Calltouch Predict

Calltouch Predict is technology of automatic recognition of quality of calls.

2019: Conversion of calls to the text indexed by a system

On November 13, 2019 it became known that Calltouch provided the updated tools using possibilities of artificial intelligence. The preiCT of Calltouch transforms all calls to the text indexed by a system which can be read in a messenger format. Read more here.

2016: Announcement of Calltouch Predict technology

On June 28, 2016 the service of call tracking and end-to-end analytics Calltouch announced development of Calltouch Predict - technology of automatic detection of quality and result of phone call.

The technology is based on speech recognition, machine learning and linguistic analysis. Its use helps automatically, to define without participation of the person - what of calls were target, i.e. led to sale, and on the basis of these data to optimize expenses on advertizing channels.

Representation of Calltouch Predict, (2016)

According to developers, Calltouch Predict will help the companies to save up to 30% of the marketing budget due to exact determination of quality of calls, their advertizing sources.

In case of the order of goods online, the company knows about the buyer almost everything: the advertizing channel, search query, the search system via which the user came to the website. However in the industries with difficult, high-marginal and expensive products to 70% of visitors prefer to call by telephone specified on the website. At a call to the company the source is defined by means of call tracking (determination of a way of a call).

It gives information on quantity of calls from each advertizing source. But a call outcome – sale or a call to service - the staff of the company should record independently: right after the conversation or listening to all entering telephone traffic.  

The specified methods are inexact because of a human factor, or are expensive: for listening of each call it is necessary to increase staff. Therefore no more than 10% of the companies in general analyze quality of phone calls.

Services of call tracking will rise the companies in 3-4% of an advertizing budget, and together with recognition of quality of calls - 4-5%. The net savings in this case can make up to 30% of an advertizing budget, at the same time the number of transactions and income will remain at the previous level.  For medium and small business of Calltouch Predict it is relevant as the companies need to analyze quality of incoming calls.

Action of a system

Processing of a call Calltouch Predict technology consists of several stages. In the beginning it is required to create the training selection of calls. For this purpose the manager of the company listens to calls and notes them as selling (target) or inappropriate. As a rule, it is enough to mark manually no more than 300 - 500 calls.

Further these calls will be recognized and will be transformed to the text - on the basis of speech Yandex SpeechKit technologies. Then, using linguistic analysis and algorithms of machine learning, Calltouch Predict analyzes the training selection and reveals the speech templates corresponding to different types of calls. On the basis of the revealed patterns, a system begins to mark incoming calls in completely automatic mode. Accuracy of determination of intentions of buyers is not less than 95% and grows with quantity of the recognizable calls.

Scope of new Calltouch technology is much wider: she allows to analyze the dialogs happening in online consultants, chats, texts of requests. An algorithm it is possible to teach to reveal rudeness of personnel, to control compliance of behavior of the call center agent to the provided scenarios of a conversation, to reveal different behavior models of clients.

We made a hypothesis that people who are configured to make purchase can have certain patterns on which they can be integrated in groups. For this purpose we together with our clients listened to a set of calls and found out that from a conversation it is always possible to draw quite unambiguous conclusion on whether the person is going to make purchase or just wants to obtain some information. Further we began to reflect how we could estimate automatically a conversation essence without listening of phone calls. Here we were come to the rescue by a speech sensing technology from Yandex SpeechKit. Then we connected algorithms of machine learning which, analyzing text arrays, automatically determine, the client for what purpose called. Mikhail Fedorinin, founder of Calltouch

40% of purchases in  Homeme  is made via phone. We always carefully analyze each of our advertizing tools and its efficiency as online, and in offline. Besides,  regularly we listen to calls  of clients for the purpose of quality control of customer service. Testing Calltouch Predict, we understood that we can solve two problems one tool: reveal sources which give purchases and also reduce the volume of resources on quality control at phone calls.