Look who’s talking! Triumph of the chatbots

Mr. Kraume, artificial intelligence seems to be a preoccupation for research and companies, and everyone is talking about chatbots. Are these just buzzwords and fads?

Karsten Kraume: No, chatbots are already a reality and are gaining acceptance among all age groups. The use of new channels like live chat or social media is increasing; at present, it stands at 2.4%. This is one finding of our representative 2018 CRM omnichannel panel, which examined the channel preferences of eight core industries and ten contact channels.  And all channels are becoming increasingly automated. This shows us two of the drivers of chatbot growth.

So are users somewhat slow to adapt the new technical solutions?

Karsten Kraume: I think it’s all a question of usability. The deciding factor when a user chooses a channel is whether they can use it to resolve their issue without difficulty and, especially, receive a solution quickly. So one basic prerequisite for chatbots is technology that performs well. And the better they can meet customer demands, the more readily they’ll be accepted. User acceptance ultimately depends on the structure of the dialogue, that is to say, the manner in which the bot responds. A chatbot can also be useful when no live agent is available, because experience shows that live chats tend to be terminated more quickly when the customer has to wait for over a minute.

We recently set up a chatbot for more than 6,000 users at Münster University, and clearly people weren’t reluctant to use it, considering that over 36% of the users switched from the web channel to the chatbot.

Professor Wünderlich, why is it that the manner of the dialogue is so important?

Nancy Wünderlich:

Human communication is more than just the efficient transmission of information. And every contact alters a customer’s perceptions of a company. Am I being taken seriously? Do they understand the urgency of my issue? Does the company seem authentic? We explored that in a study onthe role of artificial intelligence in customer support. Users tended to regard their interaction as less authentic when they believed they were communicating with a chatbot. Perceived authenticity generally has a positive impact on a user’s behavior and attitude. Although good chatbots are capable of determining the mood of the customer and responding accordingly, the success of this always depends on the context of the service situation: the way a customer judges the support specialist is more important in a banking context as opposed to looking for a product in an online shop.

How can chatbots be given something resembling a personality?

Nancy Wünderlich: Of course customers tend to perceive something like “personality,” all the more so as the technology advances further and the chatbot comes across as more and more natural. This personality has to be appropriate to the character of the company and, even better, to the specific application and context, but it also has to come across as authentic. Extremely cheery dispositions and robotic monotones are poor choices. Having said that, it’s important to bear the so-called “uncanny valley” in mind, the acceptance gap: if a chatbot seems too human, this can also cause irritation. Exactly how the chatbot should behave, how strongly it takes after the corporate brand, which speech patterns would be best for communicating with the customer in order to seem authentic and competent — these are all questions that are best answered on the basis of the specific context of its application. So our first step is to examine the preferences of the specific target audience in order to create a concrete blueprint.

So the context of its application plays a crucial role …

Karsten Kraume: Exactly. The concrete use case is extremely important and is always central in our thinking. Digitization and the deployment of chatbots at all costs is not automatically the best solution to a problem or for a company. So the question of technology is never the first step for us. Instead, we concentrate on the customer journey and how a company seeks to make it as smooth and mutually beneficial as possible. While that can call for a purely technical solution, it doesn’t always. The last thing that customers expect is to be confronted with a poorly designed chatbot existing in its own little bubble with no connection to the brand’s other touchpoints. For me, there are certain parallels between this and the hype around apps we saw a few years ago — apps are a means to an end, not an end in themselves. Furthermore, every industry demands its own individual solution, because the reasons for making contact tend to be industry-specific. On top of this, different sectors typically have markedly different backend environments. I’d just like to briefly go back to the project with Münster University as an example of the education sector. On an abstract level, a student information system is comparable to a CRM system, but it’s sometimes completely different from the backend systems in other industries like healthcare or air travel, or a dealer management system in the automotive sector.

What does that mean for companies?

Karsten Kraume: Companies have to decide how they want to map out and optimize their customer experience. That can mean automation and chatbots, but it doesn’t have to. Arvato CRM Solutions is showing interested parties the digital possibilities that exist for companies in this regard at its CRM studio in Gütersloh and also in Berlin, using the fictional travel company QuantosX as an example. We are also actively conducting research in our digital lab at the European Research Center for Information Systems (ERCIS).

Can you give us an example of a real solution?

