Decide where to try AI first by looking at how your week's hours are spent, not at what a tool can do.
You try an AI assistant on whatever's on your screen that afternoon. Sometimes it helps. Often it's bland, wrong, or slower to fix than to write.
A few weeks later you'd say AI is good for some things, but you couldn't say which. Your week looks the way it did before.
Start from the other end. Before you pick a tool or a technique, look at where your hours go. A saving only counts if it lands on a task that takes real time. And you only know it's real if you timed the task before you changed it.
Say you work in operations at a mid-sized firm. Monday morning goes on a weekly report that pulls figures from three spreadsheets, about two hours. Email takes two hours a day, which is 10 hours a week.
You sit in seven meetings, and each one leaves 20 minutes of notes to write: 7 x 20 = 140 minutes, a bit over two hours. Once a month there's a deck for management. In between come one-offs, like checking out a new supplier or learning a system the company just bought.
Laid out like that, the week starts sorting itself. Email is frequent, mostly routine and quick to read back, so it's a strong first try. Meeting notes are frequent and dull, but they carry decisions and names that have to be right.
The report comes round every week too, but a wrong figure in front of management is expensive. The supplier research is occasional and depends on current facts, which an assistant may not have and may make up.
Three questions do most of the sorting. The first is how often the task comes up. A prompt you reuse every week pays back fast: save 15 minutes on a weekly task and that's 15 x 52 = 780 minutes, or 13 hours a year. Save 30 minutes on something you do twice a year and you get one hour back.
The second is how fast you can check the output. A draft email takes a minute to read. A reconciliation you'd have to redo by hand to trust saves you nothing.
The third is what a mistake costs. A clumsy internal note costs very little. A wrong number in a client proposal costs a lot. Microsoft says it plainly about its own assistant: AI answers aren't guaranteed to be 100% factual, so check them before they go to anyone else.
Run those three across your week and the same kinds of task usually come out ahead: drafting, reformatting, summarising, and first-pass research you then verify. Each one gives you something you can read against what you already know in a few minutes, so a weak result costs you little. Tasks that hang on fresh facts or on judgement about people, like a performance review or a reply to an upset client, need more care. Some should stay with you.
The caution isn't only common sense. A 2023 Harvard Business School and BCG study of 758 consultants found GPT-4 made them faster and better on tasks inside what it could do, and 23% worse on a problem picked to sit outside it. Tasks that look equally hard to you can land on either side of that line, so you find yours by testing on work you can check.
One more filter before you try anything: data. Some of your longest tasks will involve client names, salaries or contracts, which you may not be allowed to paste into an assistant. Mark those for now and keep them out of your first tries.
Then make the list. Open last week's calendar and your sent folder, because they give better hour counts than memory does. Write down the ten tasks that took the most time, with rough hours next to each.
Keep the labels specific: "weekly report from three spreadsheets", not "reports", so you can tell later why each one earned its mark. Mark each task try first, needs care or keep yourself, and circle any that touch client or staff data.
The operations week, made into a finished list, looks like this:
| Task | Hours | Mark | Circled | |---|---|---|---| | Email | 10 | Try first | | | Monthly deck for management | 3 | Needs care | | | Notes after seven meetings | 2.3 | Needs care | | | Weekly report from three spreadsheets | 2 | Needs care | | | Researching a new supplier | 2 | Needs care | | | Learning the new system | 1.5 | Keep yourself | | | Summary of a client's contract renewal | 1.5 | Needs care | Yes | | Checking staff overtime claims | 1 | Needs care | Yes | | Reformatting a supplier price list | 1 | Try first | | | Reply to an upset client | 0.5 | Keep yourself | Yes |
Rough hours are enough to sort with. Once you've picked the three to five tasks you'll test, time them properly before you change a thing. Otherwise you'll never know what you saved, only that it felt quicker.
Open last week's calendar and sent folder and list your ten biggest tasks with rough hours. Mark each try first, needs care or keep yourself, and circle any with client or staff data.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).