The short version
- Sort by consequence, not by arrival time.
- Three buckets: answer now, draft and approve, needs you personally.
- Sweep in ten-minute blocks; do not read the whole inbox first.
- Write macros that sound like a person, or don't use macros.
- Track first response time, reopen rate and the reason codes.
Sort by consequence
A refund dispute, a safety question and a ‘where is my order’ are not the same kind of thing, even when they arrive minutes apart. Triage by what happens if it waits: money, safety and legal risk first; logistics next; praise and feature requests last.
The three buckets
- Answer now — routine, answerable from a template plus one fact.
- Draft and approve — needs a judgement call: a discount, an exception, a delicate tone.
- Needs you — complaints with legal or safety implications, VIP customers, anything you would not want a new hire to send.
The ten-minute sweep
Do not read everything and then start. Sweep once, classifying only — no replies. Then clear the ‘answer now’ bucket in one pass. This beats a chronological slog because the risky tickets surface before you are tired.
Macros that sound like a person
A macro should carry the facts (order number, dates, next step) and leave room for one human sentence. If a customer can tell it is a template, it is costing you trust — usually more than it saves in time.
Escalation rules
Write down the triggers: refund above X, any safety claim, any mention of a regulator, repeat contact on the same issue. When a trigger fires, the ticket goes to a named person with a deadline, not into a queue.
Measure the right three numbers
First response time, reopen rate, and reason codes. Response time is expectation; reopen rate is quality; reason codes tell you what to fix upstream so the ticket never exists — which is the only permanent win.
Checklist
- Consequence-based priority, documented
- Three buckets with named owners
- Ten-minute classification sweep, no replies
- Templates with one human sentence
- Written escalation triggers
- First response, reopen rate and reason codes tracked
If you would rather hand this over
AMAN's scribe (Katib) drafts replies against your policies, escalates anything matching your red lines, and returns the inbox as a queue of approvals — 36 drafted, 5 escalated, in the example on the home page. Nothing sends without you.
Questions people actually ask
How fast should a small store reply?
Same business day is a fair baseline; under four hours during peak. Publish the promise and keep it rather than aiming for a number you cannot hold.
Should support be automated?
Classification and drafting, yes. Sending without review, no — the cost of one wrong refund or tone-deaf reply outweighs the saved minutes.
What are good reason codes?
Five to seven categories you can act on: delivery delay, sizing, damage, refund, product question, complaint. More than that and nobody uses them consistently.