Reply Labeling

Why it matters for outbound teams

Consistent labels are what make a reply queue searchable and reportable; without them, a team cannot easily answer basic questions like how many meetings were booked this week or how many prospects asked to be removed. Labels also power lead routing, since a reply tagged as interested can be automatically sent to the right rep or CRM stage. Inconsistent or manual labeling tends to break down as reply volume grows across multiple sending domains and reps.

How reply labeling works

Every reply gets one primary label that describes what the sender wants. A label is only useful when it triggers something: a filter, a report line, an alert or an automation. That is the test for any label you add: what happens when a reply gets it?

Labeling can be done by a person, by rules, or by a model. A person is accurate but slow. Rules, such as matching the words "out of office", are fast but narrow. A model reads the full message and decides, which handles the wide range of ways people write. Most teams end up with a model doing the first pass and a person fixing the occasional miss. The decision behind the label is intent classification.

How to set it up

Start with a short list and grow it only when you have to.

• Positive group. Interested, meeting request, asked for pricing, asked for more information.
• Neutral group. Not now, referral to someone else, wrong person, question.
• System group. Out of office, auto-reply, bounce.
• Negative group. Not interested, unsubscribe, do not contact.

For each label, write one line that says what it means and one line that says what happens next. Keep the definitions short enough that two people would label the same reply the same way.

Master Inbox ships thirteen default labels, and you can use your own taxonomy instead, running on your own OpenAI, Anthropic or OpenRouter key. Every email and LinkedIn reply is labeled by intent as it lands. See AI labeling for how the labels are applied and edited.

A worked example

This is an illustrative example with round numbers, not measured data. An agency receives 100 replies for one client in a week. After labeling: 12 interested, 6 referrals, 10 not now, 40 not interested, 8 unsubscribe requests, 24 out of office or auto-reply.

The labels now drive the week. The 12 interested replies go to a rep and a Slack alert. The 6 referrals get a new contact added to the sequence. The 10 "not now" replies get a reminder in 60 days. The 8 unsubscribes are suppressed so they are never emailed again. The 24 system messages are closed without anyone reading them. The client report shows these counts directly, with no manual tallying. This is how labels feed lead routing.

Common mistakes

• Labels that overlap. If "interested" and "meeting request" mean nearly the same thing to your team, merge them.
• Labels with no action. A label that triggers nothing is clutter.
• No place for the unexpected. Keep an "other" label and review it weekly, so the model and the team learn new patterns.
• Never reviewing. Spot-check a sample of labels each week. A few corrections early keep accuracy high as volume grows.
• Forgetting LinkedIn. Use the same label set across email and LinkedIn so reports compare cleanly. LinkedIn Inbox applies the same labels.

When a new kind of reply appears often, such as "send this to our procurement team", add a label for it and tell the team. A label set that grows slowly from real replies stays useful.

Frequently asked questions

What is the difference between reply labeling and intent classification?

Intent classification is the decision about what a reply means. Reply labeling is the tag that records that decision on the message. In practice the terms are close, and the label is what you filter, route and report on.

How many labels should I use?

Ten to fifteen is enough for most outbound teams. Each label should map to a clear next step. If two labels lead to the same action, merge them, and if a label never triggers anything, remove it.

Can I use my own labels instead of defaults?

Yes, in most reply tools you can define your own taxonomy. Write a one-line definition and a next action for each label, and test the set on a few hundred past replies before you switch it on for live traffic.

Should a person check automated labels?

Yes, at least at the start. Review a small sample every week, correct mistakes, and add labels when a new type of reply keeps showing up. Accuracy improves with a short feedback loop.

← All glossary terms