Building A Customer Service Chatbot In Just Ten Steps

No Comments

Have you been thinking about the possibility of utilizing chatbots in your business?  Perhaps you understand that bots would relieve the workload from contact center agents by helping answer customer questions and solving issues. The idea sounds great to many who do not have the slightest idea where to start.

Below are some ideas meant to help you start planning to build a customer service chatbot.

1.  Identify the role of the chatbot – The first thing to do is set goals, identifying what you want your chatbot to do.  Think about the questions that are answered in your contact center.  Which questions are the top 20 percent answered, and make up 80 percent of the volume of questions?  Do you want your chatbot to very narrowly or broadly interpret questions?  Narrowly deflects fewer questions from the contact center but will be very precise.  Broadly will deflect more questions but may answer incorrectly.

2. Search for and select a channel – Text based bots live on any channel (communication) which can carry dialog. It can be social networks such as Twitter, chat embedded on a website, a traditional mobile carrier channel (SMS,USSD), or a fresh channel.

3. Design the chatbot’s conversational architecture – Keep in mind that user messages are part of a larger conversation that cannot be analyzed in isolation. Bots are about conversation that has the possibility of any number of responses between user and bot.  Figure out how you are going to handle interactions which might lead to follow-up steps which refer to previous communication between bot and user.

4. Create storyboards and dialog flows – This step involves you organizing your content. You must start thinking about how to best word the chatbot’s answers.  Going into detail with the dialog flow is what your developer will need to implement the bot.  It represents each juncture and branch of the conversation.  For frequently occurring dialogue, you will want to designs variations of the same message.  This random prompting is important in order to make the user’s experience feel more human and less robotic.

5. Gather chat data – Think about the variations of questions asked by your customers. Collect these questions from as many different agents/customers as possible.

Customer Service

6. Now for the integration – Data and backend integrations can be as varied as the applications you are looking to automate.

7. Select a development approach and a platform – Key tasks for most chatbots are to figure out the intent of the sentence, and to take date from the sentence. There are basically two approaches to these tasks.  Either based on creating rules from the top down, or by using machine learning algorithms to learn the task from a collection of transcribed communications.

8. Engineer the natural language understanding and implement the natural language understanding – If you are working with a linguistic rules based platform, the rules you craft will represent the characteristics that determine that a certain sentence bellows to one intent or the other intent. If you selected a platform based on machine learning, you will provide your example sentences for each possible intent.  The more examples, the better the algorithm will learn the variations used for each intent and will be able to better distinguish between the intents.  Remember to save some of your example sentences for the next step of testing.

9. Use case detection testing and revision – Your collection of example sentences will now be used for automated testing. You will want human tester diversity for this real user testing.  Test and revise until you are happy with the accuracy level.

10. Deployment and revisions – In this stage you will want to reword your bot’s responses as you listen to questions from your customers asking for clarification. You want to avoid customers having to ask unnecessary questions.  Gather the data, review the data, and apply the data to your chatbot’s design.  Don’t forget to log all of this for future reference.


About Gregg Troyanowski

Gregg Troyanowski is president of Promero, Inc.  Founded in 2001, Promero is a leading customer care -call center software expert.  Promero provides valuable insight to customers when selecting a call center technology platform. Promero supports companies of any size or industry and addresses strategic, operational and technological issues always with the focus of providing a solution that is right for the client’s business.  Promero is an authorized managed service hosting provider and reseller of the world’s best in class solutions including Oracle, Aspect, Interactive Intelligence Vocalcom, Five9, CallMiner Speech Analytics, Salesforce, Pipkins, and Riverstar.  Promero’s client list includes companies on Fortune’s Most Admired Companies list.  If your business is considering an application enhancement, replacement or in need of technical support, please contact Promero for a free, no obligation consultation.

Leave a Reply

Your email address will not be published. Required fields are marked *