Author: Sutapa Amornvivat, Ph.D.
Published in Bangkok Post newspaper/ In Ponderland column 31 October 2018
The smoother and deeper our conversation with Siri, (or with Alexa, Bixby or Cortana for that matter) the more realistic Dan Brown's Winston in Origin becomes. Many experts are predicting that demand for new mobile apps is slowing down with the next trend for user interface being chatbots. Of course, much more work in conversational AI will be necessary before we reach Winston's level of chatbot advancement. At least now, users are becoming more familiar and as a result, these interactions provide a constant source of data for machines to "learn".
Many retail companies are utilising a chatbot platform for various use cases ranging from customer support to recommendations of products and services. According to Oracle, 80% of businesses polled in France, the Netherlands, South Africa, and the UK last year said that they will resort to chatbots by 2020. In fact, one French-based multinational beauty chain already has three chatbots, one to assist in makeover reservation, one to help customers match colour shades, and the other to offer makeup tips and how-to videos.
It is no surprise that Thai businesses are jumping on the bandwagon, especially with the Thai population being so reliant on chat apps that we frequently chart in the top 10 for global social media usage. When such clear business value is involved, the rate of developments in Natural Language Processing (NLP) techniques within AI can be expected to accelerate. Many cloud services are already expanding to cover Thai language and offering standard tools to guide non-programmers to build chatbots themselves.
We at SCB Abacus have launched a number of chatbots this year. One of our first chatbots, "Puek Hom", is an expense tracker chatbot on Facebook Messenger. We learned that Puek Hom is more versatile than expected, attracting not just millennials but also Gen X users. This gives us hope in the potential of chatbots within the competitive market of personal finance. The most recent chatbot, "Perm Poon", developed for SCB Asset Management, can now handle lengthy interactions and advise on complex products, such as mutual funds.
Our experiences in developing these chatbots tell us that making a great chatbot is more of an art than science. Here are five things I want to share with Thai corporations looking to build chatbots in the future.
A great chatbot needs a persona.
You need to do research and understand target persona for an effective communication style. Diverse skills are required, such as user experience and content design, not just software development. All these skills are an essential part to a user-centric product design because they ensure that you serve your users' best interests. Humanisation is even more crucial — chatbots should be able to replicate how your best agents interact with different types of customers, especially as the bots may not always know what to answer.
A great chatbot is personalised.
A great chatbot learns from each individual's conversation history and other peripheral data and responds accordingly in real-time. Imagine a food delivery chatbot that can recommend restaurants based on previous chat history, previous orders and current location, instead of asking users to go through the same back-and-forth chat every time. Building a chatbot capable of personalisation requires technical knowledge of database architecture, and the ability to integrate internal and external data systems.
A great chatbot must be able to hold a conversation.
To keep users coming back to use the chatbot, it is crucial to keep the conversation flowing and engaging. This means that, like in a good human-to-human conversation, there is never "dead air". One way to do so is by adding value to the conversation by anticipating users' needs based on previous chat history, without being explicitly asked. A seamless experience can also help. A good chatbot should have omni-capability where conversation flows seamlessly across channels.
A great chatbot should know its own limits.
Having a chatbot for customer service is quite well-received globally, with 45% of users around the world preferring chatbots to humans, according to a 2017 poll by a market research firm Grand View Research. However, this will differ from industry to industry, as it can be difficult to create a personalised experience for certain products. So where do you draw the line? It goes back to the first point I made about knowing your target audience. Often it can be difficult to know for certain before the chatbot launches. Therefore, one must have be quick to reiterate the chatbot in short periods to retain and engage customers.
Developing and maintaining a great chatbot is a long journey.
Like other AI products and especially NLP tools in Thai, a chatbot needs continuous improvement and huge data training process to enhance its ability to handle various types of conversations. It is also important to note that a chatbot cannot solve everything. Companies such as Amtrak has created an additional chat-tree where, if reached, users can get in touch with a real customer service representative for questions that require human attention.
A great chatbot is a significant investment in terms of cost and time. It is arguably more beneficial to companies with large customer bases, but it can also benefit smaller companies looking to expand from niche to mass markets.
Not only will we see chatbots becoming the interface of choice, but they will also continue to disrupt our search behaviour. Currently, Siri and its Android counterparts are capable of offering a list of general information, but lack domain expertise to provide "the one right answer". On the other hand, chatbots that businesses build will possess such domain expertise, but how do we ensure that those chatbots are discovered?
Instead of having to browse through a list of information provided by Siri, perhaps we will see a future where multiple chatbots will collaborate with one another to deliver that one right answer straightaway.