The future of the chatbot (or conversational robot) seems to be very bright: It is projected that by 2020, 80% of companies would be using them for their interactions with customers. As a result, the chatbot market is expected to grow by 37% in the next 4 years . But, what are they really?
Enhancing one’s customer experience with a Chatbot
Are there any constraints? How to set up a chatbot? Faced with these questions, Claire Sorel of Visionary Marketing interviewed Frédéric Canevet, Product Manager at Eloquant.com (customer relations specialist).
Tip # 1: Make a qualitative and quantitative assessment of the most relevant use case, then customize the conversation design accordingly
The problem must be voluminous and recurrent. Frederic reminds us that it’s not a question of volume only, sometimes a FAQ can be sufficient, but the value is also important. The good news is that value is attainable, thanks to a connection with the information system the bot is able to deliver personalized answers in a very simple way to the users. It is therefore important to frame the use case appropriately.
Tip # 2: Comprehend the operations
Although setting up a bot is a management decision, it is required to comprehend the operations well in advance to better understand how to solve a problem. Following this methodology, we must surround ourselves with people who will first ask a dozen questions to the bot, which would create a use case of data. Then, the collaborators begin what is called a training phase. For half a day, they keep asking the bot questions. This would allow you to accumulate a number of typical questions to feed the bot with. You must also keep in mind that the bot must learn continuously, i.e., it should evolve so as to reach the required response rate, which lies between 85 and 95%.
Tip # 3: Create chatbots specializing in domains
Frédéric reminds us that a bot which responds to just anything is not a good one. The responses must correspond to precise use cases that resonate with demand in terms of volume and value. His advice is to have a bot specialized in a particular domain: a bot for customer service, order tracking or for other special cases such as helping the person in his ordering process. It is simple enough, we can customize the bot as needed and especially so that it does not mix up everything- for Frederic indicates that if the sentences are slightly similar, the bot is likely to be wrong.
Tip # 4: Make the bot direct the conversation
While designing the conversation structure, one must try to guide the user to avoid being deceived by the bot. For example, when the conversation with the bot begins, according to Frédéric, the bot shouldn’t say “hello, how can I help you? but rather, “hello, do you want to track your order? The bot must begin with an unambiguous, to-the-point statement and it should always be able to steer the conversation.
use case of a chatbot: the Eloquant example
Eloquant is a French-based SaaS CRM software vendor and is working with a French company that had just obtained new missions from the state. The missions resulted in large volumes of applications to be managed by the employees in a short time. In order to meet the requirements of its missions, the call centre increased its workforce from 60 to 160 employees. But, despite having an augmented workforce, processing times and delays were on a rise. The administration then used Eloquant to ease the workload of its employees and to focus only on Level 2 and Level 3 claims.
Identification of the levels of demand:
- Level 1 request: very simple and recurring requests
- Level 2 request: semi-complex issues which could be handled by a chatbot
- Level 3 request: requests too complex to be managed by a bot
At first, a FAQ is set up- a program on the Internet that allows people to do “self-service”. But this isn’t enough. A chatbot is needed and its setup requires a special process, which is as follows:
- A workshop to define the needs
- Feed the chatbot: Use the semantic analysis solution on 70 of the received emails and phone reports in order to analyze the conversations. Thus, we end up identifying and quantifying the most recurring themes and the exact formulations of the questions asked by the customers. By now, the chatbot is fed with some initial learning.
- Take the human factor into account:It is not just a software that is setup, rather a whole project that is built. It is vital to use imagination now to come up with certain hypotheses and anticipate content that the bot may encounter in future. For this, a discussion workshop is organized with the operators so as to get familiarized with the real problems facing the customers. Another workshop is conducted with the IT department to discern the feasibility of the connections to the information system to obtain a response with added value.