Ask a team what Copilot does and you will usually get two answers: it writes emails, and it summarises Teams meetings. Both are real. Both are also the shallowest end of what people are paying for.
The trap in the other direction is a training day that tries to cover everything. People leave with 40 features and a use for none of them. So here is the shorter list, by app, of what has been worth the training time in the sessions we run.
In short
- Outlook is the best place to start because the work is short, frequent and low risk if Copilot gets it wrong.
- Excel gives the biggest single time saving and needs the most structure to be reliable.
- PowerPoint is best used to get to a rough draft fast, not to produce a finished deck.
- Coverage is the wrong goal. Three features used weekly beats twenty features demonstrated once.
Outlook: start here
Outlook is where we start almost every session, for a practical reason. The tasks are small, they happen many times a day, and a bad output costs you fifteen seconds rather than a client relationship.
Three things earn their place:
- Summarising a long thread. The 40-message chain you have been added to on a Friday afternoon. This is the feature that converts sceptics, because the value is obvious in one click.
- Drafting a reply from a short instruction. Not "write my email for me". More like "decline this, offer the week after, keep it warm". The instruction is the skill worth teaching.
- Checking tone before it goes out. Useful for anyone who writes when they are annoyed, which is most of us.
What we skip: the more elaborate scheduling and triage features. They demo well and get abandoned.
Excel: the biggest saving, and the most setup
Excel is where the hours are. It is also where Copilot most needs the underlying work to be tidy, and where a confidently wrong answer does the most damage.
Copilot in Excel can now edit the workbook directly rather than only talk about it. It will write and apply formulas across sheets, build charts and PivotTables, sort and filter, apply conditional formatting, and answer questions about what the data is doing (Microsoft Support).
What we train:
- Asking questions about a dataset in plain language before touching a formula. This is the habit shift. Most people reach for a PivotTable out of muscle memory when a question would have been faster.
- Getting a formula written and explained. The explanation matters more than the formula. It is how people stay able to check the work.
- Cleaning and reshaping. Unmerging, splitting columns, making an inconsistent export usable. Unglamorous, and it is where the afternoons go.
The rule we teach alongside it: Copilot does not get to be the only thing that checked a number. If a figure is going to a board, a client or the ATO, a person verifies it.
PowerPoint: get to a rough draft, then take over
PowerPoint generates the most excitement in a demo and the most disappointment afterwards. A deck built end to end by Copilot usually looks like a deck built end to end by Copilot.
Where it does earn its place is the first 60 per cent. Turning an existing document into a starting structure. Getting slides out of a long report so you can argue with the shape rather than stare at a blank file. Pulling a summary out of a deck someone else sent you.
We are direct about this in sessions: use it to stop staring at nothing, then do the thinking yourself. Teams that expect a finished deck stop using it within a fortnight.
Word: fewer features than you would expect
Word tends to get less attention than it deserves. The useful pattern is not "write this document for me". It is drafting from source material you point it at, then rewriting specific passages that are not landing.
For teams that produce proposals, reports or file notes from a consistent structure, this is the second biggest time saving after Excel.
Why covering everything fails
A full-day session can demonstrate forty features. People will remember four, and only if they used them on their own material during the session.
So we cut the list deliberately. Three or four tasks per role, chosen with the team before the day, practised on their actual documents. It looks like less training. It produces more change.
Coverage is a comfortable measure for whoever booked the training. It is a poor measure of whether anything is different in six weeks.
Where this goes next
All of the above is still Copilot as an assistant sitting beside one person doing one task.
The newer capability, where Copilot takes a multi-step job and runs it to a finished result, changes the question from "which features should my team learn" to "which work are we comfortable handing over, and who checks it". That is a team and governance question rather than a feature question, and we have written about it in what Copilot Cowork changes for teams still learning the basics.
The order still matters. Teams that have not made the basics routine do not get more out of the newer capability, they get more output they have not built the habit of checking.
Sources
Product claims on this page were checked against these sources on 11 September 2026. This part of Microsoft's product moves quickly, so verify anything you are about to act on.