How artificial intelligence assists our Relation Center agents
Using machine learning to manage the flood of emails
Juri and Jürgen, you are experts in the field of artificial intelligence and deal with text mining in your daily work. This makes me think of mines, mining, blasting. Of course, text mining is about something completely different. What exactly is it about?
JURI PÄRN: First of all, it's important to understand that not all text mining means the same thing. Depending on the environment, the term has different meanings. Generally speaking, text mining involves extracting new information from texts.
JÜRGEN JÄGER: That's right. Mining basically means that you can gain knowledge from big data - and you can do that with the help of machine learning.
JURI PÄRN: In our case, we are talking about Natural Language Processing (NLP). This is about understanding language in a wide variety of contexts. Currently only written language, in the future perhaps also spoken language.
"A computer can only get as good as the data you give it to learn."
How is this Natural Language Processing applied in the Relation Centers?
JURI PÄRN: Let's take a fictitious example: You receive 1,000 e-mails, all of which belong in different categories - for example, returns or delivery status. However, they still have to be sorted into the right mailbox so that the e-mails get to the right contact person. Our AI service is the "sorting center" that classifies the large volume of incoming emails from customers according to different categories in real time. The emails are then routed to the appropriate agents without any detours. In this way, the AI can provide the agents with targeted and optimal assistance, and the e-mails end up directly with the right expert. Currently, we are talking about an average of 17,500 e-mails per day. This spares the agents monotonous and repetitive work, so that they can fully focus on the personal exchange with our customers.
That sounds like a great relief for our colleagues. But are there also times when e-mails end up in the wrong mailbox?
JURI PÄRN: Of course, mistakes can and will happen - a 100 percent accurate model is unrealistic. A computer can only become as good as the data you give it to learn from. So if a classification is created spongily, the machine will not be able to deliver the incoming e-mails correctly either. If irony, sarcasm or double meanings come into play in the e-mails, even we humans are at our wit's end from time to time.
"There will be a lot more time for the cases that really need face-to-face interaction and expertise."
What's next for AI in customer support?
JÜRGEN JÄGER: With the help of AI, we are trying to relieve our colleagues in the Relation Centers as much as possible and therefore want to further expand automation. But only in terms of activities that are repetitive and monotonous and can therefore be easily automated. In the future, for example, an e-mail that is classified as a return credit could trigger an automatic reply indicating that the processing of returns is currently taking longer and asking the customer to be patient. We are currently working on the possibility of an automatic reply.
In the future, our goal is to gain more knowledge in the field of machine learning to reduce the clutter of emails so that the agents can focus on the personal consultation. So there will be much more time for the cases that really need personal interaction and expertise.
If you could wish for something that AI or even Natural Language Processing could do at some point, what would it be?
JÜRGEN: Frankly, I'm happy as a clam at the moment. Nevertheless, we receive new information from research and applications in the field of artificial intelligence and NLP every week. We follow this closely and integrate relevant advances into our OTTO solutions.
JURI: My wish for AI would be that we can soon optimize computers so that they "understand" language. In other words, that we can interact with them in colloquial language.
And if anyone is now worried that machines will take over the world at that point: Understanding in this context does not mean that computers will develop an identity and, based on that, their own goals. And understanding will not apply to the entire world, but to sub-areas such as e-commerce. What might that look like? For example, when I'm shopping online, I can ask my PC: "Which jeans are available in which shape and color at otto.de? Show me the third one from the left. That would be my wishful thinking.
