Reading your own conversations, on purpose, once a week
Once a week, read ten conversations end to end. Not to score people - to find wrong answers given confidently and promises nobody kept. Neither appears in any metric, and both are expensive.
Which ten
- Two of the longest. Length usually means something went round in circles.
- Two that were reopened. Somebody thought this was finished and it was not.
- Two escalated. Was the escalation necessary, or was the answer findable?
- Two at random. To see the ordinary case.
- Two from the newest person. Where coaching is cheapest.
What to look for
| Look for | Why it matters |
|---|---|
| A confident wrong answer | The customer acted on it. Worse than 'I'll check' |
| A promise with no follow-through | The most damaging single failure in support |
| A question asked twice | The answer was in the conversation already |
| A copy-pasted reply to a specific question | Reads as not listening, because it is |
| Silence over ten minutes mid-conversation | The customer does not know you are still there |
The most valuable output is not feedback to a person. It is the list of questions the team keeps getting wrong, which is a documentation problem wearing a training costume - and fixing it once fixes it for everybody.
How to give it back
- Specific and quoted. "In this conversation, here" - not "you sometimes".
- Separate the systemic from the personal. If three people gave the same wrong answer, it is not three people's fault.
- Say what good looked like. Show a conversation that went well, from the same week.
- Fix the cause the same day. A review that produces notes and no changes stops being taken seriously by the third week.
Do it with the team, not to them
The best version is everybody reading two conversations and bringing one thing they would have done differently. It costs half an hour, surfaces more than a manager reading alone, and nobody has to be told anything.
Questions people actually ask
How many conversations should I review?
Ten a week, chosen deliberately rather than at random - some long ones, some reopened, some from each person. Volume is not the point; regularity is.
Should agents know their conversations are reviewed?
Yes, and they should see the results. Secret QA produces defensiveness and teaches nothing; the point is to find the questions the team keeps getting wrong.
Njiwa
One inbox for WhatsApp, Facebook and Instagram, and an API that sends over either WhatsApp rail. 7 days free, no card.