Work out the net time saved on each task from your logs, counting waiting, checking and fixing, and label each result.
[[versus Timing the AI part | Timing the whole task | One supplier reply: 1 min / 4.5 min > B | Waiting for output: Left out / Counted > B | Checking and fixing: Left out / Counted > B]]
The left column feels true.
---
[[flow Baseline, before AI: 6 min | New way, start to sent: 4.5 min | Net saved: 1.5 min]]
Start to sent means prompting, waiting, checking and fixing. Waiting only drops out if you did other work while it ran.
---
[[guess You send 40 supplier replies a week. Each took 6 minutes before. Now your average is 4.5 and your best run is 3. How many minutes a week do you save? | 60 | Use the average: (6 - 4.5) x 40 = 60. Your best run would say 120, and a manager plans around the figure you give. | 120]]
---
| Task | Before (min) | Now (min) | A week (min) | Label | |---|---|---|---|---| | Supplier replies, 40 a week | 6 | 4.5 | +60 | saving | | Monday report, weekly | 120 | 45 | +75 | saving | | Meeting notes, weekly | 25 | 16 | +9 | saving | | Regional update, monthly | 60 | 65 | about -1 | loss |
Monthly tasks: divide by 4.
---
[[big 143 min | a week: 60 + 75 + 9, minus about 1]]
Leave your first learning-week runs out of the average and log them as setup. Slow runs after that stay in.
---
[[checklist How many are true of your logs? | Runs timed start to sent | Waiting counted, unless you did other work | Checking and fixing included | Average run, not best | Learning runs logged as setup | Monthly tasks divided by 4 | Losses kept in the total | 0-3 => Re-time a week of runs before you add anything up | 4-6 => Close. Fix the gaps, then add up | 7 => Your figure will hold up when someone asks where it came from]]
---
Try one change to the method. Still slower? Go back, and write why on its line, so nobody retries it next year:
Regional update, loss: the figures come in a format the assistant always misreads, and checking takes longer than writing by hand.
---
[[rewrite Meeting notes: 9 min saved a week | Meeting notes: 9 min saved a week. Missed action items before: 2 in 4 meetings. Since: 0 in 6. | Same minutes, plus what changed]]
Fewer edit rounds count too, and so do errors caught earlier.
---
?? Your weekly client update took 30 minutes before and still averages 30. Your manager now asks for one round of edits, not two. What's the label? Drop it from the report, since no minutes were saved + No change, with one fewer edit round noted A saving, because it feels quicker now => The label follows the minutes, and the quality note sits beside it. A task with no time saved can still be worth keeping.
---
[[order Your figures, in order | Open your baseline and logs side by side | Fix or drop any run that only timed the AI part | Average each task, learning runs left out | Multiply by how often, monthly divided by 4 | Label each: saving, no change or loss]]
With your baseline sheet and your logs open side by side, work out each task's net minutes a week from your average run, then label it saving, no change or loss.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).