Time three to five work tasks your usual way, with quality notes, so any saving from AI later is measured, not just felt.
?< Would this number hold up? | Holds up / Doesn't Colleague, a month into AI | It saves me about two hours a week, easily. > Doesn't => Nothing was timed before the change, so the two hours is a guess. Your log, week one | Monday report: 125 minutes, one interruption, manager fixed one figure. > Holds up => It was timed on the day, and the quality is noted beside the minutes. You, from memory | My weekly summary used to take about an hour. > Doesn't => Remembered times drift, so the real figure could be 40 minutes or 90. Your log, two weeks | Meeting notes: 25 minutes, then 28 minutes. > Holds up => Both runs of the same task were timed when they happened. Team chat | The AI draft came out in ten seconds. > Doesn't => That's only the draft, not the time spent fixing it. Your log, week two | Monday report: 150 minutes, phone rang four times. > Holds up => The interruptions are noted, so nobody will read this as 150 minutes of steady work.
---
You'll probably remember the ten-second draft, but not the ten minutes you spent fixing a client's name in it.
Once you've changed how you work, the old times get hazy too, so it's best to log them first.
---
[[mark Tap the three claims nobody wrote down | My report takes *about an hour* now, down from the *two hours I remember*. AI drafts it *in seconds*, so I'm saving loads. | All three come from memory. With nothing logged before the change, there's no number to beat.]]
---
[[timeline Your baseline | Day 1 Pick three to five tasks and set up the log | Day 7 Daily tasks have enough runs | Day 14 Weekly tasks have two runs | Day 15 Start trying AI on them]]
---
[[sum One morning's email replies: 9 sent in 55 minutes. How many minutes a reply, to the nearest minute? | 6 within 0.5 | 55 => That's the whole batch. Divide it by the 9 replies. | 9 => That's how many replies. Divide the minutes by it. | => 55 / 9 = 6.1, so about 6 minutes a reply.]]
---
| Task | Date | Minutes | Quality notes | AI used | |---|---|---|---|---| | Monday report | 2 Nov | 125 | 1 interruption, 1 figure fixed | No | | Emails, batch of 9 | 3 Nov | 55 | None sent back | No | | Meeting notes | 4 Nov | 25 | Missed 1 action item | No |
---
[[rewrite Report: 60 minutes | Report: 60 minutes, sent back twice | The send-backs are part of the cost]]
Log the edit rounds and any errors found later.
---
?? On day 5 you try a prompt on the morning's email batch, and it takes 30 minutes. What goes in the log? Log 30 minutes like any other run + Mark the row and leave it out of the average Average it in, since one run out of several barely moves it => A run with a prompt in it measures neither your usual way nor the new one. Keep the row, but mark it and leave it out.
---
[[flow Baseline weeks: AI used says No | Same tasks with AI: Yes | Compare the No rows with the Yes rows]]
If a task already involves AI, do it your old way for a week, or baseline one that doesn't.
---
?= Build your summary of a baseline Time three to five tasks your usual way for one to two weeks | Time every task you do for a month | One timed run of each task is enough Note interruptions, edit rounds and errors beside the minutes | Minutes are the only figure worth logging | Quality notes only matter once you use AI Log task, date, minutes, quality notes and AI used on every run | Log a weekly total and skip the single runs People remember the fast AI moments better than the slow fixes | Remembered old times are close enough to compare against | A strong sense of saving time counts as evidence
Open a spreadsheet and add five columns: task, date, minutes, quality notes, AI used. Fill in the first row the next time one of your chosen tasks comes up.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).