A number that arrived a day late
Until this spring, our sentiment reading was a line on a weekly report. Every thread got one score when it closed, and nobody ever changed anything because of it.
The trouble was the question, not the model. A score for a whole thread is written once the thread has ended, so it can only describe a customer who has already gone.
The two-turn rule already existed, but it waited on a score that came at the end. The v2.1 reader, shipped in April, scores every turn as it arrives, so two negative turns in a row now put a person on the conversation while the customer is still there, with the summary written.
What the average hides
An average is good at hiding a bad minute. On an ordinary call about a bill, the agent quotes a policy that does not fit, the customer says so, and the next turn is worse. People are polite, so the call still ends with a thank you.
Score that thread as a whole and it lands near neutral. Score it turn by turn and there are two negative turns in a row, exactly where a person should have joined.
When we replayed 800 closed threads through both readers, the turn reader flagged 23 that the thread score had called neutral or better. Those were the conversations the old report said had gone fine.
Why two turns and not one
One negative turn is usually about the situation, not the conversation. A customer whose flight was canceled sounds unhappy whatever the agent says, and handing all of those to a person would teach the team to ignore the flag.
Two in a row means the answer did not land and the customer has said so twice. That is a fact about the conversation, and it arrives while a person can still turn the thread around.
Keep the scoreboard somewhere else
A team judged on a mood line starts managing the line, closing on a pleasant turn and asking for a rating only when things went well. None of that helps the customer one turn away from needing a person.
Leave the report to CSAT. Put sentiment in the routing rules, beside the channel and the hour.
CSAT is what the customer tells you afterward. Sentiment is what the agent hears while there is still time to do something about it.
What to do on your floor
Swap the sentiment average on your weekly report for the count of sentiment handoffs, by the intent each thread started on.
Read ten handoffs a week, starting at the turn before the first negative one, where the cause usually sits.
Keep CSAT and sentiment apart, and never blend them into one quality score.
If you tighten the rule to one turn, expect every customer upset about their day to reach a person.
A score on a report tells you how last week felt. Read on the turn, it tells the agent who needs a person now, and the customer only ever feels the second.
