Every message looks identical on arrival, so each one gets opened just to be sorted. Here is how to remove that cost, and the line between classification and suppression.
The cost of a shared or executive inbox is rarely the volume of mail. It is that every message looks identical on arrival, so each one has to be opened to be sorted, and that sorting cost is paid dozens of times a day by somebody expensive.
Rules first, model second, and never suppression
Most of what arrives can be sorted deterministically: sender, domain, subject pattern. A client domain is a client. A notification address is a notification. These need no intelligence and should never reach a model, because a rule is free, instant and always right.
A model earns its place on the residue: summarising a long thread, drafting a reply for editing, or suggesting that something reads as urgent. What it must never do is decide something does not need a human.
A misfiled message is recoverable in seconds. An auto-archived client complaint is not recoverable at all.
How to build it
1. Write down what actually arrives, for one week
Not categories you imagine. Tally every message into buckets. Most people find four or five account for the overwhelming majority, and two of those are things nobody needs to read individually.
2. Route notifications out of the inbox entirely
System alerts, tool notifications, newsletters. These should never touch the main inbox. They go to a folder read deliberately on a schedule, or they are unsubscribed. This step alone usually removes the largest share of volume.
3. Sort the rest by sender and domain with plain rules
Client domains, vendors, internal, recruiters. Label and file automatically. No model involved, no ambiguity, no cost.
4. Add a model only for summarising and drafting
A long thread summarised at the top, or a suggested reply presented as a draft. Both keep the human in control. Both are optional, and both should be easy to turn off if they stop being useful.
5. Never automate archiving, deleting or replying
This is the line. Classification can be occasionally wrong because the cost is a moment of confusion. Suppression cannot be wrong at all, because the cost is a client who thinks you ignored them.
6. Give the executive three buckets, in order
What needs them today, what needs them this week, and what they may want to know. Anything outside those three was handled or does not need them, and that boundary is the actual product.
Tools and what they cost
| Option | What it costs | Honest trade-off |
|---|---|---|
| Native rules and filters (Gmail, Outlook) | Included. | Free, instant and reliable for the deterministic majority. No summarising or drafting. |
| Apps Script over Gmail with an optional model call | Free with Google Workspace, plus small model usage. | Full control including thread summaries at the top of a message. You build and maintain it. |
| AI email assistants (Superhuman, Shortwave and similar) | Typically tens per user monthly. | Polished summarising, drafting and triage suggestions. Verify exactly what they do automatically versus what they suggest, because that distinction is the whole risk. |
| A second inbox for notifications | Free. | Crude and remarkably effective. Costs nothing and removes the largest volume category immediately. |
What it is actually worth
Measure the sorting, not the reading. For one week, count how many messages you open purely to work out what they are. That number, times the seconds each takes, times fifty weeks, is what deterministic routing removes.
The larger effect is on interruption. Research published in Harvard Business Review in 2022, conducted with Soroco, observed rather than surveyed knowledge workers and found they toggled between applications around 1,200 times a day, spending just under four hours a week re-orienting after those switches, roughly nine percent of the working year. The sample was 137 users across twenty teams at three firms, which is small and is worth stating. Inbox fragmentation is one contributor to that toggling, not all of it.
I have deliberately not used the widely quoted claim that workers spend 28 percent of the week on email. It traces to a McKinsey Global Institute report from 2012, it is frequently misquoted as a share of the day rather than the week, and it predates most of the tools people now use.
How it breaks
Rules bury something important. A client emails from a personal address and lands in the wrong folder. Review folders weekly for the first month and keep an exception list of individuals who always reach the main inbox.
A model auto-archives. The failure this article exists to prevent. Check what any tool does automatically before enabling it.
The notification folder is never read. That is usually fine and occasionally not. Some notifications matter, so put the ones that do through the alerting path instead, where they get an owner.
Drafts get sent unedited. A suggested reply is a starting point. If people send them unchanged, the tone drifts and clients notice before you do.
How to tell whether it worked
Messages reaching the main inbox per day, which should fall sharply. Time to first response on genuine client mail, which should improve because the signal is no longer buried. And the number of messages misfiled and found late, which should be near zero and is the safety measure.