CallChecker

We were approached by Mycredit, which needed to develop a service that would improve the quality of its existing call center without hiring additional staff. The company has 4 departments that work with clients: the total number of calls per month is 70,000. The current team of supervisors involved in call quality checks handled an average of 250 calls per day (7,500 calls per month). A large number of calls were left unchecked, making it impossible to analyze the work of call center operators and make objective and informed decisions about the development of the call center.

It was necessary for the service to analyze audio recordings of telephone conversations, then provide transcription and analysis of the conversation, and calculate the scoring value of the quality of the call. The quality score should be calculated both for a specific call and for the manager.

The analysis process looks like this: Artificial intelligence analyzes the transcription of the call and answers the questions that are embedded in its promo, after the analysis, each question is answered: "Yes" or "No".

Examples of questions:

1. Did the operator introduce himself?
2. Did he/she tell about the company?
3. Was the operator polite?
4. Did he/she talk about the service?
5. Was there swearing in the conversation (on both sides)?
6. Emotions (laughing, crying, shouting);
7. Key phrases ("I don't need anything", "I'm busy now")
8. More than 2 voices in the conversation
9. Silence for more than 10 seconds
10. Grammar, assessment of the clarity of pronunciation of words
11. Excessive conversation time (10-15 minutes is enough to resolve the client's issue, if more, it is already off-topic communication)

After that, based on the answers to the questions, we built a Score for each audio recording, which allowed us to understand the quality of the conversation and not waste time listening. Based on the Score of each audio recording, the manager's Score was built.

To make the system flexible, we made it possible to create and edit rules, each rule consisted of two parts: the wording of the rule itself and the promo for the AI service.

We added versioning in order to track changes in the rules and retrospectively analyze how it affected the caller and manager scores.