Audit Checklist for Conversations With More Than Three Exchanges Without Resolution

A conversation that went three exchanges without being resolved is not, by itself, a mistake. Some inquiries legitimately need back and forth: a quote with measurements, an appointment with availability, an order with variants. The problem appears when that pattern repeats and the customer ends up leaving without an answer or without buying. This page is not meant to be read straight through: it is meant to be printed or copied, to open the inbox and check off. The idea is that in twenty or thirty minutes you can go from "there are many long conversations" to "the problem is in this specific stretch."

Before starting, a methodological warning. Do not measure this audit with the date the message was saved in your system: use the date and time that the message itself carries from WhatsApp, which is the one that reflects when the customer wrote it. If you look at the internal upload date, any analysis of schedules will give you a false result, because imports and migrations dump everything together at the same time. The message timestamp is the only one that works to know whether the delay was real.

How to build the sample before looking at anything

Auditing all long conversations is impossible and also unnecessary. What you are looking for is a pattern, and a pattern is visible with a small, well-chosen sample. Define the universe, the size, and the period, and only then open the first conversation.

• Define "unresolved": that there is no confirmed order, no scheduled appointment, no quote sent with acceptance, or an explicit closing like "let's leave it there." If you cannot define it in one sentence, you are not ready to audit yet.
• Filter by number of exchanges: take only those with four or more messages from the customer and four or more from your team. Less than that is not a stuck conversation, it is a normal inquiry.
• Limit the period to seven days. Further back, the context changes (campaigns, promotions, holidays) and the pattern gets diluted.
• Take between 15 and 25 conversations. Fewer is not enough to see repetitions; more makes you lose the day.
• Separate by entry channel if you have more than one: a conversation that started from an ad behaves differently from one that came in from a spontaneous message.
• Write down the category or the reason for the inquiry for each one in a separate column. Without that, you will not be able to cross-reference later.

What to review in each stretch of the conversation

A long conversation gets stuck in a specific stretch, not in the whole conversation. That is why it is worth reviewing it in four stages and noting which one it broke in. Most cases concentrate in a single one, and that tells you exactly where to touch the process.

When you finish classifying the sample, count how many fell into each stage. If most get stuck in development, the problem is information: the team does not have at hand what the customer needs. If they get stuck in decision, the problem is the offer: nobody is proposing a concrete step. If they get stuck in closing, the problem is follow-up: the conversation dies on its own.

The pattern to look for (and the four most common ones)

An isolated case is not a pattern. A pattern is the same failure repeated in different conversations, with different customers and sometimes with different agents. These are the four that almost always appear, in order of frequency.

• Mirror question: the team returns a question instead of an answer. "What size do you need?" when the customer already said the size, or "When do you want it for?" without offering any date. It is detected by counting how many team messages end in a question mark without having provided a piece of information before.
• Data requested twice: the customer gave a piece of data in exchange 2 and the team asks for it again in exchange 4. It is the most costly pattern, because it makes the customer feel that nobody read them. It is detected by underlining the data the customer provided and crossing it out when the team asks for it again.
• Option that never arrives: the conversation moves forward, the customer gives everything they are asked for, and at no point do they receive a price, a date, an alternative or a concrete step. The customer is left waiting for something they do not know they have to ask for.
• Silence after the customer's last reply: the conversation ends with a customer message and no team message. It can be a natural closing or an abandonment; the only way to tell them apart is whether the reason was recorded.

In addition to the content pattern, there is a time pattern worth cross-referencing. If the stuck conversations are concentrated in a time slot or on a specific day, the problem is not the script but staffing: there are fewer people than needed at that moment. That is measured with the message time, not with the internal upload time, and it is solved with shifts, not with longer replies.

Checklist to tick during the audit

1. 1Define and write down what counts as a "resolution" for your business before opening the inbox.
2. 2You filtered the conversations from the last seven days with four or more exchanges.
3. 3You trimmed the sample to 15-25 cases and separated them by entry channel.
4. 4For each case, you noted the stage where it got stuck (opening, development, decision, close).
5. 5For each case, you marked whether there was a mirror question, data requested twice, an option that never arrived, or final silence.
6. 6You counted how many pieces of data the customer provided and how many the team asked for again.
7. 7You checked whether at any point in the conversation there was a concrete proposal (price, date, alternative, or next step).
8. 8You reviewed whether the conversation was left with a tag, note, or closing reason.
9. 9You cross-referenced the stuck cases with the time slot of the customer's message.
10. 10You noted whether the same agent appears in more than half of the stuck cases.
11. 11You separated the cases that are a process problem from those that are legitimately long inquiries.
12. 12You wrote a single improvement action for each pattern found, with an owner and a date.

What to do when something fails

Finding the pattern is half the work. The other half is correcting without breaking what works. These are the corrections that correspond to each finding, and the signal that indicates it's no longer enough to fix it behind closed doors.

A clarification about templates, because they often come up as a quick solution: if the finding is resolved with a message that starts the conversation outside the service window, that message has to go through approval. The only timeline that can be stated is up to 48 hours, and there is no way to guarantee a shorter one. If the correction depends on that, plan it with a margin and don't promise it for the same day.

How to know if service improved after the correction

The audit doesn't end when you apply the correction: it ends when you can compare. Repeat the same exercise two or three weeks later, with the same filter, the same sample size and the same period. What you compare is not the total number of long conversations, because that number depends on the volume of inquiries, which goes up and down on its own. What you compare is the proportion: what percentage of conversations with more than three exchanges ended in resolution.

• Before the correction: note the percentage of long conversations that reached resolution.
• After the correction: measure again with the same filter and the same sample size.
• Also compare the stage where they got stuck: if it moved from development to closing, it improved even if the percentage stays the same.
• Check whether the time distribution of stuck cases changed.
• Don't mix this measurement with traffic, review or social media metrics: they don't measure the same thing and aren't useful for making decisions about customer service.

If you want the suggested replies to be based on what your team has already answered well before, tools like Wando learn from the business's real answers and propose them so that someone can approve them with one click. That helps the corrected pattern hold, but it doesn't replace the audit: you still have to do the list above, because the pattern changes with the business.

Common mistakes when doing this audit

• Looking at the whole conversation instead of classifying it by segment: that way everything seems "a little slow" and you don't find the cause.
• Counting messages instead of exchanges: five consecutive messages from the customer are a single exchange.
• Measuring with the internal upload date instead of the message date: the times come out wrong and so do the conclusions.
• Auditing a single long conversation and drawing a general rule from it.
• Correcting three things at once: afterwards you don't know which one worked.
• Confusing a naturally long inquiry (a quote with measurements) with a stuck conversation.
• Assuming the customer left because of price when in reality they never received a concrete option.