Statista reports that the size of chatbot market is was 113 million USD in 2015 and projected to be 994.5 million USD in 2024 McKinsey reports that in 2015 30% of all customer care interactions took place through chat, social media, and email; this is projected to rise to 48% in 2020.
While chatbot use and investment are booming, they have a long way to go. Currently, the most popular chatbot platform is Facebook messenger, which has over 100,000 chatbots embedded into the messenger. Recently, Facebook launched its own chatbot using its developer kit called “Assistant M;” however, it has a 70% failure rate. This is ultimately because of lacking technological advancements within Artificial Intelligence.
Chatbots of the future will have advanced capabilities in five key areas: natural language processing (NLP), natural language understanding, contextual awareness, anticipate customer needs, and sentiment analysis.
Natural Language Processing (NLP)
Natural Language Processing is the process a machine goes through when translating, summarizing, contextualizing, and analyzing text – or the same process that Google Translate uses to translate text. Chatbots with NPL should understand language in plain text or speech including intent and parameters. They should also be able to process the intent with parameters and execute the next action; this usually yields response or another question by user. In addition, bots should maintain the context and its state with all parameters during single session in order for user to get result he/she is looking for.
Natural Language Understanding
With natural language understanding, developers can analyze semantic features of text input such as categories, concepts, emotion, entities, keywords, metadata, relations, semantic roles, and sentiment. Web developer kits like IBM Watson’s chatbot kit allows for easy use of NLU.
The machine does so by extracting keywords and phrases from a user’s inputted text that help the bot perform a command. For example, a weather chatbot like Poncho extracts the user’s location based on what kind of text the user has inputted into the bot.
Chatbots that are able to extract user data such as users do not want to repeat themselves when interacting with a chatbot For example, Mode.ai, a chatbot that functions as an online personal shopper, allows for a more personalized shopping experience by giving clothing recommendations on Facebook Messenger’s chatbot platform.
As Mark Zuckerberg said at the latest F8 developer conference, “I don’t know anyone who likes calling a business. And no one wants to have to install a new app for every business or service that they interact with. We think you should be able to message a business, in the same way you would message a friend.”
Anticipate Customer Needs
Chatbots of the future will be able to obtain user information seamlessly. For example, users of today won’t want to forego a series of questions when interacting with a chatbot – the bot of the future will already have the user’s browser history, social media information, past purchases, etc. Users tend to go to chatbots for simple questions, product recommendations, or to quickly purchase or book something.
If a user can easily click one or two buttons and reorder something, they will prefer not to speak to a chatbot or virtual assistant. Alternatively, if a consumer used a bot to book flights, the bot would most likely ask the user too many questions and would defeat the purpose of the bot. While advanced natural language processing is important, most people do not go on the internet to simply chat – they want the bot to effectively execute whatever task it is asked to do.
Sentiment analysis includes parsing out specific target phrases and the sentiment of the entire text. Sentiment analysis takes an input string that the user has typed into the bot and assigns it to a sentiment rating range [0-4]. Chatbots use machine learning, NLP, and statistics to identify and extract value from text. For example, product reviews tend to be easier to analyze whereas books, movies, art are more difficult because they are more open ended and harder for the bot to categorize.
Sentiment analysis functions off of Support Vector Machines (SVM) which is a type of algorithm that performs pattern recognition to analyze the tone of the user input. SVM is generally good at combining diverse information sources and does not assume feature independence but are generally sensitive to sparse and insufficient data.
Even major banks like JPMorgan Chase are looking to the future with their use of chatbots. COIN analyzes complex financial contracts and effectively reduces loan-servicing mistakes due to human error by lawyers. COIN was able to save JP Morgan 36,000 hours of manpower last year. In the future, banks could use chatbots to analyze credit-default swaps and banking regulations.
Chatbots of the future will have the intelligence to effectively negotiate and the ability to reason in a human-like way. Facebook recently launched its own Facebook messenger chatbot called Assistant M, as well as a open source developer toolkit for bots. One of the most notable releases within the toolkits were the ability for bots to negotiate through a feature titled “dialogue rollouts,” in which bots could think ahead and decide best strategies for negotiation with humans or other bots.
These bots were put forth in Amazon’s Mechanical Turk, or an online crowdsourcing marketplace that enables individuals and businesses to bargain. Most users believed that they were interacting with a human and not a bot. We’ve heard of bots who have the ability to play games like chess and Go; however, improvements within the ability for bots to reason and negotiate through dialogue will be key in the future.
If you liked what you read, please share it with friends.