Chatbots are the talk of the town with the announcement of Facebook to launch their Messenger platform with chatbots in April 2016. Here a short summary of why we need it, what shapes and forms a chatbot can take and where to start with it.
Why do we need chatbots? Simple answer is simplicity; When I want a simple question answered, I do not want search for my specific App and search within the App for the information I need, or browse through a set of pages on a website.
A couple of years ago I was working abroad, and with a phone with roaming costs you do not have internet always continuously at hand. I’m supporting my football team Ajax, and wanted to know the scores whenever they played. So I used a (Dutch) tool (NOS teletext), punched in 818 (games played today) and got me at a blink the game scores at bare minimum network.
Chatbots are about the same, just give my information or help me with an activity in the most basic configuration, what is my flight info, what are opening hours, I’d like to transfer money etc. This is all done from a command line type of interface, which could be Facebook messenger, Slack, Telegram, Text Messages, etc.
In summary a chatbot delivers a simple chat interface, and enables a conversation that can be guided by rules and potentially with artificial intelligence.
Just like Apps and Websites, chatbots can take different shapes and behaviour. The three most common forms a chatbot can take are Reactive, Scheduled and Predictive:
Reactive chatbots deliver their info in a predictive mode, just like the Airport service bot shown below. These reactive chatbots are guided by a fixed menu or a Rule engine, delivering answers based predefined questions. Upfront the chatbot designer tries to de define the questions that need to be answered by the chatbot. This works fine in the majority of customer questions, or simple transactions protocols.
Scheduled chatbots deliver their information in a set timeframe, every morning I need to know the outstanding orders, who filled in their time forms too late or just deliver the main news of the day. Here an example of Digg, a news feed providing me every morning with the latest news.
Predictive chatbots deliver their information based upon behaviour knowledge or previous conversions (with you or other persons). These chatbots are the ones everybody is interested in, and also are known as the Artificial Intelligence type of chatbots. The intelligence can be added as part of a conversation, the bot tries to understand your questions, which are free format, and gives personalised answers. Or the chatbot can give you information outside a conversation, for instance in your daily rhythm of going to work and trying to avoid the traffic jam, it can send a message when you walk to your car that the situation on the road is good (or bad..).The chatbot is here a learning system. This still is in a experimental stadium, and loads of errors are made in the implementations with the bots from vendors providing racist answers.
Where to start?
As tempting it may sound to build your own AI system answering every question, there a multiple use cases that with the help of a simple reactive chatbot can improve the interaction between the end-user, callcenters and the backend systems.
Improving customer interaction
Starting with the very basics can already improve the interaction with customers. Customers want to know the opening hours, is a specific item in stock at the store near to me and what is the status of my order. When you include these conversations already 70% of the customer questions is helped, the other questions can be relayed to the callcenter, making the effectiveness of the callcenter larger.
Interactions with external parties that are not formalised
Most companies have some sort of ERP system sitting in the backbone handing financials, ordering and payments. A lot of communication with external parties is handled via email, phone and letters. Chatbots help simplify the ad-hoc interaction with the external parties. When an order is placed with an external supplier, the supplier wants to payment status, and get information handed over from the financial system about the order payment. This transparency can aid in a better trust relationship with the external parties and enables a faster interaction between the parties.
Designing your chatbots, the conversation designer
With a website or an App you can divert your audience with a beautiful look and feel. With a chatbot you’re thrown back to the bare basics, with no look and feel is involved. It all boils down to a conversation with your end-user, and in some cases it needs to be a damn good one…
This all asks for a renewed approach for User Interaction design, which is conversation focused and learning from previous conversations. What type of people are you dealing with? Since a conversation involves two parties, you should know who you’re interacting with when designing the conversation, and also what bot platform should you focus on, Facebook messenger (mainstream public), Telegram (privacy focused audience), Slack (techie focused), WhatsApp (mainstream), Line (Asia), WeChat (China), people who still like to go to web pages (oldies ;)) etc.
In the next blogs we will dive into (Oracle) technology related to chatbots. See already first experiences with the Oracle chatbot delivered by the Oracle A-Team.
Léon Smiers, Soham Dasgupta, Jan-Willem van Doornspeek
Information Analyst and responsible for the www.politie.nl website development and active member of the Capgemini Center of Excellence Oracle PaaS group