Run your day · Lesson 6.2
A Seven Day Inbox Trial: What to Track and What to Expect
Does an AI sorted inbox work outside a demo? The only fair way to find out is to try it on your own mail for a week and measure. This post is a plan for that week. It tells you what to set up, what to write down, and how to judge the result.
The trial follows Lesson 6.2, A clean inbox every morning.
Measure two things before you start: how long your morning inbox takes, and how many important emails you reply to late. Then compare after seven days.
Before day one: the baseline
Spend two ordinary mornings working as you always do. Write down:
- Minutes from opening mail to finishing the first important reply.
- How many emails you replied to late or nearly missed in the last week.
- How you feel about your inbox, from one to five.
These three numbers are your baseline. Without them you will only have an impression, and impressions are easy to fool.
Day one: set up
Write your list of what counts as important, in plain words. Set up three buckets: Needs me, Read later, Archive. Give the AI your list and ask it to sort overnight mail, flag the five that need a real answer, archive the noise without deleting, and write a short summary.
Do not let it send anything. Not this week.
Days two to four: review hard
Each morning, read the summary and check the sort before you act on it. Record:
- How many messages it put in the wrong bucket.
- Whether anything important was missed.
- Minutes from opening mail to your first important reply.
Expect mistakes. The first sort is usually rough. Each mistake tells you something about your list, so fix the list the same day.
Days five to seven: loosen the review
If mistakes are rare, check the flagged list and spot check the archive instead of reading every sort. Keep recording the same three items.
Look for patterns. Does it mishandle one sender? Does it flag too many from one client? Add a line to your list for each pattern.
What usually goes wrong
The list is too vague. "Important clients" means nothing to an assistant. Name them.
Too many flags. If it flags twelve, your list is too wide. Narrow it.
Quiet misses. An important email lands in Archive. This is why you check the archive and why nothing should be deleted during the trial. In Gmail, deleted mail goes to Trash and is permanently removed after 30 days, so archiving is the safer choice.
Trust arrives too early. If you stop reviewing after day two, you will find the first big miss at the worst time.
How to judge the result
At the end of day seven, compare to your baseline:
| Measure | Baseline | Day seven |
|---|---|---|
| Minutes to first important reply | your number | your number |
| Late or nearly missed replies | your number | your number |
| Feeling about the inbox, 1 to 5 | your number | your number |
| Sort mistakes per day | not applicable | your number |
Ask three questions. Did the time drop? Did late replies drop? Did the mistakes fall across the week? If time dropped but misses rose, the routine is not safe yet. If both improved and mistakes are falling, keep it and extend your trust slowly.
What it might cost
I cannot give you a cost, because it depends on the tool and plan you use, and on whether you already pay for it. Write down the price and your set up time, and decide whether the saving you measured covers it.
When to stop
If after a week the sort is still wrong often and your list will not fix it, stop. Some inboxes are mostly unusual one off messages that need human judgement. That is a real finding, and it saves you from a tool that does not suit you.
What to do after the trial
When the week ends, write three lines: what you will keep, what you will change, and what you will stop. Then schedule a second trial week a month later, because inbox habits drift. Mail volume changes, new senders appear, and a list that worked in November may be out of date by February.
Questions people ask
Is seven days long enough?
It gives a fair first read, since it covers a full working week. Run a second week before you rely on it.
What if my mail volume changes during the week?
Note it. A quiet week makes any routine look good, so try to test during a normal one.
The lesson that teaches this set up is coming soon. See the lesson page to join the waitlist.
An illustrative week, day by day
Say you run a small design studio. This is an example of how a week might look, with invented figures chosen only to show how to read them. They are not real results.
On the baseline mornings you note that it takes about forty minutes to reach your first important reply, and that two clients were answered a day late in the last week. Your feeling about the inbox is two out of five.
On day two the AI puts a client message about a deadline in Read later. You add the client's name to your list. On day three it flags eight emails, not five, so you tighten the list. On day four it gets one wrong. On day five you spot a pattern: it mishandles forwarded messages, so you add a rule about forwards.
By day seven, say the time to first reply is lower and no replies were late. You would still not call it solved. You would run a second week, because one week is a short sample.
The point of the example is the reading, not the figures. Time and late replies are the two numbers that matter. Mistakes per day tell you whether to trust the sort.
A daily log you can copy
Keep a simple note each morning:
- Date and number of new messages.
- Minutes to first important reply.
- Sort mistakes found.
- Anything missed.
- One change you made to your list.
A week of this note is more useful than any opinion about whether the routine works.
Related reading
If you want the routine behind the trial, read the Sort, Flag, Clear framework. If you are unsure whether to try it at all, read by hand versus with AI. The matching lesson is Lesson 6.2.