Tags

What is a Tag?

A tag is a label on a conversation: Returns, Delivery, Complaint, Wholesale. Simple on the surface, and the single most reused piece of configuration in the product, because tags are what the AI, the routing, your automations and your reports all read.

Tags are managed under Settings → Tags. Each one is more than a name.

Anatomy of a tag

• Name, what everyone sees.
• Allow AI to use this tag 🤖 whether the AI may attach this tag by reading the conversation. On by default. Turn it off for tags your team applies by hand, for example an internal workflow marker.
• AI hint 💡 a short line telling the AI when the tag applies. Optional, capped at 100 characters, and worth it for anything ambiguous. It is read on every conversation, so keep it tight: "customer wants to send an item back", not a paragraph.
• Risk level ⚠️ 0 to 100. See below, this one has teeth.
• Channels 💬 by default a tag is available everywhere. Restrict it when a tag only makes sense on one channel, for example "Voicemail" on phone.

Risk level: the auto-send brake ⚠️

Risk level is how you tell the system that a topic is too sensitive to answer without a human.

Set 50 or above and drafts on conversations carrying that tag are never auto-sent. The AI still writes the reply, but a person has to send it. Below 50 the tag does not gate anything.

This is the cheapest safety net in the product. Put a high risk level on Complaint, Legal, Chargeback, Press, and the AI keeps helping your team on those tickets without ever speaking on your behalf.

How tags get applied

• The AI, while reading the conversation. As part of enriching a new ticket it picks from your tags, using each tag's name and hint. Tags with AI enrichment off are never offered to it, and tags restricted to other channels are left out.
• Your team, by hand, in the helpdesk.
• Workflows, with a Tag block, so a branch can label what it handled.
• Automations, as the "set tag" action on a rule.
• Your IVR, where a call flow node can tag the conversation it creates.

Note: if your team has no tags at all, the AI creates a default set the first time it enriches a conversation. Fine as a starting point, but a list you shaped yourself is worth much more, because everything below reads it.

Where a tag changes behaviour

This is why tag hygiene pays off. One label reaches into eight places:

• Auto-send, through risk level, as above.
• Access, through capabilities: a rule can include or exclude tags, so a tag can define an agent's slice or keep them away from a topic.
• Queue order, through the flow queue: in Custom mode a tag can boost a conversation up the queue, which is the usual way to prioritise VIP or refund tickets.
• Policies, whose rules can require a tag before the policy applies.
• Workflows, which can branch on it and set it.
• Automations, which can trigger on it and set it.
• Phone routing, where an IVR choice tags the ticket it creates so the rest of the system can act on it.
• Reporting, where the tag breakdown is how you see what your customers actually write about.

Keeping the list useful

• Aim for 10 to 25 tags. Under ten and the reporting says nothing; over fifty and neither your team nor the AI applies them consistently.
• One dimension per tag. "Return" and "Urgent return" fight each other. Reason belongs in tags; urgency is already measured separately.
• Name them the way you would ask about them, because these names become your report rows.
• Write a hint for anything ambiguous. "Complaint" without a hint catches every mildly annoyed customer.
• Review the tag breakdown monthly. A tag that never appears is dead weight; a spike is usually a real operational problem.

Good to know

• Renaming a tag keeps it attached to every conversation that already carries it, and the AI starts using the new name immediately. The report rows follow along.
• A tag cannot be deleted while an IVR flow or a capability references it. The error names what is holding it. Tickets and analytics never block deletion, since those are history, not configuration.
• A conversation can carry several tags. For auto-send, the highest risk level on the ticket wins.
• Turning AI enrichment off on a tag does not remove it from past conversations; it only stops the AI applying it going forward.
• Restricting a tag to a channel hides it from the AI on other channels, so a phone-only tag never shows up on email tickets.