Robotic bots are dead, long live the conversational bots

Have you found any conversational bot in Facebook Messenger lately?  While many companies are putting efforts in developing bots to communicate with their clients they seem to have forgotten that UX is the real key to gain people's loyalty.

Why is the conversation feature so important? Because what customers really want is to converse with a bot or with a Virtual Assistant as if they were talking to a human being. They want the possibility to choose something outside the provided menu or collection of buttons, and they want to be able to change their minds throughout the interaction without having to restart the process.

Next question is: Is it possible to build a conversational chatbot that understands complex requests and personalized orders like "I don't want my alarm to sound on Sundays"? By our experiences with Alexa or Google Homepod it certainly wouldn't seem so.

Complaining is very easy and we can describe the lack of conversational bots as a problem, but what really matters is whether we have a solution. We can say that at Bitext we have developed one. We have created a fully conversational bot.

Creating a conversational bot step by step:

Since major chatbot platforms such as, LUIS or Amazon Lex fail to fulfill the two main requirements through standard training: understanding natural language and responding in a human way; at Bitext, we have developed a several-step process that dramatically improves the chatbot performance.

The first step is training a chatbot with the intents and entities relevant to the area of interest. Our training consists mostly of supplying the bot with very basic training templates, rather than millions of full sentences. For example, if we are working in the tourism sector, our training templates might look like the one below:

{ACTION}{OBJECT}{PLACE}  this template covers sentences like:

I want to book a flight to San Francisco

I need to cancel my hotel reservation in Paris

Once we have our chatbot ready, the second step is using our middleware to simplify the input given by users. In this way, sentences like:

  • “I want to book a flight to San Francisco”
  • “Can I book a flight to San Francisco?”
  • “Can you please help me to book my flight to San Francisco?”

These are all mapped to the bare sentence: “book flight San Francisco” which contains all the relevant information and leaves out any confusing structures that generally present problems to these chatbots.

Our middleware helps the chatbot work around the usual problems that standardly-trained bots fail to solve. Some of these challenges are:

  • Complicated long structures: “could you please help me change my flight to San Francisco for one to Los Angeles?”
  • Coordinate sentences: “I want a restaurant that serves sushi and steak” vs “I want to find a sushi restaurant and a steak house”
  • Negation: “book a hotel that is not near the airport”
  • Double intent “I want to buy a flight to San Francisco and book a nice hotel in Santa Barbara”Untitled Diagram.png

Now that we have a simplified output and a bot trained to understand it, we can easily integrate these two services with Facebook Messenger using Chatfuel. As seen in the chart below, we create several blocks in Chatfuel:


  • Greets the user in Messenger and asks how we can help
  • Returns the user input to Chatfuel
  • Sends the user input to BLOCK 2


  • Sends the user input to Bitext webhook
  • Our webhook sends the user input to our middleware: Bitext API
  • Bitext API returns the user input, now as simplified input to the webhook
  • The webhook sends this simplified input to or whatever platform we have chosen to build our chatbot.
  • ai returns an answer to the simplified input
  • The Bitext webhook returns the platform’s answer to Chatfuel
  • Chatfuel sends that answer to Messenger where the user receives it.

Building two simple blocks in Chatfuel we have managed to integrate the Bitext middleware into Geoffrey, a bot created in a major platform with Facebook Messenger, and now our users can interact with a fully functional chatbot.

If you want to see it live just click here!


Try Geoffrey 


Subscribe Here!