What are chatbots?
The term ‘chatbot’ is a combination of the words ‘chat’ & ‘robot’. In contrast to classic robots, however, a chatbot is not a physical machine, but a software programme or a service from a Cloud.
Roughly speaking, it is possible to differentiate between two evolutionary stages of chatbots
- Simle chatbots: They recognise predefined keywords or sentence fragments and react to them with fixed linked answers. The sentence elements and answers are entered into the chatbot’s database and thus form the repertoire of response options.
- Self-learning chatbots: They continually analyse feedback from users and “training conversations” enable them to identify gaps in question recognition and any missing answers. In this way they can understand questions in different contexts and provide complex reactions to them.
Furthermore, there are also so-called action bots. This is the name given to systems which support users in small tasks and carry out actions. For example, they help customers on e-commerce platforms to find the right item or to place an order.
Laura Reiser from Zendesk talks about the difference between bots, chatbots and answer-bots:
“People frequently equate bots with chatbots without touching upon the differences in their characteristics. The term “bot” has its origins in the word “robot”. If we think of such a robot we think of a programme that always carries out identical processes according to a defined, programmed logic. It’s upgrade, the chatbot, is characterised by its ability to understand language or text. It’s programming allows it to recognise the content and aim of the questions posed to it and to handle them accordingly.
The answer-bot uses machine learning to answer customer queries with content from its knowledge database. If a customer contacts a company, the bot steps in: it scans the message written by the customer in order to understand what it is about. Then it recommends the relevant articles to the customer whilst they wait for an answer from an agent. It can be used on various channels such as the web, on mobile devices and in instant messaging services such as Slack, and it also has an open API. If the customer still needs help, their question is answered by an agent as usual. Feedback is automatically collected in order to improve future recommendations. That takes pressure off the employees whilst at the same time providing a 24/7 service for customers.”
Which channels can chatbots be installed on?
Currently chatbots are primarily used in instant messaging systems, websites, support forums and on social media sites. They can greet customers, answer small queries or they precede a customer service advisor. In this way the customer feels as though their request is being dealt with immediately.
Chatbot expert Matthias Mehner from Messenger People gives us his tips on getting started:
“The most important thing to note first of all is that chatbots are neither a strategy nor a channel. Chatbots are small programmes which automatically implement things which would otherwise have to be done by a person. In this case it’s “chatting”. In order for a chatbot to even be used in the first place, the company must therefore have previously engaged with messenger services. And every company that still communicates with customers by telephone, email or social media should do that.
Many chatbots are emerging on Facebook Messenger, as almost every company today has a Facebook page. However, a Facebook page must be managed differently than a chat, and in Germany right now, Facebook Messenger is not the most popular option. 81% of Germans use WhatsApp and companies should also keep that in mind when developing their channel strategy. Chatbots for WhatsApp are not as easy to develop and don’t offer as many options as on Facebook – but they are definitely the better assistants.
In order for a chatbot to even be used in the first place, the company must have previously engaged with messenger services.
The digital advisors thereby play an important role in 1st level support. They can be used to automatically retrieve requests e.g. pre-qualified or basic information such as the customer number. Chatbots can however also serve to advise customers automatically. Therefore chatbots are suitable for all companies who value a good and direct link to the customers – whether it is to offer advice before the purchase, to act as an assistant during the purchase or as a service offering after the sale.
My tip for getting started: first draw up a messenger channel which customers can use to make direct contact with you. Answer the questions from customer service manually and thus learn a) how the channel is being received and b) which questions are being posed. Using the experience, targeted chatbots can then be developed step by step.”
How can chatbots be used by companies?
Chatbots can be used anywhere where a company wants to makes contact with customers and offer assistance. Simple or recurring requests can thus be answered quickly and outside of service hours. That provides the opportunity to build and maintain good relationships with the customers.
How to successfully use chatbots
The demands placed on customer services are growing along with progressive digitalisation. Customers want quick, conclusive and knowledgeable answers to their questions.
The technology is still not advanced enough to install chatbots for very complex and cross-thematic queries. Therefore it makes sense to assign the programmes a domain and to inform the customers exactly which issues the chatbot can help them with. In this way they know which queries they can pose to the chatbot and in which instances its competence is exceeded.
