Early Warning Thresholds for Queues, Backlog and Aged Conversations

What this table is for

A WhatsApp queue doesn't break all at once: it degrades. First, unassigned conversations pile up, then assigned ones age without a reply, and only at the end does the customer write again to ask if anyone is going to help them. By the time that last step happens, you've already lost the sale and probably the contact too. Early warning thresholds exist to stop the degradation at step one or two, not at step four.

The problem is that most teams don't have thresholds: they have a feeling. "We're swamped," "there's a lot today," "we dropped the ball." That feeling can't be monitored, can't be delegated, and can't be automated. A threshold, on the other hand, is a number with a time next to it and an assigned action. This page is the table you build once and then check to know if the queue is healthy, if it's in the risk zone, or if you need to escalate.

Everything that follows are method thresholds: you plug in your operation's numbers and the table tells you which band you fall into. There are no market figures or invented benchmarks. The only hard references that appear are the ones that can be stated without an external source: WhatsApp's 24-hour window, the window of up to 48 hours for template approval, and the features that actually exist in an inbox tool with AI.

How to read the main table

The table has three axes that are measured separately and then combined: age of the unanswered message (the most important one), backlog size (how many open conversations there are in total), and overdue conversations (those that crossed a limit you defined and should no longer be open). Each axis has four bands: healthy, attention, risk, and escalate.

The "healthy" band doesn't mean there's nothing pending. It means what's pending is within the time your team can resolve it without the customer noticing the delay. The "attention" band is the signal that something is moving slower than normal and it's worth looking before it turns into risk. "Risk" is when there are already conversations that will end badly if nothing is done today. "Escalate" is when the problem stopped being about volume and became about capacity: it isn't solved by working faster, it's solved by adding people, changing the shift, or activating another channel.

How the age threshold is set, which is the one that matters most

Of the three axes, the age of the unanswered message is the one that best predicts whether the customer will leave. A large backlog with fresh messages may just be a busy day. A small backlog with six-hour-old messages is a structural problem. That is why the age threshold is set first and the others are derived from it.

The way to set it is not to copy a number: it is to look at your own history and see from what delay the customer starts writing again, asking if they saw it, or going elsewhere. That point is your risk threshold. The attention band is half that time, and the escalate band is double. If you don't have history yet, use the table as a starting point and adjust after two weeks of measuring.

An important detail: the clock does not run the same in all time slots. A message that comes in at 22:00 does not age while the business is closed. If you measure raw age, you will have false alarms every morning. The age that matters is the one that runs within the declared business hours. If your inbox has auto-replies by time slot, that same time slot is what defines when the clock starts counting.

What to do in each band, without ambiguity

A threshold without an assigned action is decoration. Each band has to have a concrete response and a responsible person. If you don't know who acts when the threshold is crossed, the threshold is useless.

1. 1Healthy zone: nothing is done. The team keeps its normal pace. If you want, log that it was measured, but do not interrupt the operation for a green indicator.
2. 2Attention zone: the shift lead checks the queue once and redistributes if needed. No one outside the team is notified. The goal is to keep it from turning into risk within the next hour.
3. 3Risk zone: shift reinforcement is activated. The oldest conversations are prioritized, not the newest. If there is multi-agent, overdue conversations are reassigned to whoever has the least load. If templates are available, one is used to reopen contact with overdue conversations.
4. 4Escalate zone: the decision-maker is notified (owner, manager, shift supervisor). Adding people to the shift, extending service hours, or activating an alternative channel is evaluated. What was done is documented, because if the escalate zone repeats three days in a row, the problem is not today's, it is staffing.

The most common mistake at this point is assigning the action to whoever is busiest. If the person handling conversations is the one who has to detect and resolve the threshold, the threshold is always crossed. Detection must be automatic or done by someone who is not in the queue: a supervisor, a manager, or an alert that arrives in a separate channel.

Overdue conversations: the definition that must be agreed on before measuring

"Overdue conversation" does not mean the same thing in every business, and that is where half of the dashboards fall apart. Before setting a number, three things must be agreed on: what counts as an open conversation, when it is considered overdue, and what closes it. If those three definitions are not written down, every person on the team will count differently and the indicator will be useless.

The closing limit is the number that generates the most discussion. Some teams close at 24 hours, others at 7 days, others never close. There is no universal correct value: there is a value that must be the same for the whole team and must be below the time in which the customer stops remembering that they wrote. If you close too quickly, you lose conversations that could still have converted. If you close too late, the backlog fills with noise and the thresholds stop alerting.

Common mistakes when setting thresholds

Almost all queue dashboards fail for the same reasons. It is worth reviewing them before publishing any indicator, because a badly set threshold is worse than having none at all: it generates alerts that the team learns to ignore.

