Microsoft Copilot is being made available to up to 505,000 NHS staff in England, after a trial with 30,000 of them. Staff in the trial put their saving at 43 minutes a working day, about five weeks a year each. The NHS limits it to administrative work, not clinical decisions.
That number is real, but it understates what is happening in places like Yorkshire Ambulance Service.
Most useful questions never get asked
Every manager has a list of questions they never ask, because the answer would cost a day. Which areas send the most patients to hospital who could have been treated at home? Joining up several reports to find common themes used to mean hours of trawling through them, so a question like that was asked once a year, if at all. When it takes minutes, it gets asked every week, and the follow-up questions get asked too.
What the ambulance service does with it
Tom Smith, an area operations manager, has Copilot put information into a form he can read and check against his own experience. He also uses it as a sounding board: if we did this, what would the impact be? Samantha Robinson, who runs digital information at the service, uses it to turn operational data into briefings for senior NHS leaders, and national guidance into clear instructions for different teams. Both treat the output as a draft and review it.
The goal is to take fewer patients to emergency departments, since many can be treated safely at home or referred elsewhere. The share has fallen from 60% to 56%. That fall comes from the service's new arrangements with community teams, including a transfer-of-care hub that arranges home care. Copilot's part is helping the team see where the openings are.
What this looks like in an ordinary workplace
Take a warehouse shift manager with weekly exports of late orders, overtime and absences. Today the question "do late orders cluster on shifts with the most agency staff?" waits for an analyst with a free afternoon, so it is asked once a quarter. With an assistant that reads the exports, it takes ten minutes. Then comes the next question: what would moving the second shift's start by 30 minutes change?
The assistant can only set out what the data suggests. The manager checks that against what they know about the floor, the way Smith checks his drafts against his operational experience. A question that cost half a day now costs ten minutes, so a manager who asked one a month can ask a dozen. That is the real gain, and it is much bigger than the time saved on emails.
Who gains
The spreadsheet changed accounting this way. The adding-up work shrank, and the work of deciding what to ask grew. Office AI will do the same to jobs that are mostly collecting and formatting reports. It will make the most of people who know their operation well enough to ask the right question, because the quality of the question is now the limit.
So the first thing to do with an office AI is not to time how fast it drafts an email. Write down three questions about your own operation that you have never asked because the answer would take a day, and ask them.