An inbox that files the routine and interrupts only for what matters
Every five minutes a Make scenario reads new mail in a business inbox and has Claude classify each message by urgency and category. Routine mail is marked read and filed under a label; mail that needs a person goes to a review label, with a Slack alert during working hours.
Where this applies
Owners and managers whose inbox mixes urgent client mail with receipts, notifications and newsletters, and who want to be interrupted only for the first.
- between inbox checks
- 5 min
- modules end to end
- 6
- runs, 17–21 Oct, one warning
- 699
The problem
A business owner's inbox mixes a client asking about tomorrow's job with supplier receipts, portal notifications and newsletters. Filters catch the senders they know, everything else arrives with the same weight, and the urgent email waits behind forty others.
Handing the whole inbox to an assistant is one answer. Most businesses want something narrower: the routine filed without being seen, and a nudge for the few messages that matter, without a phone going off at ten at night.
What we built
Small on purpose
The scenario has six modules: watch the inbox, ask Claude, route, then either file or flag. It runs every five minutes, so nothing waits long, and a poll with no new mail does no further work.
Claude classifies, the router decides
Claude returns an urgency and a category for each message, and the router does the rest. A message that needs review is moved to a review label; one that doesn't is marked read and moved to its label. The model reads; the routing is fixed and visible in the scenario.
Alerts respect working hours
A filter on the review route lets the Slack alert through only during active hours. Out of hours, urgent mail still lands under the review label, waiting for the morning, but nobody is pinged.
Labels as the audit trail
Every message ends up under a label, so the classification is visible in the mailbox itself, and anything misfiled can be found and moved by hand.
The outcome
In the History read, the scenario ran 699 times between 17 and 21 October 2026: 698 successful runs and one warning.
Where it stops
Claude's classification is trusted: a message it wrongly calls routine is marked read and filed without anyone seeing it, so the labels need an occasional look. This page covers one inbox. Digests, follow-up tracking, re-filing from Slack and the copies for other inboxes run as separate scenarios that are not described.
Built with
- Make.com
- Gmail
- Anthropic Claude
- Slack
Patterns used
- LLM classification, fixed routing
- Working-hours filter
- Labels as the audit trail
Page copy drafted with AI assistance. Skofi is accountable for it. Editorial policy
Have something shaped like this?
If any part of the above resembles a problem you are living with, a short call is the fastest way to find out whether it is the same problem underneath.
What happens on the call
- 15 minutes, no deck
- We map the process on the call
- A written scope, or an honest no