What to Test in a Microsoft 365 Copilot Business Trial

Coding Liquids tutorial cover featuring Sagnik Bhattacharya for What to Test in a Microsoft 365 Copilot Business Trial.
Coding Liquids tutorial cover featuring Sagnik Bhattacharya for What to Test in a Microsoft 365 Copilot Business Trial.

Before a Microsoft 365 Copilot Business trial ends, run five real jobs from your own week with each licensed person, log minutes saved and corrections needed, and check the admin centre usage report. Commit only if the saving beats roughly $21 to $25.20 per user per month; otherwise cancel inside the trial.

The Complete Excel Course with AI Integration

From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.

By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner

Buy now

Most owners test Copilot by asking it clever questions, which tells them nothing about their invoices, quotes or Monday meeting. A trial is thirty days of your staff's time, so the test needs a plan, a score sheet and a decision rule written down before day one. The plan below is for you to run in your own business; I give no results of my own, only the method, the figures you need to price it and the traps to avoid.

Is the Copilot Business trial free, and what should you confirm first?

Microsoft's small-business Copilot page lists a one-month free trial for the listed plans and the Copilot Business add-on, with a credit card required, converting to a paid subscription unless you cancel. Paid subscriptions have a 7-day window for a prorated refund. Trial terms vary by account and over time, which is why the first step is to read your own checkout screen and write down three things: the trial end date, the price it converts to, and how to cancel.

Three further points from the same page and Microsoft Learn decide how you run the test:

  • The promotional price excludes trials. Microsoft's page lists an $18 a user per month first-year price on the add-on, with an annual commitment, between 1 July and 31 December 2026, and says it is not available on trial subscriptions. List price is $21 a user a month on the annual commitment, paid yearly. Paying monthly on that annual commitment costs 5% more (about $22.05), and the month-to-month option that Microsoft partners sell costs $25.20, according to Microsoft's March 2026 partner pricing; Microsoft's own pricing page lists only the annual price (checked 20 November 2026).
  • Copilot Business is an add-on. Microsoft Learn describes it for organisations of up to 300 users with eligible Microsoft 365 Business licences, available with an annual commitment, so you need a qualifying base plan.
  • Some features bill by use. The page says advanced experiences such as Cowork and Autopilot are billed through Copilot Credits, and admins can set spending controls in Cost Management in the admin centre.

Microsoft has also renamed things: Microsoft Learn now calls the app "Microsoft Copilot", while the product pages and licence tables still say Microsoft 365 Copilot Business in places. If your staff see either name, they are looking at the same family of products, so confirm which licence a person has before you judge what they see.

What price must Copilot beat to be worth keeping?

Copilot is worth keeping only if the monthly value of time saved per person exceeds the monthly cost per person, with some margin for the effort of learning it. Write that decision rule before the trial, not after you have grown attached to the tool.

Take an illustrative six-person signage company with a loaded staff cost of $30 an hour. At $21 a user paid yearly, a licence needs to save 42 minutes a month to break even ($21 divided by $30 is 0.7 hours). At about $22.05, paying the annual commitment monthly, it needs 44 minutes, and at $25.20 month to month through a Microsoft partner it needs 50. Those are small numbers, which is why a lazy trial passes: almost anything saves 50 minutes in a month. So set the bar higher on purpose: ask for 2 hours a month per licensed person, about 30 minutes a week, because you also pay for training, review time and the hours spent on tasks where the tool did nothing useful.

Licence price per user per monthBreak-even minutes at $30 an hourTarget to commit (about 3 times)
$18 (first-year offer, if eligible)36about 110 minutes
$21 (list, paid yearly)42about 125 minutes
$22.05 (annual commitment, paid monthly)44about 130 minutes
$25.20 (month to month through a Microsoft partner)50about 150 minutes

The three-times margin is a judgement, not a rule from Microsoft. It leaves room for the days the tool is not useful. Change the multiple if your margins are thin or your staff cost is high. For local list prices in other currencies, see the three-currency price comparison.

What should you set up before the Copilot trial starts?

