Decision matrix: tag, assign or escalate an incoming conversation

Three actions, one criterion: what to do with each conversation that comes in

Every message that comes into the WhatsApp inbox forces a decision in seconds: you leave it queued with a label, you assign it to a person, or you escalate it. When that decision is made out of habit, the same thing always happens: the urgent waits and the trivial consumes the most expensive person on the team. This page is the decision tree that orders the three actions with an explicit rule, so that anyone who opens the inbox —an owner, a salesperson, a new agent— reaches the same conclusion with the same message.

The three actions, defined without ambiguity

Before deciding, you have to establish what each action means. If the team doesn't share the definition, the matrix is useless: two people look at the same message and do different things.

The question that decides everything: can it wait without something breaking?

The matrix has a single entry question: if this conversation waits two hours, what breaks? If nothing breaks, it gets labeled and stays queued. If a specific commercial opportunity breaks, it gets assigned. If something breaks that the team can't resolve on its own —a claim with legal risk, a billing error, a customer who has already threatened to leave—, it gets escalated. The question is not how important the message seems, but what happens if no one touches it.

That formulation avoids the most common bias: treating everything that sounds angry as urgent. An angry customer because of a delay that is already resolved does not need escalation, they need a clear answer. A calm customer who asks for an exception to the company policy does need escalation, even if they write in a friendly tone.

Decision matrix: what to do depending on the type of conversation

The table crosses the type of message with the consequence of waiting and returns the action. It is read from left to right: first you identify the type, then you look at what breaks if it waits, and the last column says what to do.

How to read the matrix in practice

The most common mistake when applying a matrix like this is jumping straight to the action without going through the middle column. The 'if it waits 2 hours' column avoids two extremes: escalating everything just in case and labeling everything to avoid commitment. If the team cannot answer that column with a concrete sentence, the conversation is not classified yet.

• If the answer to 'what breaks' is 'nothing', the action is to label. The tone of the message does not matter.
• If the answer mentions a person waiting for something concrete, the action is to assign.
• If the answer mentions money, a legal commitment or a decision that the agent cannot make, the action is to escalate.
• If the answer is 'I don't know what breaks', the conversation is labeled as doubtful and is reviewed with whoever coordinates the inbox before assigning it.

Labeling well: a label is not a drawer, it is an instruction

A useful label says what to do with the conversation, not what it is about. 'Inquiry' tells no one anything. 'New inquiry' implies that it must be answered with pricing and availability information. The difference seems small and it changes the result: with descriptive labels, anyone on the team can take over the conversation without reading the entire history.

Assigning without creating bottlenecks

Assigning is the most costly of the three actions, because it consumes a specific person's time. That is why it should be reserved for conversations where there is an opportunity or a commitment at stake. Two practical rules: no one has more assigned conversations than they can answer during their shift, and every assignment has an expiration time. If the person does not respond within that period, the conversation returns to the general queue with a reassignment label.

Shift-based assignment avoids the problem of conversations left orphaned when someone goes off their schedule. If the business serves from 9:00 to 18:00, a conversation assigned at 17:30 must have a plan for after closing: either it is answered before, or it is reassigned to the next shift with an internal note. In Wando, schedule-based auto-replies cover the out-of-hours automatic message, but they do not replace that reassignment decision.

Escalating: when to do it and when it is a way of not deciding

Escalating has a hidden cost: every escalation consumes time from the level above and, if used too much, that level ends up answering everything and the team stops learning. The signal that you are escalating too much is simple: the person receiving the escalations is answering questions that any agent could answer. The signal that you are escalating too little is the opposite: claims are answered with different criteria every time, because no one with authority ever looked at them.

• Escalate when the answer commits money that the agent cannot authorize.
• Escalate when there is a legal threat or a claim before an external body.
• Escalate when the customer has already received three answers and the issue is still unresolved.
• Escalate when the correct answer contradicts a written company policy.
• Do not escalate just because the customer is angry.
• Do not escalate just because the agent does not know the answer: first you look for the information.
• Do not escalate a question that already has an answer in the template or in the frequently asked questions section.

The decision tree, step by step

1. 1Read the entire message before deciding. A message with two requests may need two actions.
2. 2Ask yourself what breaks if the conversation waits two hours.
3. 3If nothing breaks, tag it with the corresponding operational category and leave it in the queue.
4. 4If an opportunity or a commitment breaks, assign it to a person with a due date.
5. 5If something breaks that the agent cannot resolve, escalate to the level with authority.
6. 6Record the decision in the conversation, even if it is just a one-line internal note.
7. 7Review at the end of the day the tagged conversations that no one took: they are the signal that there are not enough people or that the tag was not the right one.

Common mistakes when applying the matrix

These are the mistakes that appear over and over again when a team adopts a decision matrix and applies it halfway.

• Tagging by topic instead of by action: 'inquiry' does not tell anyone what to do.
• Assigning without a due date: the conversation stays in the name of someone who has already left their shift.
• Escalating by perceived urgency and not by real consequence.
• Using the tag as an archive instead of as an instruction: a tag that no one looks at again does not organize anything.
• Changing the matrix every week: if the criteria change, the team goes back to deciding out of habit.
• Not reviewing the conversations that were left untaken: they are the best indicator that the matrix is not being applied.

How to know if the matrix is working

A decision matrix is evaluated with three numbers, not with perceptions. The first is the proportion of tagged conversations that end up assigned: if it is very high, the tags are not filtering anything. The second is the time between the message's arrival and the first response, broken down by action (tagged, assigned, escalated): if all three are the same, the matrix did not change the flow. The third is the number of reassigned conversations: if the same conversation goes through three people, the assignment criteria are not clear.

In Wando, message and campaign statistics let you see volumes and times, and contact tags help with segmentation, but the matrix is defined by the business.