Almost all companies record conversations with clients. But to listen to all the conversations, the same number of people are needed as they "uttered", therefore, as a rule, these recordings are turned to very rarely.
This happens only in two cases. First, when it is necessary to find “who is right?”, That is, in case of disputes with customers or suppliers. Secondly, when the contact center has a process of selective audit of conversations for evaluation. In the latter case, there are statistical formulas that give us the following calculations. To say with 90 ± 5% confidence that the information about the audit is reliable, you need to listen to 214 out of every thousand conversations. In other words, one auditor with a similar work schedule is needed for every 5 operators. And if we take into account that some call centers work around the clock, then the required number of auditors increases.
Thus, it turns out that companies listen to only 1-3% of conversations. However, in dialogues with customers, you can find quite a lot of interesting things if you conduct voice analytics. You can search by the following criteria:
- obscene language,
- words-parasites in the speech of employees,
- marker words and phrases that may indicate negativity or the need to take any action. For example, the words “long”, “expensive”, “late” can become triggers to redirect such conversations for processing to managers.
What are we looking for
Speech analytics is a comprehensive search and assessment of conversations based on several criteria (so-called slices). Here are some of them.
- Customer loyalty is a comprehensive metric that takes into account interruptions, call ownership, exclamations and remarks to the operator.
- — -, .
- — , , “ ”, “ ”.
- —
- —
- — , , “ ”, “ ” .
The requirements for initial voice analytics conversations are fairly simple. This stereo recording where one channel - the customer, the other channel - employee in mp3 format with quality not lower than 16 kbit / c
Using SpeechToText technology performed speech recognition and calculation of the ownership card conversation - how much time talking on the individual subscriber and the operator, interrupted if each other, etc. Next, an analysis is carried out based on dictionaries of slices and the conversation is rated based on the number of matches and the weight of each slice. By the way, the number of phrases in dictionaries exceeds 4,000 expressions.
As a result, a map of conversations scores is compiled, requiring the close attention of call center auditors, department heads, quality officers and security personnel. Thus, the conversations are “highlighted”, indicating the reasons for the low aggregate score.
How it looks
The speech analytics workstation is a list of conversations in various slices with a cumulative rating and transcript of the conversation.
Each call is automatically rated on a 5-point scale. The lower the score, the more likely it is that the conversation needs to be listened to.
The designer of your own reports and slicers allows you to create additional criteria for selecting call records.
Conversation transcript
Full-text search for keywords in conversation recordings window
That's not all ...
As we already wrote above, there is a possibility of practicing the so-called triggers or phrases that met in a conversation and to which an urgent response is needed from the relevant services. Having recognized such signals, the system automatically launches the process in CRM or Service Desk and the necessary information is sent to the managers of the sales, marketing and service departments.
What does the company get
Instead of 1-3% of conversations, the company controls absolutely all conversations with customers. This allows you to get more high-quality data for a quick response to various situations and business development in general.