Most of us imagine that in the nearest future we all will live and work in smart houses, surrounded by robots to which we will delegate most of our routine work.

It’s logical that we’re not thinking about perspectives of Mars colonization or quantum computer that much as we think about improvement of the way we deal with our routine needs. The latter is what quality of life is in short.

That’s why the improvement of customer services is so important. We need it to be automated in a smart way. We want to achieve nearly human-human interaction while in reality it’s human-machine. Now we’ll look at what it takes to bring customer service closer to the nature.

We already have a lot of things that in one or another way make our life easier, and most of them come from customer services. With automation, CS has started to become way easier for the both sides – business and customer – however, it’s not possible yet to automate the whole process. Why does that happen?

1. Insufficient understanding

We can train a chatbot with the huge amount of data taken from different conversations. We can make it learn from its following conversations with the users (even though self-learning systems are still quite complicated in development).

However, there would still exist high enough possibility that the user can have particular issues that the bot has never faced before. This is where human is needed to intervene into the process.

There’s a bunch of such services that provide mixed conversations. The first that users interact with is chatbot. It receives inquiries and answers simple questions that it can retrieve from its knowledge base (real database, training set it used to learn from, or previous conversations).

Then chatbot sends the inquiry to the proper human consultant that is going to handle it. The most popular examples of such services are Magic and Operator.

More on commerce apps and use of chatbots for commercial purposes you can find here.

2. Lack of empathy and emotional intelligence

This is the primary reason why most of the people don’t like old conventional bots, which relies on the straightforward hard-coded logic only. The thing is that it’s natural for us to feel bad or good about talking to someone. All those moments are the key bricks of human interactions.

Whether something goes wrong or just okay – we apply the same human-human patterns to communication with colleagues, clients, service support representatives etc. The same happens when we interact with the chatbots. Above all, we would think about them as about human beings we can target our feelings and emotions at. Most of the bots make an essential difference that deprives us of the ability to treat them naturally.

You know there’s a bunch of narrow-specialized bots with very unnatural conversation flow. They have buttons inside and just guide you through the process of searching by clicking those when needed. For some purposes, they are still useful.

However, when talking about interactions with conversational agents we mean conversations that remind talking to people. This what we need to aim at and always try to provide the most seamless conversations even on the narrowest topics.

3. Bots are not universal

We can’t put versatile tasks on the “bot’s shoulders”. Great performance implies strict specialization on the particular type of work. With processing huge amounts of data, it becomes even more relevant. Such bots need to be measured in certain, usually narrow, field.

The broader field is – the more intricacies it implies. Building chatbot that is knowledgeable in the wide range of fields is the most challenging task. It would take a lot of time and computer power to create such one. However, we already have examples of conversational agents that can learn from the data they acquire from the interactions with the users.

The famous examples are XiaoIce, Mitsuku, Azuma Hiraki. You may have heard about them in the context of Loebner prize and self-learning systems. More on this topic you can find here.

Let’s make the first conclusion of what we’ve talked about. I will start. Even though artificial agents have certain disadvantages, human assistants are also far from perfection due to lack of ability to handle big amounts of data, low speed of interactions, the presence of “human factor” etc.

Then, what should we consider as the alternative solution? Well, optimum lies in the middle. We need balanced compound of bots performing most of the tasks and humans coordinating the workflow.


Taken from

What are the most important ingredients of creating customer service?

1. Chatbot that relies upon framework with the tools for speech and image recognition, natural language processing and other possible cognitive services;

We have been communicating with bots for a while already. Just recall automated bank assistants, telephone operators and different consultants, which made you hate them holding on for an indefinite period of time. But it’s over. Now we’re talking about the smart assistant who performs conversation patterns which resemble human ones.

It learns from the conversations with customers and is capable of self-improvement. It’s armed with speech and image recognition and can make conclusions on the gathered data including customers’ reviews of the service.

 2. Messenger platform for interaction

Bots are where the messenger is. Now it’s one of the easiest things in development to configure bot application with one of the most commonly used messengers like Kik, Facebook Messenger, Slack, Twitter, Skype.

All of these services propose you convenient API toolkit not to bother yourself with handling abundant HTTP requests and responses

3. Human admin part to coordinate the process

It should be possible for a coordinator to stop automated dialogue flow and to talk directly to the customer. It’s extremely useful when bot faces unusual questions it can’t handle or when something needs to undergo a correction.


The importance of finding an appropriate balance of human and robotic parts can hardly be overestimated. And we can trace it on bad examples as well when services started to misuse robotic virtual assistance by placing robots across all the mediums with very few human coordination.

Since the customer isn’t able to contact someone else when his (her) inquiry isn’t satisfied, it always feels bad.
Even worse it can turn out for the companies whose CS is supposed to be their trustworthy representation. There’s a measure of chatbots’ effectiveness, invented with their massive emergence (called “Conversation per session”).

The average number of bots is 1.5-2.5 (with few intelligent exceptions, like WeChat’s XiaoIce). It means that in average it takes up to 3 messages for a person to be over with talking to a bot.

We are impatient beings. When something is promised us, we consider it as if somebody owes us that. Consequently, we want to get it all and now. Nearly the same thing usually happens with CS when we’re told that we can turn to them 24 hours a day 7 days a week and finally we’re getting stuck in senseless conversation which only eats our time.

Let’s imagine that now you are convinced that robotic assistant is exactly what your CS needs. What to start with?


Text analysis with Microsoft Cognitive Services

I recommend to start with a messaging platform and look at the examples [link to the article “Top Industries to Use Chatbot”] of the chatbots that exist there to understand the specificity of the medium.

Here’re some useful links to set off:

Find your perfect messaging platform:

  1. Facebook for developers
  2. Slack API
  3. Telegram Bot API
  4. Skype Developer Platform
  5. KIK Bot Dashboard
  6. WeChat API
  7. HipChat for developers

Pick up the toolkit for easy development and configuration with MP:

  1. Microsoft Bot Framework
  2. Meya
  3. FlowXO
  4. Pandorabots
  5. Botkit
  6. Superscript
  7. Gupshup
  8. Hubot
  9. Chatfuel
  11. Bots UI kit
  12. Beepboop

Add functionality, make your bot smart:

  2. Microsoft Cognitive Services
  4. Amazon Alexa Skill Kit
  5. Google Knowledge Graph Search API
  6. IBM’s Watson Developer Cloud
  7. TensorFlow

Building admin panel totally depends on the personal choice of back-end developer. For instance, direct working with Microsoft Bot Framework implies rather using ASP.NET for back-end development. It has set of advantages like testing with Bot Emulator and easy hosting and publishing to Microsoft Azure. However, it doesn’t make a big difference and totally depends on what developer feels comfortable about.

The idea behind all of what you’ve read is to give you the insight of wise using of the chatbots. Misunderstanding is crucial. If you think that chatbots can substitute everything, then you’re definitely getting it wrong. Chatbots deployment can help a lot, however, they still need a human supervisor. When you come to balancing robotic and human parts, you can get incredible productivity.

Recommended reading:

Chatbots Evolution

Let’s trace together with the way from the simple device (where we all started) to a smart virtual assistant.

Top industries to use chatbots

Are the chatbots important for current technology era? Chatbots are obviously the main tech trend we’re talking about in 2016.

Pros and Cons of Chatbots

Education, marketing, healthcare, lifestyle, customer services – all of these fields suddenly turned out to be divided into automated tasks with chatbots.

Growth Hacker and Sales Hacker, MVP builder, love to run technology companies.