Chatbots Are Not For Everyone
Do not build a chatbot because you want to build one. Build a chatbot because you want to engage customers.
Posted in Thoughts on September 11, 2017 – 2 min read
The view and expressions on this post are my own and do not necessarily represent the postings, strategies or opinions of my employer.
Chatbots provide a zero-knowledge interface to clients. They can just ask what they want and get a human-like reply, rather than forms and buttons.
Several good examples of chatbots are: Cleverbot, Microsoft’s Zo and Tencent’s Baby Q
How to Build One
To build chatbots, you can take two approaches:
- Build with an existing bot framework
- Build it from scratch
Examples for (1) would be Wit.ai, Api.ai, Microsoft Bot Framwork etc. These frameworks provide pre-trained functions such as named entity search, e.g. find the location and date within the conversation. The cost for deployment is relative lower. You can plug your bot to any messenger service such as Facebook Messenger, Telegram and even Whatsapp! Do check out two Hong Kong based AI chatbot companies: Mindlayer and Claire.ai!
(2) means that you are training the chatbot by your own model. It is highly customisable as in you can, e.g. building a conversation-aware bot rather than a rule-based one. It can be trained specifically on very industry-specific domain knowledge. Say for the insurance industry, the chatbot can identify customers’ risk level by asking simple questions. In commercial sense, (2) is more for R&D of new products or for startups.
Don’t Built One If…
Do not build a chatbot because you want to build one per se. Build a chatbot because you want to engage customers.
You can always build a chatbot that can only disply canned responses and stock images. It is the training and data feed that matters for the chatbot.
When you want to roll out a chatbot solution, ask yourself the following questions:
- What kind of services do you want to integrate into the chatbot?
- Can the user make out what the chatbot can do in 1 minute?
- Can you derive useful user insight from the chatbot conversation?
- Do you have ad-hoc / follow-up services after the user leaves the chatbot?