Estimate whether an assistant saves you time on a task once you count prompting, checking and fixing.
?? If you could test an assistant on one task this week, which would you start with? The weekly report, 2 hours each week + Email replies, 40 a week at 6 minutes each Matching supplier invoices to deliveries, 90 minutes a month => Start with the emails. If each reply drops from 6 minutes to about 3, that's 3 x 40 = 120 minutes back a week. The report saves less, and the matching would lose time.
---
[[flow Time by hand | Minus prompting | Minus checking | Minus fixing | Time saved]]
Checking means reading it closely enough to put your name on it. Fixing covers the errors, the gaps and getting it into your manager's format.
---
[[big 50 minutes | saved a week on a 2-hour report: 10 prompting, 40 checking, 20 fixing]]
The saving is 120 - 10 - 40 - 20 = 50 minutes. That's worth testing, but it shouldn't be your first test.
---
[[versus The weekly report | 40 email replies | Each time: 50 min / 3 min > A | Times a week: 1 / 40 > B | A week: 50 min / 120 min > B]]
---
[[sum Writing and testing a good email prompt takes 30 minutes. Each reply then saves 3. After how many replies has it paid back? | 10 | 90 => That's 30 x 3. Divide the set-up by the saving. | 27 => You took the 3 off the 30. Divide instead. | => 30 / 3 = 10 replies, under two days of email at 40 a week.]]
For a yearly memo, those 30 minutes might not pay back before the format changes.
---
To trust an assistant's reconciliation, you'd need to compare every line with the invoices and delivery records yourself.
[[big -5 minutes | a month: 90 by hand, against 10 prompting and 85 checking, before any fixing]]
---
If a minute's reading would show you a mistake, the task is a good one to hand over.
When you'd have to redo it to find out, keep it yourself. A formula that matches records might help more there.
---
?! Would a minute's reading catch the mistakes? A summary of a meeting you sat through | Yes | No => You were there, so a wrong line would stand out. Totals matched across 300 invoice lines | No | Yes => You'd have to recheck every line to be sure. A short reply turning down a supplier's quote | Yes | No => It's short, and you know what you meant to say.
---
[[big Over 25% | faster, for 758 consultants on tasks inside what AI could do (Harvard Business School and BCG)]]
Those gains were measured on tasks that suit AI, so treat your own numbers as guesses until you've timed a real one.
---
[[order Put the steps for one task in order | Write down its time by hand | Guess its prompting, checking and fixing time | Subtract those three from the time by hand | Multiply by how often it comes up | Mark it if the result is a saving]]
Take five tasks from last week's calendar and note each one's time by hand in a spreadsheet, with your guesses for prompting, checking and fixing. Then subtract them, multiply by how often each comes up, and mark the savings.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).