It's commonly believed that call centers are huge halls where operators respond to customer requests. We rarely think about the fact that people working in call centers are faced with a large amount of information, for the processing of which huge resources are involved. This includes the work of operators, equipment, electricity costs, rental of premises, depreciation, etc.
Disclaimer: All the stats mentioned below have been derived from original studies and research done by our company.
Modern machine learning technologies help to significantly reduce the cost of providing services, as well as increase the efficiency of the centers.
Practice shows that about 70% of the requests coming to the hotline are of the same type.
Currently, the robotic services of the so-called “first line” support effectively allocate these requests. To provide high-quality services, it is enough for them to recognize the subject of the request and ask the client a few clarifying questions. This allows the company to clearly, quickly and unambiguously satisfy the client's request.
It's worth noting that a request from a client can come in various forms and through various communication channels — messengers, chatbots, a voice assistant or an operator working according to a definitely verified algorithm. And in all these cases, machine learning technologies come to the rescue. They allow you to determine the most appropriate “mask” of questions and give a more accurate answer to customers.
The most primitive robotic systems are linear chatbots
We meet them in messengers, social networks, mobile applications and on websites. The listed chatbots are untrained and work according to a certain scenario. But they are useful — with their help, you can order a pizza or a table in a restaurant, specify the cost of sending a parcel or get a ticket to see a doctor. At the same time, the average request processing time will be reduced by about 3 times.
These bots allow you to maintain customer loyalty — it is known that more than 50% of people would prefer to solve issues without communicating with people. And prompt and more detailed information undoubtedly leads to an increase in sales.
Example 1: International Delivery
As an example, let's describe the scenario of working with a chatbot of a company engaged in international delivery.
Not so long ago, there was a need to send cargo from country N to country K. We went to the company's website and chatted with a chatbot. At the same time, it turned out that due to the non-standard dimensions of the cargo, it is necessary to clarify the parameters: length, width, height of the cargo.
As it turned out, the number of places in the container that had to be paid for significantly depended on a couple of centimeters. After making a couple more calls, 20 minutes later, we successfully answered all the questions.
After that, we were switched to the operator. We have successfully confirmed all the transmitted parameters of the cargo, and he placed an order for delivery in 5 minutes. Thus, the time costs were as follows: about 30 minutes were spent on a chatbot and 5 minutes on live communication with the operator. The effectiveness of such a chatbot turned out to be equal to 85%.
Example 2: Insurance
Let's analyze another example of a first-line chatbot, in the development of which I happened to participate. The company was engaged in insurance and wanted to create its own chatbot with AI. The peculiarity of this project was the complete absence of chat, and the entry point of the request was in the voice channel. The pandemic worsened the situation and led to a severe overload of the voice channel. As a result, it was decided to configure a first-line support bot to improve the quality of service and unload the call center.
At the first stage, we just launched a chat and in a month tried to make it a single entry point for most requests. Then we analyzed and configured a system of quick widgets of the following type: information about the policy, an appointment with a doctor, an insurance case, etc. The client could choose one of these widgets and continue to communicate with a live operator in a chat. At the same time, the operator marked the dialogue with the client with tags. In fact, at the first stage, the operators carried out data markup.
Then the system put down the tags automatically, and at the end of the conversation the operator had to confirm them. This procedure took no more than 10 seconds. By the way, this option is still valid.
Every two weeks, we conducted additional training in recognizing the topic of communication.
At the final stage of the project, the chatbot completely closed 30% of requests in automatic mode without operator participation, another 35% of requests required only operator confirmation, due to the specifics of the insurance business.
Of the remaining 35%, about 80% of the chatbot clients correctly redirected to the operator, making mistakes only in 20% of cases.
As a result, the average time for solving a request in which the operator's participation is required has been reduced from several hours to 10–15 minutes.
Currently, quite a lot of programs have been developed that allow you to quickly create, implement and configure a simple linear chatbot. The low cost and ease of development allowed them to be implemented almost everywhere. Now this service is actively used by customers of telecom operators, banks, insurance companies, delivery, public services, tourism. Linear robotic systems not only respond to customer requests, but also send out the most relevant offers. Media files can also be provided as a response to the request. Also, such chatbots easily form a database of requests, which allows you to quickly respond to changing customer needs.
But will modern technologies based on machine learning be able to fully replace call center operators?
Already now we can definitely say that the answer is no.
Only a person can answer non-standard complex questions, provide the necessary psychological support, and give an emotional color to a conversation.
Nevertheless, the information initially received about the topic of the conversation, the collected client data and the answers received may allow the client to transfer work with him from machine to person unnoticed by the client. It is worth noting that no modern technologies are ready to solve conflict situations.
Author: Ilya Smirnov, Head of Data Science at Usetech