This year at the rooftop event, we looked at the future of customer service automation. We are clearly focusing on the further development of our OMQ chatbot, our Automator and the engine behind our system. At the rooftop event, our CEO Matthias Meisdrock presented these developments as a lecture.
Customers experience among other things, the following service scenario:
Something is ordered but the delivery address must be changed after it is placed. First you look on the website to get help online but find nothing.
The next logical step is to make a phone call with an employee, who tries to give the most helpful assistance. However, even after the conversation, the helping settings are nowhere to be found and customer service is contacted again by email.
A few hours later you get a reply. As for the company’s side, email replies also have to be validated and the data must be changed in the system - so there is a lot of work involved.
In this case, this clearly means one thing for customer service: a lot of manual work.
A common problem is that answers on the website are not always the same ones that you would get in the call center or by email, because different information is often circulating in a company. To give one example of these situations: FAQs may have only been updated once and are in urgent need of an update. Another reason for different information among the company can be coupon campaigns, of which some employees are not yet aware.
Common service problem:
- Long waiting periods
- Outdated information
- Different information
- Time-consuming processes
To avoid these problems, we offer an AI-based system, which stores the entire service knowledge in our central knowledge database and distributes it across all channels. This means customers can ask questions in chat or use the contact form and always get the same answer in real time.
In order to give customers an answer in real time, an AI must be able to understand the intention of the customer request and match it correctly. For recurring requests such as: Can I change my delivery address? With an NLU engine - (Natural Language Understanding) the AI can immediately give the right answer, which is stored in the database.
OMQ couples a question to a representative database. With the introduction of OMQ, a question can be asked anywhere on the website and you will get an answer immediately. Nothing has to be done on the part of the company, because OMQ takes over all processes.
This is made possible by the central OMQ knowledge database, which stores all questions and answers. All connections run through the database. So it works like a filter for the company: The customer asks a question, the software validates it and gives an answer directly. The company is only involved if a question cannot be answered. In this case, the question is forwarded to an employee, who can then enter this new problem into the database and thus guarantee that it can also be solved by the software next time.
In order to provide a better understanding of which processes run in the background and how our technology works, we would like to give you an insight into the OMQ encoder. You can think of our OMQ encoder as a prism - however, there are no light particles to be split here and there, but questions.
This means that the OMQ encoder breaks down customer questions and creates entire concepts from words and contexts. The system thus learns how the customer’s language works and which words and concepts are to be linked with one another.
For example, the OMQ Encoder contains on the one hand “change” as a concept, and on the other hand “order” as another concept. Phrase concepts are formed from these two sub-concepts, which ultimately become question concepts. The encoder not only learns the language of the service department, but also that of the customer, and because of that is able to understand the relations.
If the customer asks the eyeglass manufacturer for a “rubber” instead of a “nose pad”, the system will still know what the customer means. The OMQ encoder is set up by entering individual questions and their matching answers. Questions can then be assigned to the entries in the database, whereupon the appropriate answer is given.
When entering new questions, the system knows immediately how to deal with it, since it already knows the language. The question can be relocated and assigned directly without the system having to be re-taught.
The OMQ encoder…
- is self-learning
- works out-of-the-box
- is dynamic
Most of other companies use a system that has to be trained and therefore it works neither out-of-the-box nor dynamically. They use categorizers, which makes the entire process more complex. In addition, new questions cannot be assigned correctly, which leads to incorrect answers to the question.
We are constantly developing our engine so that our system can answer questions. The engine can recognize nuances, so it breaks down a sentence and filters out important information. In addition to the support of 32 languages, spelling tolerance and multilingual tokenization, the following functions are now also part of the engine:
- Contextual word concepts
- Cross-reference resolution
- Learns across languages
- Recognizes nuances
Already last year the OMQ Automator was part of our rooftop event. It automatically answers interactive questions that require interaction with an agent. At that time, his interactive answers were the main focus. All relevant customer data is asked by our system and an action is triggered from this information, which is then forwarded to the ticket system or the backend server. For example, a change of address can be carried out automatically without a service employee having to intervene.
In order to link the Automator quickly and easily to the backend system, we have built a virtual agent (RPA - Robotic Process Automation) instead of the complex connection via an interface, which does exactly the same thing as a service agent:
It logs into systems, fills out data, searches for the affected positions, validates responses and forwards processes. Finally, it gives the customer an answer and also informs them that the process has been completed. Our beta tests for this OMQ Automator have already started, we are looking forward to further interested parties who want to test this system.
With our OMQ Chatbot, we guarantee natural and foolproof communication in customer service. It is important to us that companies do not have to spend a lot of time implementing it.
The engine OMQ controls the fluency of the chatbot and also works continuously on its (re) evaluation and improvement. Questions and answers provided by our customers are stored in our database and are therefore available on all channels (email, contact form, chatbot, help page, service center).
In 2020 we want to replace our current engine, OMQ Space, with BeOMQ. BeOMQ is able to form an umbrella concept from different word concepts and contexts. This umbrella concept can finally be searched in the database and answers can be provided.
At the moment we are catching an average of 50% of requests. Many of these requests are interactive or can be interactive because their processing involves customer data. With our Automator we can intercept 30% more and, thanks to the new engine, we can intercept 5% better inquiries, since it offers many more features and captures detailed information.
Further advantages of the system are:
- quick implementation
- no risk project
- more effective service
- high quality