From Zero to Service in Two Hours: Live Programming of a Chatbot
How long does it actually take to prepare a chatbot for service? If you have the right partner, it can actually go quite quickly - as proven by a team from Arvato CRM Solutions during a recent live event in Berlin. The specialists were able to program a chatbot in just two hours. And the audience was invited to participate.
Every year, the CRM Forum is held prior to Call Center World in Berlin. Guests of this year’s industry event enjoyed an extra-special presentation: the live, on-site programming of a chatbot that provides travel recommendations on behalf of the fictitious company QuantosX. QuantosX is a teleportation travel service provider and, in the CRM Studio of Arvato CRM Solutions, serves to vividly present the company’s portfolio of solutions.
“Trade fairs always feature a number of chatbot demos, but we’ve never seen a live programming in which the audience was able to participate,” explains Thorsten Hanisch, Management Team Member at Arvato CRM Solutions Germany. “We wanted to demonstrate the basics of how to train a chatbot, and to show that this type of project – even with its many challenges – can be implemented relatively quickly, with the help of the right partner.”
Chatbot training by the audience
The team of Arvato CRM Solutions began by asking the guests at the CRM Forum what criteria are important to them when planning a trip. Using the various answers provided, the audience then voted on the five most important categories: the number of travelers, type of activity, type of accommodation, duration, and budget. The participants also determined how they would ask to be assisted by a human customer support representative, in case the chatbot were not able to provide satisfactory guidance with a particular query. For this eventuality, a team of customer support representatives were on standby at an Arvato service center, ready to provide further support to the “customers” and to show how seamless the transition can be between chatbot and human support.
The next step involved developing the necessary answers to the travel queries. The forum participants could select various travel destinations and augment them with information from the five previously determined categories. One specific example was as follows: Barcelona, single, party, vacation rental, weekend trip, 500 euros. This dataset represented a weekend trip for one person who wants to fly to Barcelona to party. The data could be submitted by email, WhatsApp or text.
Live in just two hours
In this way, around 100 datasets were developed and entered into a database, and then fed into the chatbot in structured form. In just under two hours, the time had finally come: the bot had to prove what it could do. To facilitate the demonstration, the Arvato team had prepared a website with responsive design, on which users could ask about travel offers. For instance, those looking for a weekend trip with a high party factor for singles, costing up to 500 euros and including accommodations in a vacation rental, would receive Barcelona as a recommended travel destination. Other suitable recommendations could also be displayed, if desired. And, of course, the chatbot also successfully transferred customers to its human colleagues at the service center, if requested. Test passed!
More than technology
So, what are the key factors for successfully implementing a chatbot? “In our view, there are three crucial criteria,” explains Thorsten Hanisch. “First, an interdisciplinary team composed of IT process analysts, machine learning specialists, dialogue designers, knowledge engineers and computer linguists. They should be managed by a ‘chatbot architect’ who is familiar with all of these disciplines and can guide the team accordingly.” Secondly, according to Hanisch, the employed technologies need to fulfill certain prerequisites; for instance, they should be compatible with other systems and channels. It should also be possible to integrate CRM systems and backends. “And thirdly, the chatbot and the human customer support representative must be able to work well together. The human needs to be able to immediately take over in the event of complex queries, but also to hand simple tasks back over to the bot.” The connection of an analytics solution brings even more added value, since it provides support such as the right next best actions.
Author: Editorial team Future. Customer.
Image: sdecoret – AdobeStock