In 2016, Facebook Messenger announced they will open their Messenger application to bots. The phenomenon chatbot was born.
- Who has not Messenger or WhatsApp on the first page of his smartphone?
- Who doesn’t check his messages every day?
- Who has notifications disabled?
The messaging services, it is, for the Quadrimester 2016:
- WhatsApp: 900 million users;
- Messenger: 700 million users;
- foreach: 30+ billion messages.
Many enterprise want to be were their customers are. But hire employees to respond to this mass of messages should be to expensive. It explains the chatbots’ growth. And the best thing is, most part of its api are free!
What is a chatbot?
I need to understand the messages!
This messages can be a little bit ambiguous. To understand the user, the bot need to use NLP (natural language processing) or NLU (natural language understanding) .
I do manage the conversation and a unlimited number of questions
Prima facie, we could think to use Regex. But imagine, the thousand of rules you need to do (worst, they are not scalable…). The lost of time! So, the fast and easy way (but not the only, it exists some “bot Regex”) is to use a A.I. Some are already build and free, like Siri , Home or Alexa . This AI have many features, like NLU to understand the users’ intensions.
I have to answer
This one is directly connected with the previous point. You have to create something like a “user’s story”. To do this, Recast should be very helpful. But currently, Skynet doesn’t exist! (We are saved!) And a bot cannot do everything. So you can use a customer service Zendesk/ , Intercom or anything else, to help your bot then he is in trouble.
I want to learn with each interaction
This is another work of the AI like build his own rules. Humans can do it but it needs so many time (and money). The feedback is really important to perfect the bot. “Did you like our conversation? My service?”. Smileys could be really helpful too.
Traps and good practices
You can implement buttons like in a classic website to lead the story user but may be should open your mind.
- It is not necessary to answer to all the questions. “Sorry, this not my work, i am here to help to…”
- Just some questions are importants. To find them, you have just to detect the intentions.
- To achieve its goal, a user should never answer to more 5 questions. After, you are pretty sure to loose him.
- For a better user’s experience, if the bot has many answer, you should simulate writing. It lets the time to read the first answer before the next one is given by the bot.
Imagine a bot…
The user start a chat with the bot. The bot can have the user’s email, name, firstname, gender, birthday and picture form Facebook. So, you can know if this user is one of your customers.
A story user:
- User: “I want to loose weight…”
- Bot: “Ok Benjamin! To help you, i just have few questions to ask.”
- Bot: “1- Currently, how much do you weight?”
- User: “103kg”
- Bot: “Thanks!”
- Bot: “2- And how much want you loose?”
- User: “23kg…”
- Bot: “Ok, Ok.”
- Bot: “How many time?”
- User: “3 month?”
- Bot: “I can help you!”
Here, the bot can send a link, view a little description to an article in your website. If this user is already your customer, he will use more you application. If he is not, he may be a new user, and this new user has tested your service without downloading your ap! Other possibility, if sometimes you send articles to your customers, you can send them with your chatbot in a messaging service.