Why tag chats at all
An untagged transcript archive is a record you can read but not measure. Tagging — attaching a short category to each conversation — is what lets you answer questions like “how much of our volume is billing?” or “are shipping complaints rising?” without reading a thousand chats by hand. Tags are the difference between having transcripts and having data.
Keep the tag set small and consistent
The most common tagging mistake is too many tags. When agents face a list of fifty options they either pick inconsistently or skip tagging altogether, and the data becomes noise. Start with a short, mutually understood set — a handful of topics that map to real decisions — and only add a tag when you know what you would do with the count. A small, disciplined taxonomy beats an exhaustive one nobody applies the same way twice.
Tag at the end, not the beginning
Ask agents to tag a chat as they wrap it up, when they actually know what it was about, rather than guessing from the opening line. Tagging at close takes a few seconds, keeps the category accurate, and doesn't interrupt the flow of helping the customer. If tagging feels like a chore that competes with the conversation, agents will do it badly — so make it the last, quick step of the interaction.
A tag count is only worth collecting if it changes something. A spike in one category is a signal to fix the underlying cause — a confusing page, a missing help article, a broken process — not just to staff up. Review the distribution on a regular cadence and let it drive where you invest: the categories that dominate your queue are your product and content roadmap, written by your customers.
How MyLiveChat fits
MyLiveChat lets agents tag conversations and keeps those tags with the transcript, so you can see how your volume breaks down over time. Pair it with a review habit and the tags stop being decoration and start telling you where to spend your next hour of improvement.
Design the tag set backwards from the decisions
Most tagging schemes fail because they were designed as a description of everything that could
happen rather than as an input to decisions someone actually makes. Start from the other end:
list the questions you want the data to answer, then create the smallest set of tags that
answers them.
In practice that is usually a handful of questions. What are people contacting us about? Which
of those are avoidable? Which product areas generate the most confusion? Which conversations
turn into revenue? A tag that serves none of these is overhead on every chat and will be applied
inconsistently within a month.
Keep it small, flat and mutually exclusive
Ten to fifteen tags is a workable range for most teams. Beyond about twenty, agents stop
choosing carefully and start picking whichever plausible option appears first, at which point the
data is worse than none because it looks authoritative.
- Write definitions, not just labels. Everyone knows what billing means until two agents tag the same chat differently.
- Make categories distinguishable. If two tags overlap, the split between them is arbitrary and both become unreliable.
- Avoid deep hierarchies. Three levels of nesting guarantees the lower levels are unused.
- Prefer one primary tag for the reason, with optional secondary tags for detail. Chats tagged with five things cannot be counted.
- Include an other and read it. A growing other pile is the signal that the set needs a new category.
Make tagging quick or it will not happen
Tagging competes with the next conversation in the queue, and it loses whenever it is
awkward. Keep it to a single click at the end of the chat, put the common tags first rather than
alphabetically, and never require free text where a choice will do.
Be careful about mandatory tagging. Forcing a selection guarantees a value in every row but
not a correct one — under pressure, agents pick the top option. If you do make it
required, keep the list short enough that the right answer is genuinely easy to find, and audit
occasionally to see whether one tag is suspiciously popular.
Audit consistency, or the trend is fiction
Tag data drifts. New agents interpret definitions differently, categories blur as the product
changes, and a reorganisation quietly changes what a label means. Since most of the value is in
comparing this month with last, drift is the main threat to the whole exercise.
A light quarterly check is enough. Take twenty recent chats, have two people tag them
independently, and compare. Substantial disagreement means the definitions need tightening, not
that the taggers were careless. At the same time, review whether any tag has stopped being used,
whether other is growing, and whether a category has become so large it no longer
distinguishes anything.
Tagging only pays for itself at the point where a number causes a change. Build the loop
explicitly: review the distribution monthly, pick the two or three largest avoidable categories,
and assign each an owner and a fix with a date — a page rewrite, a navigation change, a
product change. Then check the following month whether the volume actually moved.
Share the results back with the agents who did the tagging. A team that sees its tagging
produce a fixed page tags carefully; a team that suspects the data goes nowhere stops trying,
and the quality collapses quietly.
What to measure
Watch the distribution across tags and how it shifts over time, with an eye on concentration
— a few categories carrying most of the volume is good news, because a small number of
fixes can move a lot of contacts. Track tagging coverage, since a metric computed on half the
chats is not representative. And keep an explicit list of changes made because of the tag data;
if that list is empty after a quarter, the scheme is costing agent time and returning nothing.
Alongside the tags your team applies by hand, the console derives a subject of its own. The topic badge your console puts on a chat covers how that automatic label is scored, and why it is a prompt to tag rather than a replacement for tagging.