18 thousand M.Video-Eldorado’s employees throughout Russia use a single IT infrastructure. A huge amount of technical and business changes take place regularly, requiring the support team to respond to requests of company’s staff efficiently twenty-four-seven. To solve these tasks, CROC IT Company has developed a chatbot based on artificial intelligence. We discussed it with Felix Skvortsov, the head of robotic process automation at CROC.
Felix has been responsible for the project since the beginning. He told us about chatbot features, challenges faced during its development, bot’s impact on M.Video-Eldorado , and its further evolution.
On April 9, Felix Skvortsov will speak at AI Conference in Moscow on the topic: User service robotization using artificial intelligence.
Interviewer: AI Conference (AIC).
Respondent: Felix Skvortsov (FS).
AIC: Why did M.Video-Eldorado need a chatbot for internal use?
FS: M.Video-Eldorado has approximately 30,000 Russian employees working at 940 shops. Frequently, people do not have their own workplaces: for example, sales personnel should be in the room along with customers. Clients have various issues that should be constantly solved. Therefore, M.Video-Eldorado required a bot for the ITSM system (IT service management – Ed.). When team members should solve a certain problem, they address the ITSM system via the chatbot using their smartphones.
AIC: What features do the chatbot perform?
FS: Currently, the chatbot conducts the most sought-after features for which the first line technical support is responsible: registration and classification of requests as well as answers to common questions such as where to find a necessary document. Besides, we have added the possibility to coordinate and edit requests through the chatbot. In other words, employees can create, edit, or withdraw a request, and check a progress status. An application coordination and rejection option is available for chiefs.
M.Video-Eldorado is going to expand the chatbot functionality and to delegate obligations of the second line support to it. This will be basic tasks, including granting access to workers, account unblocking, etc. For instance, each employee of M.Video-Eldorado’s head office will be provided with an access to external email – Outlook Web Access, but not everyone can have it because of different reasons. One is now discussing the matter of providing a default access.
AIC: How did you design the chatbot?
FS: Initially, we launched a pilot project using machine learning elements. We took the selection of data on customers’ requests and cleared them out. Based on the obtained dataset, we taught the neural network to properly recognize and classify requests. The pilot results showed that our idea was viable, while the classification accuracy of all requests passing the rapidly trained neural network exceeded 70%. After the pilot launch lasting two or three months, the customer decided to implement the project.
Indeed, the project was more comprehensive and interesting: we took an extended dataset and integrated a chatbot platform (it was our Open Source-based solution). Then, we united artificial intelligence, chatbot platform, Service Desk system, and made a single entry point via Messenger.
AIC: What challenges did you face during chatbot development?
FS: The key problem was related to data. To get an appropriate dataset, one should clean up data for correct system training. Generally, customers provide unrefined data: they open and close requests, but not always manage to control the quality of application filling out.
There were a number of instances after the system launch when customers said, “The system made a mistake here.” We began to investigate, compare activity of system and operator, and figured out that the error was made by the operator rather than the system. The system has lots of such wrong data that should be cleaned out.
AIC: How did the chatbot affect M.Video-Eldorado operations?
FS: The robot processed up to 30% of text requests at the stage of testing. In the future, the chatbot will process more than half of requests sent to the technical support service due to self-learning algorithms. The common request standard will increase the quality, while the user-friendly, especially for young members, interaction channel – Messenger will enhance the execution process.
The IT request coordination feature allowed to involve various chiefs to the project, as all approvals were done on the portal. Today, one can examine a request status and promptly coordinate it within several seconds on Messenger.
AIC: Is it possible to transfer your solutions so that the other company could use them?
FS: When it comes to the platform, these best practices can be easily transferred. With regard to some kind of data (e.g., AI classification), they cannot be shifted, it is private information. It is impossible to train the neural network using data of one client and just offer the summarized version to another.
The chatbot consists of numerous characteristic aspects, abbreviations, reductions, and internal processes. Therefore, transferring a chatbot platform is possible, but transferring brain is not. The system should be trained according to the needs of each customer.
AIC: Are you going to advance the chatbot?
FS: We do want the chatbot take on more features. Eventually, the first line support should include as few people as possible. Not only should the bot accept and register requests, but it also should carry out conventional operations, the amount of which can reach 80% based on our experience. For instance, employees need an access to certain folders, but this access is allowed. They address the bot that makes a request to the access management system, and the rights are given automatically. This means both the first and the second line support.
We will expand bot features by integrating it with customer’s systems. Thus, we will enhance the availability of support and the operation speed, because the chatbot does not take a rest, never gets sick, and works the clock round.