Solution Architect,

Softweb Solutions

Ahmed Zubair

Ahmed Zubair

Subject Matter Expert,

Softweb Solutions

About Webinar

Chatbots have become much popular among enterprises lately. They allow businesses to immediately communicate with their customers, which is one of the crucial factors of business success. Enterprises are now aware of the benefits that they might reap by early adoption of chatbots. However, they lack the information on best practices and implementation strategies of these conversational interfaces.

In this webinar, we will discuss how enterprises can get started with chatbot implementation and the best practices to adopt while doing so. You will also see how these new chatbots can leverage machine learning and cognitive services to provide richer context to the user.

Attendees will also be introduced to different variants of chatbots, so that they can pick the one that suits them the best. Our experts will provide demo's of how things can and will change for enterprises once they start using conversational UX to reach their customers.


  • Conversational UX best practices
  • Bot implementation strategies
  • Chatbots with deep machine learning
  • Smart chatbots with cognitive services
  • Recent updates in MS Bot framework
  • Chatbots benefits for enterprises
  • Demo
  • Q&A session

Questions & Answers

The following are the answers to the questions that were asked during the live webinar.

Question 1: How important is control over the learning environment? Take for example one of Microsoft's early chatbot experiments that learned to be racist from its interactions with an uncontrolled learning environment

Answer 1: We know about Tay - the Microsoft bot that managed to learn all the wrong things in a single day. Luckily, it was contained.

Controlling the learning environment of NLP is very important. The incident mentioned here was of their Twitter bot, Tay. It’s a tad bit baffling that Microsoft didn’t ponder the possibility of what Tay would learn in an environment such as Twitter. Luckily, it was contained and we are sure Microsoft learnt enough from that debacle, because we sure did.

Training an NLP is like mentoring a student or parenting your kid. They observe new things quickly, so it is advisable to train the NLP with relevant intents and utterances.

Question 2: How can you build a chatbot with unlabelled data? So far, I have seen examples of building chatbots with labelled data only.

Answer 2: The role of AI development needs further deliberation in order to create chatbots that can deliver sought-after user experiences. The main goal here would be to create a more smarter and flexible AI that supports rapid application delivery with minimal hand-coding using either labelled or unlabeled data, or both.

The development approach must be one that improves on the existing human-based coding, able to maximize the collaboration between man and machine. The only way, as of now, is to use existing technologies like machine learning models, cognitive services, among others.

Question 3: What about chatbot integration with Skype for Business?

Answer 3: Skype for Business is integrated as a channel with MS Bot Framework now. This means that the same chatbot you created for Skype or Slack can work with Skype for Business, but the only downside right now is that they are only supporting the online version of Skype for Business. But I am sure that it will arrive soon for the desktop version as well.

Question 4: What will the bot do if the user input is not one of the listed options?

Answer 4: We can configure and train the NLP with conversational UX design for the bot to even provide appropriate responses in scenarios where the user is not choosing from the selected option. It is definitely achievable technically.

Question 5: Can the bot be programmed to pass the chat to a live agent based on user input - e.g. if the user is unable to enter the question in a way the bot can respond to - without the user requesting a person? Thinking about the user experience, I'm wondering how criteria are developed to increase the odds of a positive outcome/experience

Answer 5: This is possible as we showcased in the demo during our webinar for customer care chatbot. We need to train the NLP so that on detection of certain utterances from the user, it can pass the line to a human agent. Also, we can include a moderator, who can switch the line between chatbot and human based on certain user inputs.

Question 6: Can a bot be trained by analyzing chat logs between users and human responders - essentially, do they look at previous interactions (both positive and negative) to learn how to recognize question intent and generate responses which lead to positive outcome without human interaction?

Answer 6: We would need a bit more research on this. As of now, No NLP model provides a way to supply intent and utterances from outside using some APIs. If this is possible then language model can be retrained from chat conversations happened previously.