| News and Trends

Putting artificial intelligence to the test: Watson – a good customer service adviser?

Putting artificial intelligence to the test: Watson – a good customer service adviser?

IBM Watson can process enormous quantities of data from a range of different sources. It understands human language, recognizes images and can even learn independently. But how well suited is it to being a customer service adviser? This is precisely what Arvato CRM Solutions, IBM and Lufthansa Miles & More investigated as part of a pilot project.

Can IBM Watson correctly identify written customer requests, extract relevant data from them and then correctly classify them? That’s what Arvato CRM Solutions, IBM and Lufthansa Miles & More are now testing, of course taking into account all the relevant data protection provisions. The tests are based on older free text requests made by Miles & More customers to which customer service advisers had already responded in the traditional manner. To begin with, Watson was able to learn from part of this data. It was first fed requests as well as the correct and incorrect outcomes. It was then fed new requests and was then tested to see whether it correctly processed the requests and correctly categorized them.

The difficulty here lies in the fact that even just changing the address or the name and address is enough to change customer data. In the first case, entering a new address is sufficient. The second case is somewhat more complex. For example, when someone gets married they have to submit certain documents, and a new customer card even has to be issued. In addition, a request often involves multiple issues that must be addressed. The fact that every person expresses themselves differently adds another layer of difficulty.

Multiple challenges for the new technology

The pilot project demonstrated that Watson is generally good at recognizing issues, classifying them and then extracting information from them. However, there are several hundred categories into which requests can and must be categorized at Miles & More. Certain categories were well identified as part of the test. For example, if the requests were labelled with certain key words or appeared often, they formed a large database that could be used to train Watson. But there were also cases where the data was only correctly processed after a hefty period working to train Watson, for example when a request involved multiple issues. The customer service sector would need to balance the time and expense needed to train it and the desired success rate before using it in practice.

Major potential identified

Another finding: IBM Watson has major potential to provide support to customers in the self-service sector and reduce the burden on customer support representatives in responding to routine requests. It’s able to handle natural language and does especially well with German, plus the technology is scalable. But the initial expense is still very high to keep the error rate as low as possible.

The tests did more than just prove Watson’s technical performance capabilities, however. “We also gathered valuable experience in how to use this kind of innovative technology as a basis for new service components,” says Thorsten Hanisch, Management Team Member at Arvato CRM Solution. “Artificial intelligence is a trending issue. We’re now one of the first service providers to work with the broad range of applications this trend has to offer in addition to traditional one-on-one human customer service.” For example, chatbots can be made more powerful using cognitive methods.

Author: Editorial team Future. Customer. 
Image: liuzishan – Fotolia/Adobe Stock

Tags for this article Artificial Intelligence (85) CRM (108) Digitization (167)


Nearly 500K passionate people


languages and dialects spoken


Countries have a TP presence


Markets served

We're global and local. So you can be close

We're global and local.

We speak your language. And we’d love to hear from you