Alexander Braun from Creative Construction talks about the potential uses of chatbots:
“The ideal potential uses arise within a clearly-defined domain and along a decision tree with clearly definable options at the forefront. A good example of this is Ada Health: in this case, the domain “health” is transparently selected in advance for every user. The user will thus not attempt to discuss the current political situation or films with the chatbot, which limits the scope of knowledge to be shown in the chatbot and is therefore more easily displayable. The scope of conversations which break down or fail to be understood by the chatbot due to it having insufficient knowledge is thus limited. Because Ada identifies the possible illness along a decision tree on the basis of symptoms, the possible response options are also clearly narrowed down here, which in turn reduces the risk of the dialogue breaking down.
For the user, a chatbot should be clearly limited to the domain in which it can also effectively guarantee the strongest possible answers.
A negative example is Apple’s Siri: this chatbot was introduced with the marketing message stating that it is not limited to a specific domain and therefore is open to all potential applications. As anyone who has ever once used Siri knows from painful experience, the reality looks very different: “I’m sorry, I didn’t quite get that” is probably the most frequent response. Although the chatbot is supposed to improve usability compared with a graphic interface, the result is the opposite: whilst in a graphic interface, the user can at least clearly recognise the potential options, on a purely speech-based interface, they are not apparent to the user and must be explored via trial and error. This can lead to terrible usability. The consequence of this is even more negative: after the user has now learnt via trial and error what the chatbot can do – e.g. enquire about the weather or make a call from an address book – and what it can’t do – so almost everything else –, these are the only use-cases which the user will use in the future. So if the knowledge base of the chatbot further develops over time and opens up new use-cases, these are not apparent to the user and they continue to only use the features in the learned, limited domain. This therefore limits the use of the chatbot from the very beginning, despite its continued development.
For the user, a chatbot should thus be clearly limited to the domain in which it can also effectively guarantee the strongest possible answers. In this way, the user’s disappointment is minimised. In addition, depending on the specific application, a carefully selected combination of speech-based communication and graphically-represented selection options is recommended in order to guide the user to the desired goal as efficiently as possible.”
Customer acceptance of chatbots
The acceptance of chatbots amongst customers is growing steadily. In fact, only 24% have communicated with chatbots in the past, however around half of users can imagine doing so in the future. Amongst 18 to 29-year-olds, it is the majority at 60%. It is important for customers to know that they are communicating with a chatbot and that a human employee is on hand if their query exceeds the competence of the bot.
Chatbots and data protection
Particularly in the era of the GDPR, the topic of data protection may not be disregarded. A study by Pegasystems has shown that almost a third of respondents (27%) have concerns about data protection and security with regard to chatbots.
Lawyer for the “IT-Recht” law firm, Max-Lion Keller, explains the legally compliant use of chatbots and gives us a glimpse of the future:
“Chatbots will revolutionise the paths of customer communication in the future. However, exchanges using such dialogue systems necessitate particular legal data protection measures in accordance with the GDPR, which also depend on the operating modes of the bot. Firstly it is necessary to enquire about the legal basis for data processing. For live support bots, which function via a dialogue field on the website and answer general questions to provide a service, data processing may be regularly justified by the legitimate interest of the operator in providing effective customer service in accordance with Art. 6 (1) (f) GDPR. The same should also be possible for contract-specific requests about orders or their implementation against the backdrop of the provision of customer data.
On the other hand, for chatbots which go beyond a support service and thus process extensive contact details for specific queries or provide the processing for advertising purposes, the express consent of the user to the data processing must be regularly sought before the activation. In this case it must be ensured in particular that the consent is recorded and that the respective briefed user can easily revoke their consent at any time.
For chatbots which process extensive contact details for specific queries or provide them for advertising purposes, the express consent of the user to the data processing must be regularly sought before the activation.
Finally, a so-called order processing agreement in accordance with Art. 28 (3) GDPR must be regularly concluded with the chatbot developer, in which the rights and obligations of the parties are regulated. This is necessary as the chatbot developer – as the agent of the website operator – collects data via the bot and makes the latter available again.
This status quo regarding data protection law could yet be intensified by the planned European e-Privacy Regulation. However, a consolidated version does not yet currently exist. Experts therefore anticipate that the regulation will not take effect before 2022.”