You will sort your recurring work into three groups and set a check for each.
This exercise turns the three groups from lesson 7.3, Decide what to hand over, what to assist, what to keep, into a map of your actual work. You'll list your ten most common tasks, sort each one, write the check you'll do for anything AI drafts, and book one task a week to do by hand.
It takes about 25 minutes. You'll need your calendar, your sent email and your recent files open, because memory alone tends to miss the routine tasks that fill most of a working week.
Look back over the last two weeks. Go through your calendar, your sent folder, your recent documents and your chat history with any AI assistant. Write down the ten tasks that came up most often.
Be specific. "Emails" is too broad to sort. "Replying to supplier queries about delivery dates" and "writing the weekly update to my manager" are different tasks with different stakes. Include the small ones too, such as formatting reports, booking meetings, or summarising calls.
If you don't currently use AI for a task, still list it. The map is about your work, not only about where AI already is.
For each task, choose delegate, assist or keep, and write one short reason based on the two questions from lesson 7.3: how much does it matter if the output is wrong, and does doing it yourself keep a skill you need?
A reason should be specific enough that someone else could see why you chose it. "Low risk" is too thin. "Errors would be obvious and it's only for me" tells you something.
Don't worry about getting every one perfect. If a task could go either way, put it in the stricter group for now. You can move it later.
For every task in the assist group, write the specific check you'll do before the output is used. This is the most important step, because it's where lesson 7.1, Why we trust confident machines too much, gets turned into habit.
A good check names what you compare against what. "Compare every figure to the source spreadsheet" is a check. "Review it" isn't. Other examples: read the client's original email again and confirm every question has been answered; check every name, date and amount against the booking system; run the code on test data before deploying.
If the output leaves your team, also write who the named reviewer is.
Here's part of the map Siew Ling, the accountant from lesson 7.1, made. Her tasks are examples.
Task one: reformatting the trial balance export for the monthly pack. Delegate. Reason: errors would be visible immediately when the totals don't match.
Task two: monthly variance commentary for managers. Assist. Reason: it goes to the finance director and errors could mislead decisions. Check: compare every direction word and figure to the variance table. Reviewer: Siew Ling.
Task three: replying to supplier queries about payment dates. Assist. Reason: a wrong date could damage the relationship. Check: confirm each date against the payment run schedule before sending.
Task four: deciding which overdue customer accounts to escalate. Keep. Reason: it's a decision about customers and involves judgement she's built over years.
Task five: reviewing the bank reconciliation. Keep. Reason: final numbers, high stakes, and it's the core skill that lets her spot errors everywhere else.
Task six: summarising new accounting standard updates. Assist. Reason: she needs to know them accurately. Check: read the official summary from the standard setter for anything the AI summary says affects the company.
Her by-hand task for the week: writing the first draft of the variance commentary without AI every Monday of close week, then comparing it with what the assistant would have produced.
Pick one task from your keep group, or one assist task whose underlying skill you identified as fading in lesson 7.2, Skills you lose when you stop practising them. Commit to doing it once a week entirely without AI.
Put it in your calendar as a recurring event and name the task in the title, so that on a busy Monday you can see exactly what you promised yourself rather than a vague word like "practice".
By the end, your map has ten rows, each with the task, its group and a one-line reason. Every assist row carries a written check, plus a named reviewer wherever the output leaves the team, and one recurring by-hand task sits in your calendar for the coming weeks.
Siew Ling found that three of her ten tasks had been sitting in the wrong group. She'd been treating two assist tasks as delegate, sending them on after a glance, and was using AI for the reconciliation review she now keeps. Your map may turn up something similar. Open your calendar and sent folder now and begin the list.
Complete a ten-task map with groups, reasons and checks, and schedule one weekly by-hand task in your calendar.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).