Decision matrix: when to escalate, when to resolve with a template and when to transfer to another area

What this matrix solves (and what it doesn't)

This matrix does not decide whether a conversation is handled by text, by call or in person: that is another decision and it is already covered on another page of the site. The assumption here is that the conversation is handled on WhatsApp and the question is different: who closes it. Three possible outcomes, nothing more. Resolve with a template: the operator sends an already approved response and the conversation ends there. Refer to another area: the conversation stays alive, but another team handles it (billing, logistics, technical support, after-sales). Escalate to a supervisor: someone with more authority than the operator makes the decision, because there is money, a commitment or a risk that the operator cannot take on alone.

The matrix applies per conversation, not per customer. The same customer can generate three conversations in one week and each one can fall into a different outcome. Nor is it a one-time rule: a conversation can start resolved with a template and escalate midway if new information appears. What matters is that the outcome is written before the message arrives, not improvised at eleven at night.

The two variables: criticality and customer type

Criticality measures what happens if the response goes wrong or is late. It does not measure how angry the person got or how much they wrote in capital letters. A price inquiry answered wrong is fixed with a message; a duplicated charge answered wrong generates a complaint, a chargeback and a review. Criticality has four levels and is determined by impact, not by tone.

Customer type is not how much they spend, it is how much a mistake with that person weighs. A new customer who hasn't bought yet weighs differently than one who has already paid and is waiting. A recurring customer who is billed every month weighs differently than one who bought once. And there is a category that is often forgotten: the customer who already had a problem before. If this is the second time they write about the same thing, the conversation goes up a level of criticality even if the topic is trivial, because the first response did not work.

The complete matrix: the twelve combinations

Crossing the four criticality levels with the five customer types gives twenty combinations. Many repeat in practice, so the useful table has twelve rows: the ones that actually appear every day. The output column is the default decision; the exception column is when that decision changes.

How to read the matrix in practice

1. 1Determine criticality by impact, not by tone. If the wrong answer creates a cost, it's C3 or above.
2. 2Identify the customer type with the data already in the conversation: whether they bought before, whether they are recurring, whether they already complained.
3. 3Find the row. If there are two possible rows, the one with the higher criticality wins. Always.
4. 4Execute the default output. If the condition in the exception column is met, apply the exception instead of the output.
5. 5Log the chosen output in the conversation (tag or internal note) so the next operator doesn't repeat the diagnosis.
6. 6Review weekly the rows that were used most and the ones that were never used: the ones never used usually indicate that a category is missing.

Step three is the one that almost always gets skipped. When a conversation falls into two rows, the temptation is to choose the most comfortable one. The rule that the higher criticality wins exists so that decision doesn't depend on the operator's fatigue at seven in the evening.

What each output means, concretely

Resolving with a template means sending a message that is already written and approved, without adding anything. The condition for this to work is that the template covers the entire inquiry: if the operator has to write three more messages to complete the answer, it wasn't a template, it was a routing in disguise. Templates are approved once and reused; if a template is always used with a manual clarification, the template is poorly written and needs to be corrected.

Routing to another area means passing the conversation to a different team, with the context already summarized. The most common mistake is routing without context: the customer explains everything again and the area starts from zero. Routing must include what the customer asked for, what has already been answered, and what is expected from the area. If the area has no way to see the history, the routing is just a forward and the customer notices.

Escalating to a supervisor means the decision leaves the operator's hands. It is not a punishment or a sign that the operator did something wrong: it is the recognition that some decisions require authority the role does not have. An operator who escalates well is more valuable than one who solves everything alone, because the latter will sooner or later authorize something they shouldn't have.

Edge cases that block the decision

• The customer writes in all caps and with insults, but the issue is trivial. Criticality does not go up because of tone. If the tone prevents the operator from responding well, that is an operator management problem, not a matrix problem.
• The customer asks to speak directly with a supervisor. The matrix does not change, but the conversation is escalated anyway: denying the escalation when it was explicitly requested generates more complaints than resolving the issue.
• The customer is not identified and it is not known whether they purchased. They are treated as a prospect until we have the data, unless the content of the inquiry indicates otherwise (for example, it mentions an order number).
• The inquiry mixes two topics, one trivial and one financial. It is split into two conversations or handled by the most critical topic. The easy part is not answered while the hard part is ignored.
• The area to which it must be routed does not exist in the company. Then the row does not apply: the conversation escalates to the supervisor, who decides who takes it.
• The customer writes outside business hours. The acknowledgment can be automatic, but the classification is done the same way and the output is executed when the area opens. A C4 outside business hours does not wait until the next day.

Common mistakes when building this matrix

The first is confusing criticality with difficulty. An inquiry can be extremely difficult to answer and at the same time trivial in impact: explaining how a product works in detail takes time, but if it goes wrong nothing happens. Conversely, a one-line answer can be critical if it authorizes a refund. The matrix orders by impact, not by effort.

The second is leaving the decision in the operator's hands without written criteria. If each one decides when to escalate based on their experience, service becomes inconsistent and the customer receives different answers depending on who attends. The matrix exists precisely so that two different operators reach the same output with the same conversation.

The third is not reviewing the matrix. Businesses change: a new product appears, a new channel, a type of complaint that didn't exist before. A matrix that isn't reviewed every so often starts to have rows that no longer apply and is missing the row that is actually needed. The review doesn't have to be long: it's enough to look at which conversations were escalated too much and which too little.

The fourth is writing the matrix and not leaving it where it's used. A table in a document nobody opens changes nothing. It has to be visible in the place where the operator works, or turned into tags and quick replies inside the support tool.

How to know if the matrix is working

• C3 and C4 conversations reach the supervisor without the customer having to insist. If the customer had to write three times, the escalation came too late.
• Templates are used without manual additions. If every use carries a handwritten clarification, the template is wrong.
• Handoffs include context. If the department asks what the customer needs, the handoff was a forward.
• Operators don't argue about the classification among themselves. If two operators classify the same conversation differently, the row is badly written.
• No conversation is left without an outcome. If one appears that doesn't fit any row, a category is missing.

If you want to put this into practice, Wando's 14-day trial without a card is enough to load the templates and see how the inbox behaves with AI suggestions on real conversations. But the matrix can be built and used without any tool: it's a table and a criterion.