Before the trial starts, assign Copilot licences to your testers in the Microsoft 365 admin centre, check what their files can see and put the cancel date in the calendar. An hour of set-up before the trial saves a week of confusion. Microsoft Learn's licence guidance says to buy the licences and then assign them to users in the admin centre, and it points to a service plan diagnostic tool that checks whether a user account meets the requirements. Use both.

  1. Pick three to five testers with different jobs: a writer of quotes or proposals, a spreadsheet user, a meeting-heavy manager and one sceptic. Avoid testing only enthusiasts.
  2. Assign the licences in the admin centre and note the date. Ask each person to sign out and back in to Word, Excel, Outlook and Teams so the new licence is picked up.
  3. Run the diagnostic for each tester if Copilot does not appear.
  4. Check what the files can see. Copilot works with what a person already has access to, so a messy shared drive with wrong permissions will show up in the answers. Fix the worst folders first; the readiness checklist covers this.
  5. Decide spending controls for any usage-billed features, or switch them off for the trial.
  6. Write the cancel date in the shared calendar with a reminder five days earlier.

The privacy question usually comes up in week one. Whether Copilot keeps your business data private explains what Microsoft says about prompts, and what remains your responsibility.

Which five jobs should you test Copilot on?

Test Copilot on five recurring jobs taken from each tester's last month of work, not five impressive ones. The test is whether Copilot makes the jobs you already do faster, or better, with the same checking you already do. An illustrative interior design studio might pick:

  1. Turn a client call's notes into a written scope and proposal in Word.
  2. Summarise a long email thread before replying in Outlook.
  3. Pull a supplier price list into an Excel comparison and flag the cheapest line per item.
  4. Summarise a Teams meeting into decisions and actions.
  5. Draft a project update for a client from a folder of notes.

A financial adviser's five might be different: summarising a long client email chain, drafting a review meeting agenda from file notes, building a meeting recap, turning a spreadsheet of fund fees into a comparison table and rewriting a technical paragraph in plainer words. Whatever your sector, three rules apply: each job must recur at least monthly, each must have a known right answer you can compare against, and none should involve unreviewed regulated or legal wording.

Time each job twice

For each job, record the minutes it took last time by hand, from the tester's memory or a timesheet. During the trial, time the same type of job with Copilot, including the time to check and correct the result. The checking time is the part that most trials leave out and the part that decides the outcome.

Tester: [name]    Role: [role]    Licence assigned: [date]

Job                         Minutes by hand   Minutes with Copilot   Corrections needed   Keep using? (Y/N)
1. Proposal from call notes     [  ]              [  ]                  [  ]               [  ]
2. Email thread summary         [  ]              [  ]                  [  ]               [  ]
3. Supplier price comparison    [  ]              [  ]                  [  ]               [  ]
4. Meeting recap                [  ]              [  ]                  [  ]               [  ]
5. Client project update        [  ]              [  ]                  [  ]               [  ]

Monthly occurrences of each job: [  ]
Saving per month = sum of (hand minus Copilot) x occurrences = [  ] minutes
Worst error found: [describe]

A filled-in row, to show what a useful entry looks like

These numbers are invented to show the arithmetic. Job 4, meeting recap: by hand 25 minutes, with Copilot 8 minutes including checking, three corrections (one wrong owner on an action, one missed decision, one wrong date). The job recurs eight times a month. Saving: 17 minutes times 8 is 136 minutes. At $30 an hour that is $68 of time against a $21 to $25.20 licence. But note the worst error, a wrong date on an action. If recaps go to clients, the checking step must stay, and the saving figure already includes it.

If the spreadsheet job on your list is the weak one, try The Complete Excel Course with AI Integration on Coding Liquids.

One job walked through: the proposal from call notes

To make "time each job twice" concrete, here is job 1 for an illustrative interior design studio. The tester pastes rough call notes into Word and asks Copilot for a first-draft proposal. The prompt below is a template to adapt. The sample output after it is invented to show what to look for, not a recorded result.

Using the notes below, draft a one-page proposal for the client.
Structure: what we understood, what we will do, what is not included,
timeline in weeks, fee. Use only facts from the notes. Where the notes
do not say, write [CONFIRM] instead of guessing.

