pilot of fifteen gets the readiness and adoption pages plus the admin centre report, so self-reported data and worked examples carry more weight.

Turning it into a per-seat number

At AU$31.40 per user per month excluding GST on an annual commitment, a seat costs roughly AU$377 a year excluding GST. Apply your own fully loaded hourly cost. If an hour of a participant's time is worth AU$60, the seat breaks even at about six and a half hours saved per person per year, or roughly eight minutes a working week. That threshold is low, and saying so plainly beats inflating the savings. The real question is whether a given seat holder will use Copilot consistently enough to clear it.

The three honest outcomes

A pilot with only one acceptable answer was never a pilot. All three of these are legitimate results.

1. 1Roll out broadly. Sustained usage across most of the cohort, examples from more than one role, and a case that clears your threshold comfortably. Expand in waves, keeping permissions remediation ahead of each one.
2. 2Roll out to specific roles only. Value concentrated in two or three roles while others barely engaged. This is the most common outcome and a good one. Buy seats where value was demonstrated and revisit the rest in six months.
3. 3Stop. Low usage that coaching did not fix, no examples with defensible numbers, or permissions that need work first. Hand the licences back. A pilot costing a few thousand dollars that prevents a whole-of-business purchase has done its job.

How do you avoid the two ways pilots fail?

The first is the pilot that proves nothing because success was never defined. Six weeks pass, the review asks whether people liked it, everyone says it was interesting, and the decision defaults to whoever speaks with most conviction. The fix sits entirely in Week 0: write down the baselines and the threshold, and get the budget holder to agree to it in advance.

The second is subtler and more expensive. The pilot succeeds on enthusiasm alone. Feedback was glowing, the business buys 200 seats, and a year later the active user rate sits under a third. Enthusiasm is a weak predictor, because a self-selected group getting hands-on coaching behaves nothing like the two-hundredth person who receives a licence and an email. Insist on one quantified example per role before expanding, and keep watching the active user rate afterwards.

Why does Copilot land harder for some roles than others?

The value is genuinely uneven, and being straight about that builds more trust than overselling. Copilot works on language and on content held inside Microsoft 365, so the closer a role sits to writing, summarising and meetings, the more it returns. Strong fit means people who draft from source material, sit in many meetings they must act on, or spend real time hunting for internal information: bid writers, account managers, executives, HR, and finance staff.

Weaker fit means people whose day sits inside a line-of-business application rather than Microsoft 365, who work in field operations, or whose output is repetitive and already templated. A warehouse supervisor or a technician on the tools will see far less, because the work does not pass through the surfaces Copilot reads. That is why role-specific rollouts come up so often: a 200-person business may find a genuine case for 40 seats and none for the other 160. Building the cohort from both a strong-fit and a weaker-fit role is what shows you where the line sits.