Hi Dogan,

Please find the explanation below.

What is a ChatBot?
 
A ChatBot, in general, is a service that is powered by predefined business rules, scripts and artificial intelligence thru a conversational interface. With the recent rise of Artificial Intelligence, ChatBots have become a lot smarter and they understand customers more accurately than a bot could ten years ago. More and more businesses are now looking into ChatBot as a practical channel to provide instant service to their customers.

In my recent experience in helping customers implement the very first Einstein Bot, the top 3 inquiries from their chat channel can now be handled by a bot. These inquiry types make up over 1/3 of their overall chat volume! More importantly, they can chat with a lot more customers at the same time and the bot does it faster than any of their best agents. With the help from the bot on these heavy volumes but simple tasks, their agents now have more time to work on complex customer issues or value-add activities. What does this mean? Less agent handle time, less customer wait time, a better user experience and more happy customers!
 
Know Your Transactions
 
So how does it work? First of all, your ChatBot is very smart, but you need to teach it what to do. Let's think like a customer for a moment. They come to your website to get certain things done, like getting order status or making appointments. These “things” are typically related to the customer's common interactions with your business. There might be a few folks wanting to check the weather forecast but chances are most customers will likely go to a weather service website.

To sort out the pecking order of things supported by your bot, here is some homework. Analyze frequent issues that your chat agents are helping customers with every day and see if those issues, like booking an appointment or checking status are simple enough for a bot to handle. These “things,” often referred to as “transactions,” will serve as the building blocks of your bot.

Each of these transactions usually takes one or more inputs and will also have output(s). Let's say your customers are accepting an invite to a party via a bot. You would at least ask if they are coming, and if so, their name and number of guests, then respond with a friendly message “Thanks for coming.” Because you are either asking or returning information one piece at a time, this RSVP transaction eventually becomes a small dialog with multiple exchanges back and forth between the customer and the bot.
The Einstein Bot builder is the new point-and-click setup tool that allows us to build dialogs in a bot. It supports different types of tasks that can be configured in a dialog. You can Ask Question to gather information, and Send Response to return the output to the chat window. After gathering customer input on “Coming?” “Name” and “Number of Guests” you would call Run Script to save their answers to Salesforce or even external systems. You can also use Process Rules to control the flow with a certain logic. If a customer says they are not coming, you don't have to ask the other two questions.



Help the Bot Understand The Customer Intent
 
Let's say you have your top five transactions identified and are ready to go. The next question is how would the bot know which transaction the customer is inquiring about, so the bot can kick off that particular dialog. This is where Salesforce Einstein, or in this particular use case, Natural Language Processing (NLP) comes into play. Your customer may phrase the same question differently. They may ask “order status” or “where is the stuff I bought?” In each of the dialogs you have configured, a separate Intent tab gives you the place to enter these commonly used phrases by customers. Then you can “train” the bot to recognize the pattern of these phrases. The goal is even if customers ask the same question in various ways, the bot will still be able to understand their intent at run time. By the way, here comes your second homework. Build a list of common phrases or utterances for the transactions you defined in the last section.

One capability I love about the Bot Builder is that in addition to the Natural Language Processing (NLP) capability, it also provides a Menu feature so you can list the transactions supported by the ChatBot and allow customers to do one click to kick it off. While these menu driven options are a way to expose your top requests as soon as someone starts the chat, it also gives a sense to the customer on what the bot can do and they can select to transfer to agent right away for other things if it's out of the scope of what the ChatBot can help them with. My recommendation - use both the NLP and Menu capabilities. The more flexible the bot is, the better it will be for your customer. This is similar to an IVR system where you want it both understand your dialect AND recognize that you just pressed #0 on the key pad.
 
Deploy the Bot to your Web Site
 
All right, so our newest team member is now well trained and has received a perfect score on the final exam (User Acceptance Test). Let's put the bot onto the big stage. Chat buttons or proactive invites typically reside on yourweb site or mobile app, and more often than not, maintained by a different team and they would have their own release schedule as well. Guess what, the good news is we can cut them some slack this time! Your live agent button now has a new attribute - ChatBot Configuration. This is where you point it to the ChatBot you just built and that's it.
Now when a customer clicks the same chat button, instead of connecting to an agent or a queue directly, they will start talking with the ChatBot right away, and if they still would like to talk to an agent, they can use the “Transfer to Agent” action and the behavior would be same as today where they click the button to talk with an agent with the right skill set. The only difference is, the chat transcript now includes the earlier conversation between the bot and the customer, so agents can review and seamlessly continue the dialogue.

Hope this helps.

Kindly mark this as solved if the reply was helpful.

Thanks,
Nagendra