Karsten Kraume: The chatbot we’ve deployed at Münster University is one example. But at this point I’d like to discuss another similar example. Kontowechsel 24, one of Arvato CRM Solutions’ subsidiaries, has become the first provider to offer an Alexa skill which can answer customers’ questions about switching bank accounts and even supports automated account switching. This new skill is a supplement to the digital account-switching service, which automatically informs all payment partners about new bank account details.

In general, how do you measure the success of a chatbot?

Karsten Kraume:

There are various metrics for that. For one thing, we measure the “adoption rate.” In the example I gave earlier, I mentioned that over 36% of the target group were using the chatbot. These users were poached from online self-service. This result speaks for itself. Another parameter we analyze is the bot’s response time. This often depends on other integrated systems. Just as with conversations with a real agent, only certain response times are acceptable.

Professor Wünderlich, as a scientist, you probably have a different perspective on bots and assistants. Where can these be deployed?

Nancy Wünderlich: As it stands, text-based chatbots and language assistants are at their strongest when dealing with the simplification or optimization of communication in the broadest sense. They respond particularly well to rule-based queries, tirelessly and around the clock, relieving some of the pressure on customer service workers. But this is just the beginning. Due to the continuous development of the underlying AI technology, we are now starting to make use of chatbots in sectors that require much more comprehensive understanding and reasoning and even planning. I believe that, in the medium term, we’ll see chatbots being used in simple consultancy, sales and even social services. I think that chatbots also have a bright future in public administration.

Some users still have reservations about chatbots and virtual assistants. In your view, what questions and risks remain?

Karsten Kraume: Bots have to be trained and supervised, because a chatbot is only as good as its “training.” They can ultimately only provide answers that have been given to them by the AI developers or that they have learned themselves. This means that providers have to do their best to understand the end customers’ situation as precisely as possible, such as by analyzing customer queries with text mining to arrive at the relevant themes. And of course a chatbot requires a human counterpart who can automatically take over complex topics where appropriate. One of the factors that will determine the quality of the outcome will be facilitating the transition from chatbot to human-human communication as seamlessly as possible, without interrupting the flow of the dialogue. We see all the time that a rethink is often required, and the necessary corporate structures have to be created in order to maintain a bot. We are happy to help with that, especially because we’ve had to carry out our own organizational restructurings, and we’ve learned from those experiences.

What contribution can research make?

Karsten Kraume: The research into artificial intelligence and machine learning is essential. Cooperation between industry and scientific research, like what’s happening in the ERCIS omnichannel lab, benefits all parties involved. This is where academic research and theory come together with global industry experience. We are also anticipating new impetus from the recently launched Initiative CLAIRE, of which Arvato CRM Solutions is a sponsor. CLAIRE aims to establish a European network of centers of excellence for AI across the whole of Europe. The goal is to give the field of artificial intelligence an impact and brand awareness similar to that of CERN, Europe’s globally renowned particle physics laboratory. As well as the technical side, I see a great deal of overlap between marketing functions and research, and experts such as Nancy Wünderlich or Manfred Krafft and Mirja Kroschke from the Marketing Center Münster.

Looking once more to the future: will human service workers eventually be completely replaced by chatbots?

Karsten Kraume: Chatbots are absolutely on the rise. However, in my view, they won’t be able to completely replace people in the foreseeable future, and I’m skeptical as to whether users would even want this in all situations. The results of our Customer Survey 2027 point the other way: the findings underline the special role of the human components and, despite the backdrop of increasing automation and digitization, speak in favor of a hybrid solution for the design of the customer experience.

Nancy Wünderlich: I would agree with that. Even with artificial intelligence and machine learning driving constant improvement, our studies show that communication with a bot is viewed as less authentic, especially if there is uncertainty as to the identity of the other person. This discrepancy is a disruptive factor. Communication purely with bots, or even between bots, is unlikely in the medium term, in my view. Particularly when it comes to complex topics or complaints, excellent customer service needs human emotion and empathy as much as it always has.

Professor Wünderlich, Mr. Kraume, thank you for your time.

Nancy Wünderlich holds the chair for business studies, specifically service management, in the Management Department at the Faculty of Business Administration and Economics at the University of Paderborn. She co-authored the study “A Nice and Friendly Chat with a Bot: User Perceptions of AI-Based Service Agents” (with Stefanie Paluch, RWTH Aachen University).

Karsten Kraume is Arvato CRM Solutions’ CIO/CSO and represents Arvato as a member of the advisory board at the European Research Center for Information Systems (ERCIS) and also at research programs such as Business Process Management RISE.

Author: Editorial team Future. Customer.
Image: © sdecoret – AdobeStock