By Dr Wang Aobo, Lecturer & Consultant, Artificial Intelligence Practice, NUS-ISS
Since the invention of computers, people have been looking at ways to talk to computers. Nowadays, with the development of artificial intelligence, speech processing and natural language processing, conversational user interfaces are making significant contributions in many business areas, such as customer service, personal assistant and language learning. Beyond how fancy the technology of machine learning or deep learning is, more importantly, chatbots are a promising tool to deliver better customer experience.
Similar to its AI neighbor, the autonomous vehicle, which has been growing fast with the rise of deep learning and reinforcement learning, chatbots are catching everyone’s eye and contributing more and more to our daily business. However, unlike self-driving cars, which is an extremely costly industry with an uncertain timeline to deliver, chatbots are much more affordable and easier to be tailored for various levels of business objectives.
According to estimates, more than 67% of consumers worldwide used a chatbot for customer support in the past year and around 85% of all customer interactions will be handled without a human agent by 2020. 40% of consumers do not care whether a chatbot or a real human helps them, as long as they are getting the help they need. Moreover, in order to achieve such a level of performance, it requires relatively less effort and resources compared to the autonomous vehicle industry. To some degree, it is because the chatbots do not necessarily need to work for an open domain, thus only requiring machine learning models, knowledge and rules within focused domains.
Real world example
One interesting example is the collaboration between Coca-Cola and Pandorabots. Expectations are for consumers to leverage on these personalised chats with Coca-Cola smart vending machines, as if they were communicating with a familiar bodega operator. As soon as consumers reach their checkout cart, their purchases would appear in their vending machine. In this case, the dialogue management system provided by Pandorabots primarily uses the mark-up language AIML, which is a rule based and pattern matching based conversation engine which requires limited programming effort. It's focus is to carefully design the dialogues and navigate the communication smoothly to the final destination. Updating one of these machines only require up to 45 minutes of on-site reprogramming.
Benefits of using a Chatbot
Apart from the rule-based chatbots for narrow domain, problems with traditional online experience have been well tackled by retrieval-based chatbots. The figure below from Drift’s 2018 State of Chatbots Report validates this point as chatbots have solved many customer problems with traditional online experiences such as getting immediate answers to simple questions instead of navigating through complicated websites, filling up poorly designed forms and waiting for services that are long-drawn.
Key motivation of using Chatbots
With the advent of chatbots, upon receiving a new enquiry, a coherent and useful response can be returned by searching a query-response repository. The best answer to the most similar question asked before will be reused to answer the new questions. This strategy is very effective given the assumption that for most of the queries, they are simple and repeating FAQ questions generated from a large user base, which is true in most scenarios. In this case, a large volume of query-response pairs are required to train the machine learning model so that the similarity between questions and context can be measured accurately.
According to statistics from Invesp, Chatbots can save up to 30% in customer support costs and can help businesses save on customer service costs by speeding up response times and answering up to 80% of routine questions. The performance of such chatbots could be improved directly by collecting more conversation data and designing more question response pairs continuously.
Trending chatbots
To date, the most in demand chatbots for the purpose of customer support and being a personal assistant are the ones driven by natural language processing technology. Popular in the market are the million of skills on Alexa by Amazon, fulfillments on DialogAI, a platform by Google and bots on facebook messengers. These platforms illustrates the great power of understanding of user utterances by simply detecting the users’ intent and extraction of parameters to fill in the slots of a particular task. Both deterministic and stochastic methods such as classic machine learning or deep learning are employed to better understand the user’s intent and extract the slots more precisely. Furthermore, these skill/fufillment/bots developers will then focus on producing the responses or providing the services according to the intent and slots parsed by the platform. Such a strategy allows the focus to be on the specific task asked by the customer and captures responses accordingly through pre-defined behaviours so that chatbots can avoid generating irrelevant or even - funny responses.
The common goal of all three types of chatbots mentioned above is to maintain a high quality of understanding of a user’s questions as well as to generate the most appropriate and controllable responses. Instead of aiming to answer and react in an open domain, the best strategy is to provide great customer experiences by focusing on closed domain business scenarios and applying these smart tips:
- Chatbots should ask simple questions with options provided
- Chatbots should acknowledge their AI identity to manage the user’s expectations.
- Chatbots should sit in the driver's seat to navigate the conversation
- Chatbots should seamlessly handover to a human agent as soon as needed
Will your business explore the use of chatbots in the near future?
Click here to find out more on Artificial Intelligence courses we offer.