Interview. Call Center Revolution: Which Office Workers Will Robots Actually Replace

Recently, Amazon opened a fully automated store , where there are no sellers, cashiers. This event triggered a new round of trend discussions on the replacement of people by robots in many areas of business (for example, here’s an article on The Guardian). I was able to talk with Dmitry Plotnikov, an expert on automating business processes (Microsoft MVP), about who actually had to start worrying about their jobs.

You work a lot in business process automation. Can you share your observations on which employees will soon be occupied by robots?

Generally, various lists of occupations that will destroy robots (for example, here and here ) have been published on the Internet for several years - they always call workers from factories, drivers or waiters. In fact, from what we see, office problems may arise before others.

This, for example, who?

For example, employees of call centers who are engaged in cold sales through telephone calls. Usually, the operator has a clear conversation script (script), which states what and how he should speak. To deviate from it is often prohibited. And even if a situation arises that is not handled by the script, the operators are supposed to repeat only those phrases that are in their script — the rest is forbidden.

That is, a person, like a robot, will repeat the same words. And if so, then why not replace it with a robot, because it will save: the program does not need to be taught, it can work around the clock and seven days a week.

But in this case, it turns out that the system will have to lead a voice conversation, and this is even more difficult than a chat bot. Are there any significant successes in this area now?

Yes, I happened to take part in a project where it was necessary to implement just such a system - an application that uses the SharePoint database as a CRM and communicates during cold calls. This system has already been launched and is working for one of the customers from the western region (unfortunately, I cannot name the company).

There are quite a few such decisions on the market; most of them use the approach in which a computer synthesizes a human voice. It almost always sounds unnatural, it’s not very pleasant for a person to talk with such a robot.

Therefore, in our project it was decided to abandon speech synthesis and use in communications the pre-recorded voice of a living person. As a result, the system reads the interlocutor suggestions from the company's script and decrypts his answers on the fly, converting them into text. One of the most important tasks here is to analyze the content of the call. Key words can be used to understand exactly how a person reacts to what he hears (this also includes searching the knowledge base), and depending on this, build further communication - end the conversation, play the next sentence of a pre-recorded text, etc. .

How effectively does it work?

Surprisingly, the system works quite effectively and allows you to seriously save and increase the productivity of cold calls. The application itself can set up statuses on the basis of conversations in CRM, which is very convenient, and one program per day can call hundreds of clients. The cost of a call center that could do the same amount of work would be very substantial.

Are there any difficulties in creating such applications?

Yes of course. In the approach that we used, there are some difficulties, for example, to reproduce the recorded speech so that it sounds natural is not so simple. For example, during a conversation between two people, long pauses occur rarely, therefore, the robot must respond to the replicas of the interlocutor quickly. At the same time, in real life, the person to whom the call is being received may be in a noisy place - and it is not easy for the robot to understand in real time when the phrase addressed to it has ended and you need to react to it.

In the same way, some words may sound the same, but have a different meaning and be recorded differently (homophones). For example, in English, the word bot and bought sound the same, but mean completely different concepts.

How can you solve these problems and improve accuracy?

Machine learning, neural networks. Any call center has a knowledge base, as a rule, quite extensive: there are conversation scripts and their records (they are often made, for example, to assess the quality of operators' work). For a start, you can create a neural network and train it on this data - the decrypted records of conversations. The result will be a full-fledged virtual operator for making cold calls.

In conclusion, can we give any advice to those who are going to create similar automation systems for routine office tasks?

In this case, perhaps the most important thing is to choose the appropriate technologies and tools. It is unlikely that you have enough resources to do everything from scratch on your own, so you have to use ready-made products and various APIs. And here it is important that they have the necessary functionality and good documentation.

For example, during the project we wanted to use the tools from Google and Microsoft, but we found out that one of them does not support the languages ​​we need, and the documentation of the second was so bad that some points were clarified only during the experiments. If this is avoided, you will save a lot of time and effort.

Interesting articles on the impact of automation on the labor market:

• Shops without cash desks, cash and staff: how the future of trade works
• Continuous learning and job cuts: how technology will impact the labor market
• What professions will disappear because of automation - a study by Bloomberg
• Workers should not be afraid of robopocalypse
• How automation increases the standard of living and saves jobs