Run your voice guide, bullet drafting and one editing pass on five real pieces, and time each against your baseline.
?< Does this count towards the piece's time? | Count it / Leave it out Before the assistant | Writing your bullet points > Count it => Your old times included the thinking, so your new times should include it too. While it drafts | Waiting for the draft to appear > Count it => Without the assistant you'd have spent that time typing, so it belongs in the total. Mid-draft | A 20-minute call pulls you away > Leave it out => Pause the timer while you're on the call, because it would have happened anyway. After the draft | Putting back the names you took out > Count it => It's part of getting the piece ready to send, so it goes in the time. Before send | Reading the whole piece once more > Count it => This last read is usually where a wrong date gets caught. Two days later | Reading their reply > Leave it out => The timer stops when you press send, so anything after that isn't part of the piece. Before you start | Choosing which five pieces to test > Leave it out => Picking the pieces is set-up you do once, not part of any one piece.
---
[[flow Timer on | Bullets | Draft | Edit pass | Fixes and final read | Send, timer off]]
Your baseline is how long the same kind of piece takes without AI. If you haven't got one, time two pieces the old way first.
---
Include a routine reply, an ask, and a hard one about a delay, refusal or complaint. With only easy ones, the results look better than they are.
Remove client names before drafting and restore them before sending.
---
For each piece, ask whether the draft could have gone out under your name without anyone noticing.
If you rewrote the opening because it sounded nothing like you, that's a no, even if the version you sent was fine.
---
?? The draft needed one word changed but had moved a meeting from Tuesday to Wednesday, which you caught on the final read. How do you log voice? No, because the draft had a mistake + Yes, and note the date fix under edits Yes, and leave the date out of the log => The voice question is only about whether the draft sounded like you. The wrong date is a fact edit, so it goes in the log under edits. Microsoft says AI answers aren't guaranteed to be factual, which is why the final read stays inside the timer.
---
| Email | Minutes | Sounded like you | |---|---|---| | Update | 9 | Yes | | Request | 11 | Yes | | Follow-up | 8 | Yes | | Complaint | 15 | No | | Reply | 10 | Yes |
Five would have taken 70 minutes and these took 53, about a quarter less, with only the complaint running over.
---
[[rewrite We're so sorry, and we sincerely apologise for any inconvenience caused. | Sorry it's late. You'll have it Thursday by 5pm. | Apologise once]]
Add that to your voice guide. A run of noes means it needs work, and heavily edited yeses mean thin bullets.
---
If you don't have a voice guide yet, paste three emails you've sent and add "write like these".
?# Your first draft request My voice guide: {paste your voice guide} Draft a {reply, request or update} to {who it's for} from these points: {your bullet points} Keep it under {how many} words.
---
[[order Every piece, in this order | Start the timer before the assistant opens | Write your bullet points | Draft with your voice guide | Run one edit pass and pick the changes | Put back redacted details and read it once | Send and stop the timer | Log minutes, edits and voice yes or no]]
This week, run five real pieces of work through the steps and log each one in the spreadsheet next to your baseline, recording the minutes from start to send, the edits, and voice yes or no.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).