back to overview

Chatbot Podcast - 100% Automation in Customer Service

We were really happy to receive the invitation from Sophie Hundertmark to talk about chatbots. Our CEO Sven Engelmann gladly accepted the invitation to be a digital guest on the podcast hosted by the chatbot expert. In this article, we have summarized Chatbot-Podcast for you.

Sophie: Before we dive into the subject of chatbots and e-commerce, I would first like to know about you and your company. Could you shortly introduce yourself?

Sven: My name is Sven Engelman and I am one of the founders of OMQ GmbH, where we combine customer service with artificial intelligence. At the beginning, we were laughed at really hard because of it, but then we managed all the challenges and developed a system in which the companies can connect to all communication channels. We equip contact forms with our system, create dynamic FAQs, support agents with their ticket processing and of course we provide chatbots.

 OMQ Wissensdatenbank mit Produkt Incons

You said that you deal with the topics of customer communication automation and AI, what is it about chatbots that makes them special? I mean, what fascinated you with them?

Chatbot is the medium through which also young people communicate. It combines both topics of artificial intelligence and new communication media. With the integration of chatbots to our OMQ world, we can reach out to new employees or new customers that we would not be able to reach before with our software. With the help of chatbots, it is possible to lighten customer service requests through another communication channel.

Does that mean then that you have introduced chatbots in order to grow your customer base?

Yes, that is also one of the reasons. Another one is that they are really interesting, technically speaking. As I said before, we connect all communication channels with one another. So our system can learn from all of them and their conversations. When our AI learns a new word through the contact form, the system knows how customers define a certain issue. When the same topic is brought up through the chatbot, the self-learning AI already knows: It has to do with a certain topic that was already learned through another channel. It is really interesting to connect all of this world together, so that the AI learns from it and can react accordingly to each individual contact point.

Today we want to talk more about e-commerce and I know that you have already achieved a lot of success. Why do you think that chatbots make sense in e-commerce and should they be used in the long term?

In e-commerce customer service, we work according to the Pareto principle.

This means that especially in e-commerce, 80% of service requests are recurring. These recurring requests can be answered with 20% of the topics the company handles.

Here we land again on the subject of artificial intelligence and pattern recognition. The narrower this funnel, which makes it possible for the AI to recognize certain topics and identify the intentions, the better it also works in the chatbot. The framework conditions in e-commerce are predestined to be processed with an AI or a chatbot. E-commerce is a business that is subject to certain fluctuations. With the help of such systems, these peaks can also be flattened in customer service so that the customer service is not overloaded. These are the possibilities offered by the chatbot.

Thank you for the explanations! It is certainly also interesting to learn about your success experiences. Why did you use a chatbot in e-commerce and are there projects you can tell us about?

We have a project which is very interesting - our 100% project. We try to bring the automation rate to 100 percent and have equipped all communication channels with our system to one of our customers. The chatbot is intended not only to provide information and answers, but also to interact dynamically. To do this, we have developed a virtual agent that can perform activities in the background that a service employee usually needs to perform (for example, entering an address change into the company’s back-end system). In this project, we let this virtual employee do these activities directly.

The big goal and our vision is to achieve a 100%.

It is exciting to see if we can achieve that. We have already achieved a solution rate of 80%, which is very good. Now it is all about analyzing the last 20% and figuring out what the topics are and how to automate them. Of course, there are also service requests that are not comprehensible with the human mind alone - how is an AI supposed to answer that? But we are on a very good path with the project. We are investing a lot in the 100% project and want to give our customers more opportunities to carry out interactive projects in the future. Our experience has shown that the system is used very frequently in e-commerce, especially whenever manual customer service is generally not available. Generally in the evening hours, the system is so busy that the chatbot is used accordingly.

I have noticed many times that customers have a different tolerance towards a chatbot. Have you seen this here or in a similar project?

What we noticed during the evaluation is that in e-commerce, the chatbot is often used as a ‘lightning conductor’ and it is massively insulted. But what must also be said is that this reduced tolerance can also be used positively to get more information out of the customer or to be more available to customers in certain situations. If, for example, the customer has payment difficulties, then it is of course a stronger hurdle to go to the phone and admit it. But when the customer talks to a machine, he knows that he does not have to reveal himself to another person. This then goes through the individual application stages step by step and in the end, the customer simply presses “Accept”. Chatbots offer the possibility to handle unpleasant cases in order to contact the company without shyness.