• Setting the threshold at the average instead of the high percentile. If your average response time is 20 minutes but 10% of conversations take 5 hours, the average is hiding the problem from you. The threshold is set on the 90th or 95th percentile, not on the mean.
• Measuring raw age instead of age during business hours. Every morning you will see conversations that look overdue and actually aged while the business was closed.
• Counting the total backlog instead of backlog per agent. A backlog of 40 conversations with 8 agents is different from a backlog of 40 with 2 agents. The number that matters is the ratio, not the total.
• Not distinguishing between an unassigned conversation and an assigned conversation with no reply. They are two different problems: the first is about distribution, the second is about capacity. They are fixed in different ways.
• Using a single threshold for all hours. Morning and night do not behave the same. If you have enough data, set thresholds by time slot.
• Defining the threshold but not the action. We already said it, but it is the most repeated mistake: if nobody knows what to do when it is crossed, the threshold changes nothing.
• Never reviewing the thresholds. A threshold that was set a year ago and never adjusted probably no longer reflects the operation. They are reviewed every quarter or when the team size changes.
• Using the same threshold for all channels. WhatsApp, Instagram and Messenger can have different response times and different expectations. If you measure everything together, the slowest channel contaminates the fastest one.

How to know if the thresholds are working

A threshold system is well calibrated when three things happen: the alerts in the escalation band are few and all end in a concrete action; the risk band is resolved within the day; and the team does not argue about the numbers, it argues about what to do. If alerts fire every week and nobody does anything, the problem is not the threshold, it is that the threshold is not connected to a decision.

The sign that they are badly calibrated is the opposite: nothing ever fires. If in three months you have not crossed the risk band even once, either your operation is exceptionally relaxed or the thresholds are so high that they are useless. In that case it is worth lowering them until they start flagging something, even if it is uncomfortable. A dashboard that is always green does not inform: it decorates.

The other sign is reopening. If the reopenings indicator goes up, it means you are closing conversations that the customer still considered open. That happens when the closing limit is too aggressive or when a conversation is closed to clean up the dashboard and not because the conversation ended. Reopenings are the best thermometer for whether your definition of "closed" is honest.

Where a tool fits into all of this

Thresholds can be measured by hand with a spreadsheet, and many teams start that way. The problem is that by hand you measure once a week, and age thresholds are useful when they are looked at in the moment. An inbox with AI suggestions and multi-agent support helps in two specific points: first, because unanswered conversations stay visible with their age and do not get lost in the scroll; second, because suggested replies lower the first response time, which is precisely the variable that triggers the risk band.

Wando, for example, uses Meta's official WhatsApp API and suggests replies that the team approves with one click, learning from the business's real replies. That does not replace thresholds: it complements them. The threshold tells you when there is a problem; the tool helps you make the problem last less time. If you want to see what an inbox with those features looks like, you can try it, but you still have to define the threshold system yourself, with or without a tool. Plans and the free trial are at wando.online/pricing.

Template to copy: minimum thresholds spreadsheet

If you want to start today, this is the minimum structure of a thresholds spreadsheet. It is filled out once per shift and reviewed at the end of the day. It does not need more columns than these.

With ten days of this spreadsheet you already have enough to set your own thresholds with data instead of intuition. You will see what your typical age is at the high percentile, how many expired conversations appear on a normal day, and in how many shifts you cross the risk band. That is the moment to adjust the numbers in the first table and stop using the reference ones.

Examples by industry of how the threshold changes

The same method produces different numbers depending on the industry, because customer expectations change. These are examples of how the age threshold moves in typical operations, not Wando values or industry standards. They are meant to help you see where to start before you have your own data.

• Food and beverage: the customer writes while hungry and with the order already decided. The age threshold tends to be the shortest of all, because the decision window is measured in minutes. If you have the food and beverage landing page, that's where you'll find the order flow examples that combine with these thresholds.
• Beauty and health: the customer writes to book an appointment, not to buy on the spot. The threshold tolerates more hours, but not more than a day, because otherwise they go elsewhere. Scheduling and reminders matter more than pure speed.
• E-commerce and stores: the customer asks about stock, shipping or order status. The threshold is set mainly on the pre-sale inquiry, which is what decides the purchase. Post-sale inquiries tolerate more delay.
• Real estate: the cycle is long and the customer inquires about several listings at once. The age threshold can be measured in hours, but the re-engagement threshold is the critical one, because the contact who writes again is comparing options.
• Academies: inquiries arrive in peaks marked by enrollment periods. It's best to set thresholds by time slot rather than a single one, because the volume during enrollment week doesn't look like the rest of the month.

If your industry isn't on the list, the method is the same: look at how long it takes before your customers start writing again and use that point as a reference. What changes between industries isn't the method, it's the number.