Good Transformer
AI for operations and admin: the quiet, high-ROI wins
Most of the day disappears into 'work about work'. That coordination layer is where AI pays off fastest: summarising, drafting, routing. Start there, not with the moonshot.
If your team is busy all day but the real work keeps slipping to the evening, the admin is winning. Most of the day goes on "work about work": searching, summarising, chasing and coordinating. That layer, not some grand project, is where AI pays off fastest.
The instinct with AI is to reach for the big idea: the custom tool, the automated pipeline, the thing that changes everything. Those take months and often stall. Meanwhile the unglamorous coordination tax runs every single day, and it is exactly the work AI already does well.
This post is about where AI quietly gives operations and admin the most time back, a simple way to find your own biggest leaks, and why starting small beats the moonshot.
What "work about work" actually costs
The coordination layer is bigger than it feels, because it is spread in minutes across the whole team. Add those minutes up and the number is uncomfortable.
In Atlassian's State of Teams 2025, a survey of 12,000 knowledge workers and 200 executives, leaders and teams waste 25% of their time just searching for answers. A quarter of the working week, gone before anyone has done the job they were hired to do.
The other half of the tax is fragmentation. In Microsoft's Work Trend Index report Breaking Down the Infinite Workday, published 17 June 2025 and built on aggregated Microsoft 365 signals plus a global survey, the average worker receives 117 emails and 153 Teams messages every weekday and is interrupted every two minutes by a meeting, an email or a notification. Nobody does deep work in two-minute windows.
This is the honest case for AI in operations. Not "AI will run the business", but "AI can absorb the searching, summarising and drafting that eats the day", so your people spend more of the week on work that needs a person.
The quiet AI wins that pay off fastest
Six back-office jobs give operations and admin the most time back for the least setup. None of them is a moonshot. Each produces a draft or a summary a person checks, never a decision made without you.
- Searching and finding. Point AI at your shared drive, wiki or inbox and ask it in plain language, so a member of staff gets an answer with a source in seconds instead of hunting through folders. This targets the single biggest line in the tax.
- Summarising. Let AI condense a long call, a document or a busy email thread into the three points that matter and the actions agreed. The reader starts from a briefing, not a backlog.
- Drafting. The routine email, the status update, the standard proposal, the job advert: AI gives a solid first draft in seconds that a person shapes and approves, rather than starting from a blank page.
- Data entry and moving data. Pulling figures off an invoice, copying details from an email into a system, tidying a list: AI can extract and format the data, and a person confirms it. Slow, error-prone typing becomes a quick check.
- Routing and triage. New requests, tickets and enquiries can be read, categorised and pointed to the right person or queue by AI, so nothing sits unseen and your ops lead is not the human switchboard.
- Scheduling. Finding a time, and the reshuffle when something moves, is pure coordination. AI scheduling can propose slots and handle the rearrangement, taking a slow chore off the desk.
The pattern is the same across all six: AI prepares, a person decides. That is what makes these safe to switch on this week.
Why these beat the moonshot
The quiet wins win because the maths is on their side. A big custom build is one uncertain payoff, months away, that many teams never finish. Summarising and drafting are dozens of tiny payoffs a day, starting now, on work you already understand.
They are also low-risk. The stakes on a routine draft are small, so a wrong answer costs a re-read, not a client. That lets your team build the habit and the judgement to spot when AI is wrong, on work where being wrong is cheap. We have written before that the right first job for AI is the boundary of the tedium: the dull, repetitive edge of the work, not its most valuable core.
There is a trap worth naming. Saving time is not the same as creating value, and hours handed back only count if they go somewhere that matters. If AI frees five hours and they refill with more admin, you have a faster treadmill, not a better business. We cover that gap in productivity versus business value; the short version is to decide in advance what the reclaimed time is for.
The admin-tax map
To find your own biggest leaks, we use a one-page tool we call the admin-tax map. It is deliberately simple: list the six coordination jobs, estimate the hours the whole team spends on each per week, note who mostly does it, and mark whether AI can help today. The rows marked "yes" with the most hours are where you start.
Copy this and fill it in for your team:
| Admin task | Hours/week (whole team) | Who does it | AI-ready today? |
|---|---|---|---|
| Searching for answers and files | |||
| Summarising calls, documents, threads | |||
| Drafting emails, updates, proposals | |||
| Data entry and moving data between tools | |||
| Routing and triage of requests | |||
| Scheduling and rearranging |
Here is the map filled in for a fictional ten-person firm (illustrative figures):
| Admin task | Hours/week | Who does it | AI-ready today? |
|---|---|---|---|
| Searching for answers and files | 10 | Everyone | Yes |
| Summarising calls, documents, threads | 6 | Managers | Yes |
| Drafting emails, updates, proposals | 12 | Everyone | Yes |
| Data entry and moving data | 5 | Ops and admin | Partly |
| Routing and triage of requests | 4 | Ops lead | Partly |
| Scheduling and rearranging | 3 | Admin | Yes |
| Total | 40 |
Forty hours a week, a full person, disappears into the coordination layer. The four "yes" rows account for 31 of them, and they are where the fast wins live.
Now the honest maths on what AI actually claws back. It does not take these rows to zero, because a person still checks the output. A realistic first pass: searching 10 down to 7, summarising 6 down to 3, drafting 12 down to 8, scheduling 3 down to 1.5. That is about 11.5 hours a week returned, more than a full working day for the firm, and it lands where the survey said it would, on searching and drafting. The gain is real because it turns writing time into reviewing time, not because it hands the work away.
Where to start on Monday
Do not try to fix all six rows at once. Take your ten minutes to fill in the admin-tax map, pick the single "yes" row with the most hours, and switch on AI for that one job in a tool your team already uses. For most firms that first row is searching or drafting.
Run it for one week with one rule: a person reads every AI output before it is used or sent. On Friday, check one thing. Did it save real time without costing accuracy or your team's own voice. If yes, add the next "yes" row. If no, you have spent a week finding out, not a quarter.
Choosing that first job well is most of the win, and it is worth getting the use case right before adding another tool.
If you want help mapping where your week actually leaks and setting up the first fix safely, book a session and we will build it around how your team really works.