RetailTrends: The secret of chatbots
They have been discussed, applauded and criticized. Either way, chatbots are the next big thing. At bol.com, Billie now handles around two million conversations with customers, more than all the phone calls in the contact center. Time to get to the bottom of all the pros and cons.
This article appeared earlier in RetailTrends, Volume 15, Edition 153. By: Niels Achtereekte.
Customers who start a conversation with Sephora via the app Kik are subjected to a short quiz, so that their virtual conversation partner gets to know them a bit better. Thereafter, they get all the beauty tips and recommendations from the French retail chain. An experience that should be as similar as possible to chatting with friends about interesting products and developments. The American luxury department store Nordstrom poses as that (girl)friend who always has the best gift suggestions, and regards Facebook Messenger as the right place to offer these suggestions. Consumers merely have to answer a few basic questions about the person they’re shopping for.
These are just two examples of the many experiments with chatbots that retailers have started up in recent years. In short, the bots are made for conversations with users, usually via text or audio. They can be used at different times in the customer journey, from information and inspiration beforehand to after-sales. This can be part of the website or your own app, but recently the application in open message apps has increased. In the western world, WhatsApp, Facebook Messenger and Kik set the tone. Interestingly, the use of those apps has massively increased and has even exceded the use of social media some years ago.
In our own country, Albert Heijn has been working with the Allerhande bot through Facebook Messenger for some time. The application helps people with the daily question: what should we eat today? The bot provides recipe inspiration every day and users can easily search for recipes. “The chatbot can usually answer the questions completely by itself”, says Jet Wieske, manager of Allerhande Digitaal at Albert Heijn. “Research shows that 79 percent of our users are satisfied or very satisfied with the recipe suggestions.” However, a human fallback has been built in the application. If a user would like to speak to a real person, you can contact a webcare employee from the application who will take over the conversation.
“Within the customer service world, bots play an important role in work preparation”, says Frank Smit, chief innovation officer at OBI4Wan, a chatbot developer. For example, questions that employees often have to ask to complete information, such as order or customer numbers. Typical first-line support. “It’s all about simple applications, after which a colleague can take over. As a result, customers are helped faster and employees can fully focus on content. ”
Bol.com puts this into practice with its virtual assistant Billie, the well-known blue man. In addition to general facts (what is your return policy?), the tool also provides specific information, such as the status of a shipment. He can also carry out a number of actions himself, such as returning an ordered product. “In addition, Billie can pass on the conversation seamlessly to a live chat employee if he is unable to manage something himself or he can make a call-back appointment”, says Elmer Hiemstra, manager of service experience. “Within our social channels, Billie asks the customer for information such as an order or customer number, so that our service experts can help the customer much quicker. We continuously strive to improve Billie by analyzing which conversations are being transferred to an employee, and how we can optimize the conversational flow to help more customers more successfully.”
When asked for insights, Wieske notes that the right tone of voice contributes to a better customer experience, so that the chatbot is more likely to appeal to the public. “Your brand already has a personality and tone-of-voice. Be yourself, but make the tone natural and human. Too many push messages and notifications are experienced as ‘spam’ and turns people off.” She also emphasizes not to build the chatbot for yourself, but for the end user.
“Your goal is not to be the first to launch a chatbot, but to add value for your customer. Examining customer needs is therefore essential. Also think carefully about the experience. What tasks do you want people to do with the chatbot? Does this match the wishes of your customer?”
Sense of language
“Make a clear distinction between a virtual assistant and an employee”, Hiemstra advises. “We sometimes encounter conversations where the customer keeps addressing the real employee with ‘Billie’ after referral, or vice versa: he thinks he is talking to an employee. This could lead to dissatisfaction, while referral to an employee must remain easy. However good your virtual assistant may be, he can’t do or understand everything, he says. In that case, a colleague must always be on hand to help the customer out. “One should also ensure a good learning course. Especially in the beginning there is a lot that the bot doesn’t understand. It is crucial to continuously analyze and improve this. The virtual assistant is as good as the data you put into it: the more conversations, the more the bot will learn, the better it will do.”
Artificial intelligence is therefore inextricably linked to the bots and ensures that they continuously become better. They learn from every conversation, sharpen their sense of language and give more appropriate answers. As a result, the bots can answer more complex questions as time passes. They do not necessarily have to come across as humanly as possible, but better understand what a customer means by their message, says Smit.
“The customer experience is the most important. A customer must be helped as well as possible. In that case, it doesn’t matter if he knows that he is dealing with a chatbot. As a branch, we have certainly taken steps in recent years and developed systems to better understand the language, but for additional applications and minimalizing costs, there is room for improvement. “
However, Hiemstra says he wants to make steps in humanizing of the bot. “Billie sometimes reacts a bit cool. Billie has basic recognition of profanities and incomprehension, but through sentiment analysis we should get better at connecting unsatisfied customers to an employee more quickly or to have the bot respond more empathetically. “At Albert Heijn explicit attention is paid to personalization. Wieske: “A number of interesting wishes emerged from research among users. For example, many of them would like it if the chatbot takes food preferences such as allergies into account.”
Artificial intelligence is also an important overarching point of attention for bol.com in the coming years. “It’s an umbrella term, but basically we want to substantially improve question recognition, skip conversation steps and make the management of the conversational flows more scalable. Our volumes are currently substantial enough to provide these models with sufficient data to actually perform better.” Existing online correspondence with customers can also help the bots do their job better. The basis of language recognition is usually supplemented by a specifically made language model.
Smit: “Target groups formulate their questions differently. Young people usually use shorter sentences, older people send longer messages. They use different words and may make fewer spelling mistakes. You can train the bot specifically for that.”
Chatbots have been around for a while, but the development is still fairly new. A chatbot implementation therefore requires its own content strategy, says Wieske. After all, the content is the interface. According to her, the production is really different than with regular editorial work. “The bot will only be successful if it adds value to the customer experience. Regular testing and adjustment is therefore indispensable in the process. ”
High-quality and clean data have a major influence on the degree of success, says Hiemstra. “Garbage in, garbage out applies to all projects where data is involved, the same goes for chatbots. You need volume and data that is well-labeled to make machine learning techniques work, so that your chatbot can learn from what he does well and does not do well.”
The future is positive. Several studies endorse the potential of the bots. For example, 25 percent of Americans use it every day. This even amounts to forty percent under millennials. Interest is growing closer to home as well. At bol.com the application now handles more conversations with customers than all phone calls in the contact center: about two million conversations in 2017. Hiemstra:
“We see that Billie is mainly used because he is fast and always accessible, 24/7. User research shows that about a fifth of our customers even explicitly prefer the virtual assistant to solve their service request – above personal contact or arranging it themselves.”
There is a wish for better integration with other channels. The virtual assistant is currently a separate entity, with the exception of a number of small experiments. “In the future, people and machines will work together in all customer contact channels, with speech playing a major role.”
Smit also sees bots developing into virtual assistants for employees in the long run. He sees systems, apart from talking to the customer, assigning messages to specialists or categorizing and deleting items. As an alternative, they could get more of a marketing function. “Especially smaller retailers could use bots as information providers. Think of a baker who lets his customers know via Facebook what the offers of the week are. That could offer a competitive advantage.”