Tagging conversations: a tag list a small team will actually use
A starter list of ten, the five rules that keep tags meaningful, the synonym problem that quietly destroys most tag lists, and what to do with a label whose count is zero.
Sohail Safdar · · 5 min read
A practice has forty tags. Three of them mean price. Two mean the customer was unhappy, and nobody can say which is the worse one. There is a tag called "jane" that somebody made in March for a reason nobody remembers, and a tag called "URGENT" alongside one called "Urgent" that are different tags and filter differently.
Nobody did anything wrong here. Every one of those tags was created by a helpful person solving a real problem in the moment. That is precisely how the list ends up unusable, and why the rules below are mostly about restraint rather than taxonomy.
A starter list
| Tag | Meaning |
|---|---|
| New customer | First conversation with this person |
| Booking | Wants to book or reschedule |
| Pricing | Asked what something costs |
| Billing | Question or problem with a charge |
| Complaint | Unhappy with something that happened |
| Urgent | Needs a response within the hour |
| Insurance | Plan or coverage question, where relevant |
| Handed off by AI | The assistant passed this to a person |
| Follow up | Needs a later action |
| Spam | Not a real customer |
Adjust the names to your trade. Keep the count near ten, and resist the eleventh for a month.
Five rules, and the one that matters most
- Write the meaning down where the team can see it. A tag without a written meaning is a tag with five private meanings.
- One owner adds, renames and retires. Not a committee, and not everybody.
- No near-duplicates. Price, pricing and cost are one tag.
- Tag as you act, not later from memory. Retrospective tagging is both slower and less accurate.
- Retire what nobody uses, monthly.
Rule three is the one that quietly decides whether any of this works, and it is the one teams skip because each individual synonym seems harmless. It is not harmless, because the damage is invisible: you filter on "pricing", get eleven conversations, and act on the number. The twenty-three filed under "price" and "cost" are not reported as missing. A tag list with synonyms does not fail loudly, it just answers your questions wrongly for a year.
Counting is the point
Once a month, count conversations by tag and look only at what moved.
A jump in billing questions in the week after you changed an invoice layout is a fixable cause. A jump in pricing questions about one service usually means that service's price is missing from the facts your assistant can quote, so every enquiry becomes a conversation. A rising "Handed off by AI" count is either a knowledge gap or a sign the escalation rules are too eager, and the transcripts tell you which within ten minutes.
This is the whole return on tagging. Tags that are never counted are decoration applied at thirty seconds a thread.
What to do with a label whose count is zero
Worth a paragraph because the instinct is wrong in both directions.
In Frontiva's inbox, the folder row across the top shows a chip per folder with its count, and folders with nothing open in them are not drawn at all. Nine chips, most of them reading zero, took a full row of a screen whose job is to show a queue, and pushed the one number anybody cared about, the total, behind eight zeroes. A filter that can only return nothing is not a choice somebody is weighing.
But they are not silently dropped either. The row ends with a quiet "8 empty", because somebody who knows there is a Billing folder and cannot see it should be told it is empty rather than left wondering whether the feature was removed. And the folder you are standing in stays visible even when you have just emptied it, so the chip does not vanish from under the cursor while you are reading.
The same judgement applies to your tag list. Hide or retire what is empty, so the list stays readable. Say that you did, so nobody searches for a tag that quietly disappeared.
Tags, folders and fields
A folder is where a thread lives and it holds one at a time. A tag is a label and a thread can carry several. A field is a fact about the customer rather than the conversation: their plan, their stylist, their vehicle.
The common mistake is putting a customer fact on a conversation as a tag. It will be right on the thread where you added it and absent from every other conversation that person ever has. See tags vs fields vs pipelines.
Telling whether it is working, after a month
Two numbers, both cheap to get.
Coverage. What share of conversations carry at least one tag? There is no correct figure, but the shape tells you something. Under about one in five and the habit has not taken, usually because tagging happens after the reply rather than during it. Over about four in five and somebody has decided everything must be tagged, which means a lot of the labels are reflexes rather than decisions, and the counts you filter on are softer than they look.
Disagreement. Take ten tagged conversations into the monthly review and have two people say, independently, which tags they would have applied. If they match on eight, the written meanings are doing their job. If they match on four, you do not have a tagging problem, you have a definitions problem, and adding more tags will make it worse.
The second test is the one teams skip and the one that finds the rot, because a tag list degrades silently. Nobody reports that they have started using Complaint for mild irritation. The counts simply drift, and the monthly review keeps reading them as though they mean what they meant in January.
In Frontiva
The unified inbox carries tags, priority, assignment, snoozing and the nine folders, and a filtered view you use often can be saved so it is one click rather than four. Contacts have their own tags in the CRM, which is the right home for anything true about the person rather than about one conversation.
The design principle worth stealing, whatever you use: a label earns its place by being acted on. Everything else is a row between your team and the queue. See shared inbox best practices.