In this case, the customer has no inhibitions to talk to this machine and puts completely aside the emotional aspects. This must be used as much as possible in these cases. There are very nice areas of application to use the machine or the features of the machine in such a way that you can get the most out of customer communication.

Yes, thank you. What I also find interesting is what you do when you create concepts on a chatbot, or if you want to develop a chatbot for a customer. Are there any ready-made steps that you work on?

We always proceed in such a way that we try to keep the customer’s effort in such a project as low as possible and to leave the most effort to us. We have noticed that the less the customer has to do, the higher will be the acceptance at the time of the system implementation. In the best case, this works in such a way that we get FAQs from customers, which we incorporate completely into our knowledge database. The information can be worked out, and then we see how you can work with certain multimedia content. With this data set, which was then entered into the knowledge base, we can then let the chatbot do its work. This already brings with it a corresponding basic intelligence and with each interaction with the customers the system then becomes more and more clever.

Whenever the chatbot does not know how to handle it anymore, it saves the conversation and at best you have a list of a thousand different communications. Another system of ours manages to sort these communications out and directly evaluate which topics are still missing from the knowledge base and need to be maintained for the chatbot. The system also detects which topics lead to an interaction with customer service or a negative result, even though the response has been displayed. These events are evaluated and our customers get a list of missing questions that are being asked by the customers, or with suggestions about what needs to be improved. This is not a project that is completed at some point - we see it more as a continuously improving process. The idea here is to work with technology and leave the evaluation to the machine, because it can do this much more precisely and faster. We build a base, integrate everything, get started and then we can improve everything piece by piece.

What is your view on the topic “chatbot personality”? How do you know it should or is allowed to have small talk? Does the chatbot talk more in a formal or informal way? How do you train your chatbot to do that?

Our customers decide on all these small-talk components. What we have to say about our concept is that we always integrate it into existing systems with our chatbot. This means that we always take existing chat systems and integrate a virtual OMQ agent, which then acts in these chat systems. The whole widget, the whole front end, is usually delivered by a chat provider, and we install a virtual agent there. This then brings a corresponding small-talk component. You can set a default setting beforehand whether you want a more formal or informal way of addressing. You can use special formulations. These settings are done within a few minutes. The chatbot then resorts to the knowledge base, in which all the FAQs are located, but also uses this small-talk knowledge base. It greets; says goodbye; asks questions; and if it does not know how to help the case anymore, then it says: I cannot answer your question. Do you want to talk to a human agent?” This is also very important with a chatbot - that you always succeed, that you never leave the customer alone, but always have the opportunity to be connected with an employee. Alternatively, a contact form or callback service can always be used to give the customer the opportunity to contact the company’s customer service.

On the subject of “seriousness” and “humor” - I can not really say whether a chatbot should be particularly humorous or serious. In other words, should it be more straight-forward?

Exactly, because the bot itself works out-of-the-box. It does not need to be trained and only needs an existing FAQ data and can then start immediately. Such a chatbot project can be done in a week. If you want to make special adjustments and specifications, then it all drags on in such a project. That is why we decided that we are taking a bot that works out-of-the-box and can get started right away. Of course, you have to make certain compromises when it comes to how humorous or serious it is.

But what can be set is the confidence score. This means that we say exactly how precise the chatbot is. With the chatbot, you have the opportunity to ask the customer again what the relevant topic is. You output either a pool of answers or a single answer. However, this only happens if the bot is sure of a certain percentage. When the chatbot has reached a certain confidence score on a customer’s question, it immediately returns information without asking again.

We have already heard a lot and you have already answered many of my questions directly. What have you learned about your achievements from chatbot projects so far?

One lesson we learned is that it is important not to send the customer somehow into a one-way street. He must always have the possibility to contact a human service agent in addition to chatbot communication - either by e-mail or by phone.

At the beginning, we underestimated that the data collected is difficult for AI to process, because it is highly fragmented. For example, phrases are entered in a contact form. In an FAQ, the search line sometimes has simple word groups and in the chatbot only words are worked with, without any connections. Making this information workable for AI has not been so easy, and we have also drawn the learning from the fact that it makes a lot of sense to use a central system and to work through all knowledge from all communication channels and then make them available accordingly.

Thank you, Sven! Until next time!

Contact us if you have any questions about chatbots or AI in customer service. We look forward to seeing you :)