Speech analytics of calls
The TARP research of the client complaints in the 200 largest companies in the world showed a very interesting findings:
Only 20-25% of complaints are caused by low qualification of operators (ignorance and non-observance of procedures, knowledge gaps, incorrect behavior)
The 15-20% of complaints are caused by the complexity of the clients themselves – unrealistic expectations or client errors
And the biggest cause of complaints is problems within the company itself, 55-65% of complaints are caused by shortcomings in products / services, marketing policies, billing errors and heavy business processes
In order to understand the reason for the complaints, somehow it is necessary to conduct an analysis of telephone conversations, which is a non-trivial and difficult task: for the 100 operators need 100 more quality controllers to listen to conversations and analyze them.
Traditional analysis of interactions with customers does not give the desired result, for the following reasons:
The 80% of quality controller efforts focuses on agent problems (behavior, competencies), but not on problems in the company
82% of companies listen to no more than 10 calls per agent per month
It is not possible to identify trends and systematic problems in the topics of interest by hand, as this requires a full analysis of all telephone conversations
NOVA IT actively implements the speech analytics solutions that are able to automate quality control and provide a 100% overview of all telephone conversations. After recording of the phone call, phonetic and semantic recognition is activated to provides the complete transcription output to text, and puts calls into categories (“complaint”, “promise of payment”, “problems with the personal cabinet”, etc.).
The analytics of calls can also include the separation of conversations on topics, emotional characteristics, silence, etc.
There are several ways to analyze phone calls:
Full-text speech analytics. In this case, we use the Large Vocabulary Conversational Speech Recognition engine (LVCSR) -complex technology, which uses basic sets of words (bigrams, trigrams, etc.) with the inclusion of linguistic and acoustic models on them. This allows to make a complete indexing of the recorded conversation and build an analyst without predetermining keywords
Speech recognition (keyword search), in this approach uses phonetic recognition – analisys of phoneme sets, which allows you to determine the necessary words when analyzing calls. An important point is to determine the keywords that will be searched.
The second approach is more difficult, because it requires a lot of computing resources and professional consulting, so it makes sense to use in projects where there are at least 100 operators. The first method is suitable even for small contact centers and it does not require significant financial investments and gives quick results.
NOVA IT offers solutions with both methods: in each case it make sense to offer one or another method of the analysis for certain purpose.
Speech Recognition allows you to analyze not only problem calls, but also good calls.
The main directions of research of problem calls:
The main directions of research of good calls:
Compliance with the script
Customer focused agents
Consent in sales
Promise of payment
The implementation of a speech analytics system in collection agencies gives the following results:
The increasing the payment promise from 6 to 15%, through the using of effective phrases
Reducing of silence by 33%
Reducing abandoned calls by 4%, etc
The opportunities of speech analytics are not limited only to telephone calls. Nowadays technology allows to analyze audio taken from any source. To do that it is necessary to have a highly sensitive microphone in the office, or sale point (branch office), to record the conversations between employees and customers. Further, audio is processed by algorithms similar to the analysis of telephone calls. At the output company have the opportunity to control the quality of service “on the ground”, which is a fairly powerful quality control tool in comparison with a callback or a “secret buyer”.
Such using of speech analytics has certain limitations in comparison with telephone calls – a noisy environment and the inability to separate the client audio and employee, but nevertheless is successfully used in many areas – from bank branches to network pharmacies.
NOVA IT has many years of experience implementing complex systems using speech recognition and speech synthesis technologies, and their integration with the customers information systems.
What technologies do we use: