An agent holding one conversation always knows who is waiting. An agent holding four does not, and the one who has been waiting longest is by definition the one who is not on screen. That is the entire problem the wait indicator solves, and it solves it with two small surfaces rather than a dashboard.
The first is a coloured age badge that appears on each chat tab in the agent console. The second is a floating pill across the top of the window that names the longest wait and jumps you to that conversation when clicked. Both are worth understanding literally, because each one measures something slightly narrower than its label implies.
The clock starts on the first message, not the last
When a visitor message arrives and there is no timer running for that conversation, a timestamp is recorded. If the same visitor sends three more messages before you answer, the timestamp is not updated. The age you see is the age of the first unanswered message.
This is the honest choice and it is worth defending, because the alternative flatters you. If each new message reset the clock, an anxious visitor typing every twenty seconds would appear to have been waiting twenty seconds, when in truth they have been ignored for four minutes and are getting more anxious with each unanswered line. The counter is measuring how long that person has been without an answer, which is the number that matches their experience.
Only real conversational messages start it. System notices, announcements and emotion events are skipped, as is anything replayed from history when a chat is reopened, so the badge does not light up for traffic the visitor is not actually waiting on.
Nothing at all for the first thirty seconds
Below thirty seconds the badge is not merely grey — it is removed from the tab entirely. There is no counter ticking up from zero.
That is a deliberate restraint. Normal back-and-forth in a live chat involves constant sub-thirty-second gaps, and a widget that flashed a timer on every single message would be ignored within a day. The instrument only speaks when the gap has grown past the length of ordinary conversation. The corollary is that its silence means nothing: a visitor who has been waiting twenty-nine seconds looks exactly like a visitor with no pending message at all.
Yellow, orange, red
Three thresholds, and they are the plain ones: thirty seconds turns the badge yellow, sixty turns it orange, and two minutes turns it red and starts it pulsing.
The age text next to the colour has its own formatting rule that catches people out on long waits. Under a minute it counts seconds. Under ten minutes it shows minutes and seconds. Past ten minutes it shows minutes only — and it never converts to hours. A visitor who has been sitting unanswered for three hours reads as 180m. If you ever see a three-digit number in that badge, translate it in your head, because the display will not help you feel how bad it is.
The pill counts stale chats, not waiting visitors
The floating pill says something like 2 visitors waiting. That count only includes conversations that have already crossed the thirty-second line.
So the pill is not a count of unanswered visitors. It is a count of unanswered visitors you are arguably already late for. Three people could have messaged you in the last twenty seconds and the pill will read zero and stay hidden. This is consistent with the badge rule and it is the right behaviour for an alerting surface, but it makes the pill useless as a workload gauge — it is an exception report, not an inbox.
The time shown beside the count is the single longest wait, not a total or an average, and clicking the pill selects that conversation. On a busy shift, clicking the pill repeatedly is a reasonable work queue: it always takes you to the person who has been ignored longest.
The beep you get once, and can spend while muted
Next to the pill is a small bell button. Turned on, it plays a short tone when a chat reaches the two minute mark — a brief eight hundred and eighty hertz note, a few tenths of a second, generated in the browser rather than loaded as a file. The preference is stored in your browser and is off by default, so nobody gets an unexpected noise on their first shift.
There are two things about it that surprise people, and both follow from the same design.
Each conversation can only ever chirp once. Crossing two minutes marks that chat as chirped, and no later escalation for the same visitor makes another sound. The intent is obvious: an alert that repeats every second for a chat you already know about is an alarm you will mute permanently.
The subtler one is that a chat is marked as chirped whether or not the sound was actually enabled. If the bell is off when a conversation crosses two minutes, that chat has spent its one beep in silence — switching the sound on afterwards will never produce a tone for it. Turn the bell on at the start of a shift, not in the middle of a bad one.
One more detail: only the single longest-waiting chat is a candidate to sound at any moment. If two conversations cross the line together you hear one tone, not two.
What stops the clock, and what does not
Your own reply stops it. So does the conversation being accepted or transferred away, which drops the timer immediately rather than leaving a badge on a chat that is no longer yours. So does the tab disappearing, which is swept up on the next tick.
Here is the one that matters on a team. Another agent replying does not stop your clock. The timer clears only when the reply comes from you. In a conversation where a colleague steps in without a formal transfer, your badge keeps climbing and your pill keeps counting a visitor who has in fact just been answered. If your team works that way, treat the indicator as personal rather than shared, and use a real transfer when handing a conversation over.
What it will not tell you
Everything here lives in the browser tab you are working in. Nothing is stored on the server, nothing is aggregated, and a refresh starts every counter from nothing. Close the console and the history of who waited how long is gone.
That means the indicator is an operational aid, not a source of data. It is genuinely good at the question who should I answer next and it cannot answer how long did people wait this week. For that you want reporting built on the stored conversation record, not a client-side timer.
What to measure
Since the timers are not recorded, measure the behaviour they are meant to change. The number worth tracking is not average wait but the tail: the proportion of replies that took longer than two minutes. Averages hide exactly the cases this tool exists to surface, because ten fast replies bury one four-minute silence.
Pair that with your concurrency limit. A red badge appearing regularly is usually not an attention failure; it is a sign that the number of chats an agent is allowed to hold at once is set higher than the number a person can genuinely serve. The badge tells you which chat to answer. The rate of red badges tells you the limit is wrong.
Put it into practice
- Turn the bell on at the start of your shift, while nothing is waiting. It cannot retroactively alert you about a chat that already crossed two minutes.
- Read the pill as an exception report. Zero means nobody is late, not that nobody is waiting.
- Click the pill instead of choosing a tab yourself. It always points at the longest-ignored person.
- Treat any three-digit minute count as an emergency — the badge will not shout at you past red.
- Use a real transfer when a colleague takes over, or your timers will keep counting a conversation that has already been answered.
- If red badges are routine, lower the concurrent chat limit before asking anyone to try harder.