Notes:
[paste call notes]

A realistic draft might come back well structured, with a sensible timeline of six weeks and a fee line that reads "approx. $4,800". Two checks matter. First, did it invent anything? If the notes never mentioned $4,800, the figure is a guess dressed as a fact, and the [CONFIRM] rule exists to catch it. Second, did it omit anything the client said? A missing "no work before the school holiday" is a worse error than a clumsy phrase. Count each as a correction. If a job regularly produces two or more factual corrections, either improve the prompt (the [CONFIRM] rule often cuts them) or accept that this job needs full human review and recalculate the saving on that basis.

Before and after on one paragraph

Illustrative: the call notes say "client wants matte black fittings, budget tight, install after 12 March". A weak first draft says, "We will install premium matte black fittings throughout in March at a competitive price." The fixes are plain. Premium is the tool's word, not the client's. "Throughout" widens the scope beyond what was agreed, and "in March" ignores the 12 March start. The corrected line reads, "We will fit matte black fittings in the rooms listed in section 2, starting after 12 March, within the agreed budget of [CONFIRM]." Three edits, about a minute. That is the kind of correction to log.

What checking rules should a Copilot trial follow?

A Copilot trial needs four checking rules written on the score sheet before anyone starts, because a trial that skips checking will look better than the real thing:

  1. Every number, date and name in a Copilot output is compared with its source before it leaves the building.
  2. Anything sent to a client is read in full by the tester, not skimmed.
  3. Legal, tax and employment wording is not drafted by the tool without an adviser's review.
  4. The tester logs each correction in one line, because the corrections list is how you improve prompts.

This is also where small businesses learn most. After two weeks, the corrections list shows which jobs are safe to speed up and which need a person at every step.

What to score beyond minutes

Time saved is the main number, but three other things decide whether staff keep using it after the novelty fades.

  • Quality: would a client notice? Mark each output as better, same or worse than your normal work.
  • Trust: how many errors did the tester find, and how severe? One wrong figure in a quote matters more than ten clumsy sentences.
  • Friction: did the tester forget it existed? A tool nobody opens saves nothing, and the usage report will show it.

The sceptic's notes are the most valuable. Ask them to write one sentence per week: "The thing that annoyed me most was...". If the same annoyance appears twice, it is a real limit of the tool for your business.

What does the Copilot usage report show?

The Copilot usage report shows enabled users, active users and the active-user rate, and splits enabled and active users by Microsoft 365 app. Microsoft Learn describes it in the admin centre under Reports, then Usage, where it is now called the Microsoft Copilot usage report. Enabled users are people holding a Copilot licence in the period, and active users are those who actually tried a Copilot feature. Microsoft Learn says a day's activity typically appears within 48 hours of the end of that day (UTC), up to 72 hours after the activity itself, so a day-one check will look empty. Menu labels change, so if you cannot find it, search the admin centre for "Copilot usage".

Use the report three times:

  1. Day four: confirm every tester appears as enabled. If someone does not, the licence did not assign properly.
  2. Day fifteen: compare active users to enabled. A tester who is licensed but inactive needs a conversation, not a verdict.
  3. Day twenty-five: note the apps in use. If Word and Outlook are busy but Excel is untouched, ask why before judging the whole product.

The report shows activity, not value. A person can be "active" by trying a feature once. Treat it as a way of catching empty licences, and treat your minutes table as the proof of savings. Active-user rate near 100 per cent with a poor minutes table means the tool is being tried and found wanting; a high saving with a low active rate means the benefit is concentrated in one person.

A four-week schedule that fits around real work

WeekWhat the testers doWhat you do
0Nothing yetChoose testers, assign licences, fix permissions, record by-hand times for the five jobs
1Do each of the five jobs once with Copilot and log itCheck usage report on day four; run a 20-minute stand-up on what is confusing
2Repeat jobs on live work, not practiceCollect the worst errors and add rules to the checking steps
3Try one new job each from a colleague's suggestionCheck active users against enabled; ask inactive testers why
4Final scoring and one-line verdictsAdd up minutes, apply the decision rule, cancel or confirm before the date

The schedule deliberately uses live work in week two. Practice prompts flatter the tool; a real quote with half the details missing does not.

Three tester stories the sheet can show

Here are three results you might see, with what each means. The people are invented illustrations.

