What is Microsoft Copilot good for in a professional services firm?

Why does this question keep coming up?

Many firms bought Microsoft Copilot licences with real hope. The tool sits inside Word, Outlook, Teams, Excel and PowerPoint, the software the firm already lives in, so it seemed the obvious way to bring AI into the building without a new system. A few months on, the same conversation happens in a lot of partner meetings. Some people use it every day and swear by it. Others opened it twice. And nobody can point to a job that now takes a fraction of the time it used to.

That gap is usually read as a problem with Copilot or a problem with the staff. We think it is neither. It is a mismatch between what the tool was designed to do and what the firm hoped it would do.

Where does Copilot genuinely help?

It is worth being fair, because Copilot is good at a set of things that matter. In general terms, it helps with three kinds of work.

• Drafting. Getting from a blank page to a first version: an email reply, the outline of a note, a paragraph explaining a point in plainer language. For a fee earner who stares at an empty document at the end of a long day, that is a real help.
• Summarising. Condensing a long email chain, a lengthy document or, where the firm's set-up allows it, a meeting, into the points that matter. Coming back from leave to a thread with dozens of replies is exactly the situation it suits.
• Searching within Microsoft 365. Finding the file, message or note a person already has access to, by describing it rather than remembering where it was saved. In firms where documents are scattered across mailboxes, shared drives and team sites, that saves genuine time.

Used well, those three things make an individual faster and less frustrated. For a partner who handles a lot of correspondence, they can give back a meaningful slice of the week. That is not nothing, and firms that write Copilot off entirely are throwing away something useful.

Where does it stop?

Everything in that list has the same shape. One person, one task, one prompt, one result. Copilot helps whoever is sitting at the keyboard with whatever they have asked for at that moment. It does not know that your firm's monthly client report has seven sections, that the fee section must match the engagement letter, that the regulatory wording changed last quarter, or that a named partner has to approve anything with a valuation in it.

The work that eats a firm's week is rarely a single task. It is a process: a sequence of inputs, drafts, handoffs and checks that repeats every week or month. A proposal, a client report, a file review, a compliance pack. To cut that kind of job substantially, something has to hold the sequence together, apply the firm's rules every time and deliver a finished draft to the right reviewer. A general assistant, however capable, is not set up to do that on its own.

So what usually happens is this. A fee earner uses Copilot to speed up one step of a ten-step job. That step gets quicker. The other nine stay as they were, and the job as a whole barely moves. Multiply that across a firm and you get the familiar result: lots of small, invisible savings and no change anyone can point to. We make the wider argument in a tool is not a process.

Why does quality vary so much between people?

Because the output depends heavily on the person asking. A fee earner who knows exactly what good looks like, gives precise instructions and checks the result carefully will get far more from Copilot than a colleague who types a vague request and accepts the first answer. That is true of every general AI assistant, not just this one.

For a firm, this is a problem of consistency. Two associates producing the same kind of document with the same tool can end up with very different results, in different styles, with different gaps. The firm's standard has not been built into anything. It still lives in whoever wrote the prompt and whoever reviewed the output.

There is also a point about access that is easy to overlook. Copilot works within the permissions each user already has. That is the right design, but it means that if a firm's file permissions have grown loose over the years, Copilot will surface material that people could technically open but were never meant to find. Tidy permissions are worth checking before rollout, not after.

So what should a firm expect from it?

Treat Copilot as a strong personal assistant, not as a strategy. It will make capable people somewhat faster at the tasks they already do. It will help with drafting, reading and finding. It will not, by itself, change how the firm produces its recurring work, and it was never really claiming to.

If the licences are not being used, more training rarely fixes it. People do not avoid the tool because they cannot work out which button to press. They avoid it because it does not obviously fit into the job in front of them. The honest reasons rollouts stall are covered in why AI rollouts fail.

What turns the licence into a changed week?

Start from the job, not the tool. Pick the one recurring process that costs the most senior time, map exactly how it runs today, and rebuild it so that the firm's templates, rules and review steps are built in. That rebuilt process can run inside Microsoft 365, alongside Copilot, in the same tools the team already opens every morning.

At that point the two do different jobs. Copilot handles the unpredictable, one-off tasks that arrive during the day. The rebuilt process handles the work that repeats, the same way every time, with a named person signing off the result. Firms tend to get the most from both when they stop expecting one to do the other's work. If you are weighing a general tool against something built around your own process, custom or off-the-shelf sets out the trade-offs.