From virtual assistants to messaging tools, chatbots are all around us. Almost by stealth, these algorithmic tools have become deeply embedded in the lives of people who often have little idea of what they are or how they work.
For the uninitiated, a chatbot service accepts input strings like speech or text. It uses artificial intelligence to match database contents with that particular input string, then responding accordingly. Common examples of chatbots include voice-controlled assistants like Siri and Alexa, or the “can I help?” pop-up boxes on web pages. Facebook alone has over 30,000 active bots, while experts predict virtual assistants will increasingly dovetail with smart technology to control domestic appliances and electrical systems. Scheduling and shopping are expected to be other significant growth areas for bots in the coming years.
The reason these tools have crept into our lives with such little fanfare is because chatbot development often focuses on making them as human-like as possible. While some operate following pre-determined rules, others use machine learning to evolve and expand their knowledge – like humans. Both systems pose a number of challenges for chatbot development teams.
Here are six common pitfalls to avoid when creating your own bot:
- Avoid weaker APIs. Platforms for chatbot development include IBM Watson, Live Agent and Microsoft Bot Framework. Watson has a billion Wikipedia words in its directory, understands user intent and comes with various developer tools. There’s no reason to use a limited or basic builder when such powerful tools are freely available.
- Don’t assume people know what to do. Many people are unfamiliar with chatbots, and a survey last December revealed 73 per cent of us wouldn’t use one again if we had a bad experience. As the public face of your company, it’s crucial to provide step-by-step instructions plus a prominent ‘help’ button to reduce any confusion or frustration.
- Make it clear audiences aren’t talking to a person. According to Mindshare, three quarters of people want to know they’re dealing with a bot while half of us think personalised bots are “creepy”. Giving bots human names or ‘borrowed’ photographic avatars is inadvisable, but adding character and humour will usually be well received.
- Never forget what the bot is designed to achieve. There’s no point developing a chatbot if it won’t improve the user experience in some way. Don’t get sidelined with cool features – instead, draw an interaction flow chart to determine what actions the bot can complete. Minimise the length of input strings and focus on desired outcomes.
- Don’t use bots to replace people. Chatbots should only ever be a supplement to human interactions, with scope to delegate a conversation to a person. Consumers will be disappointed if bots operate in a vacuum, misunderstanding input strings or failing to respond to a question. Have customer services staff on standby, ready to take over.
- Never launch a bot without extensive beta testing. Remember Tay? Microsoft probably hopes you don’t. Its Twitter-based bot described Mexicans as an evil race and gave the Holocaust 10 out of 10. Despite being provoked by malicious users, Tay demonstrates the importance of rigorously testing input fields before going live.
It’s predicted that by 2029, chatbots should be able to convince people they’re talking to a real person. Of course, that might not be welcomed by consumers who may launch a backlash against automated communications. Nonetheless, it demonstrates the rapid pace of chatbot development, with huge improvements in AI expected throughout the next decade.