  • The proposal writer saves 200 minutes a month and finds two errors. That is a commit for this role, with the checking step written into the process.
  • The spreadsheet user saves 15 minutes and spends more time explaining the data to the tool. That is a cancel for this role for now. Spreadsheet help depends on tidy tables, and the site has a separate guide on which licence Copilot in Excel needs.
  • The sceptic never opens it after day six. That is not a failure of the tool alone; find out whether the job mix was wrong or the habit never formed.

The outcome you want from the trial is not a yes or no for the whole firm, but a table of who benefits.

Should you commit, partly commit or cancel after the trial?

Commit when most testers save at least three times the licence cost with no serious unchecked errors. If only some roles save time, buy licences for those roles and cancel the rest; if savings are near break-even for everyone, do not commit.

What the sheet showsDecision
Savings of 3 times the licence cost or more for most testers, no serious unchecked errorsCommit, and roll out in stages
Savings for two roles onlyBuy licences for those roles; cancel the rest
Savings near break-even for everyoneDo not commit; use the free Copilot Chat and revisit in six months
Serious errors on client-facing workCancel, or restrict to internal tasks with a named reviewer
Strong results, but spend on usage-billed features risingSet spending controls before committing; see how credits and metered billing work

Before you confirm, also check the paid-licence price for your situation. A bundle with Business Standard or Premium can differ from an add-on, so the bundle against add-on cost is worth a read, and the wider worth-it question puts the decision in context.

What mistakes spoil Copilot trial results?

The commonest trial mistakes are testing on practice tasks, forgetting the cancel date and letting one enthusiast speak for everyone.

  • Testing on practice tasks. Use this week's real work.
  • Forgetting the cancel date. The trial converts to a paid subscription unless you cancel, so the reminder is part of the plan.
  • Letting one enthusiast speak for everyone. Score each person separately.
  • Counting generation time but not checking time. A fast draft that needs twenty minutes of repair is no saving.
  • Assuming the offer price survives the trial. The promotional price excludes trial subscriptions; ask before you start.

If you would like someone to look at your real jobs and choose which five are worth testing, that is a fit for an AI implementation consultation. Otherwise, copy the sheet, set the cancel date and run the month.

Copilot Business trial: questions owners ask

Is the Copilot Business trial free?

Yes, for the trial period, with conditions: Microsoft's small-business page lists a one-month free trial for the listed plans and the Copilot Business add-on, with a credit card required. It converts to a paid subscription unless you cancel during the trial. Terms can change and vary by account, so read the checkout screen and note the exact end date.

Does the $18 offer apply during the trial?

No. Microsoft's page says the promotional price is not available on trial subscriptions, runs from 1 July to 31 December 2026, needs an annual commitment and covers the first year only. Starting the trial can therefore change which price you pay afterwards, so ask your reseller or check before you click.

How many people should be in the trial?

Three to five people with different jobs beat one enthusiast. Pick someone who writes a lot, someone in spreadsheets, someone who runs meetings, and a sceptic. A sample of one tells you about that person, not about the team.

How long until the usage report fills up?

Up to about 72 hours after the activity: Microsoft Learn says a day's Copilot activity typically appears within 48 hours of the end of that day (UTC), so a check on day one will look empty. Look at the report from about day four, and again in the last week, using the enabled and active users figures.

What if the results are mixed?

Keep licences only for the roles with a clear saving and cancel the rest. Copilot Business is sold per user, so the choice does not have to be all or nothing, subject to the minimums and commitment terms on your agreement.

What should you test in a Copilot Business trial?

Five real recurring jobs from each tester's last month, such as turning call notes into a proposal in Word, summarising an email thread in Outlook, comparing supplier prices in Excel, summarising a Teams meeting and drafting a client update. Time each job by hand and with Copilot, log corrections, and check the usage report.

Further reads

Sources: Microsoft's small-business Copilot pages, Microsoft Learn licence options page (updated 4 Nov 2026), Microsoft 365 admin centre usage-report documentation, checked at the time of writing. Trial terms vary, so check the checkout screen.

Want a second pair of eyes on your setup?

Bring the process you're trying to fix. On a 1:1 call we'll work out which AI tools should handle it, which parts stay manual, and how to roll it out without disrupting the team.

Book a 1:1 call with me