Top 4 Best Practices for Implementing an IBM Chatbot

1854

So, you have identified how an AI-based chatbot can add value within your organization and you’re finally ready to start implementing it. During a 3-month project for a large insurance company, we realized that building a chatbot is easy, but building a highly-functional chatbot requires dedication, a lot of structure and teamwork, several prototypes, and consistent quality-control. I will walk you through the top 4 lessons we learned during our project, which involved setting-up a proof-of-concept using IBM Watson algorithms. My goal is to share some insight on how you – too – can take your chatbot from prototype to production.

Atlassian (Publisher). (2017, Jan). Smells Like Teamwork [digital image]. Retrieved from https://www.atlassian.com/blog/

Here Are Some Best Practices to Help You Implement a Chatbot Sustainably

1. Set clear guidelines!

Within Watson Knowledge Studio, the team of experts had to annotate the documents to link annotator components to entities in the text. Since many people will be annotating different documents, it is important to set clear guidelines. These guidelines improve the quality of the annotation being created and, by doing so, will establish synchronicity and will help avoid miscommunication in the future.

2. Keep everybody involved!

Building a chatbot relies on the help of many people within the organization. You need to always keep people involved in the building process to gain valuable feedback and to encourage change. For the insurance company, we made it our mission to make the employees feel involved and informed throughout the process. We constantly explained the project we were working on and held monthly demonstrations. It resulted in receiving a lot of useful feedback and users being motivated to train the chatbot.

3. Set up a prototype as soon as possible!

It is very possible to set up a working chatbot within 30 minutes using Watson Conversation. However, having a prototype available provided us with the relevant insight to make the improvements we needed. It showed us which formulations of questions performed better than others, and – ultimately – helped us improve on the algorithm at hand. Additionally, the prototype enabled us to make the project more useful for the employees who will be working with the chatbot and to get people excited about its contribution to being able to work more efficiently whilst still delivering the required quality.

4. Always track quality!

Constantly keep track of the quality of the different models being applied. Doing so helps to identify what models need to be modified to improve their accuracy levels.

A chatbot is only as good as the team building it. It is extremely important to work together since many people will be interpreting vast amounts of information at once. Establishing clear guidelines before you start is also critical to the success of your chatbot, as well as involving your employees or users to provide feedback as they have experience in the field, it is cricial to engage the team on this change. Since they will be the ones handling the chatbot on a daily basis, their feedback is a valuable asset to creating prototypes and to identifying which series of questions worked the best and to make the proper algorithmic adjustments. This will be a game changer when you’re trying to decide on and track the accuracy levels of the models being tested. A succesful implementation of a chatbot all comes down to the power of teamwork and establishing clear goals that the team can follow through with to be able to run the chatbot efficiently for years to come.

Stamplay (Publisher). (2017, Aug). Chatbots [digital image]. Retrieved from https://blog.stamplay.com/making-customer-support-more-efficient-with-ibm-watson-2c082415eb86

If you would like to know more on implementing and embedding a chatbot into your organization, please contact Ivo Fugers, Data Scientist and IBM Watson Expert at ORTEC Consulting, via ivo.fugers@ortec.com or +31 (0)88 678 3265 or via our website: www.ortec-consulting.com.

*This article was written in collaboration with Wim Jansen.

· · · · · · · · ·
https://ortec.com

Ivo Fugers
Ivo is a Data Scientist at ORTEC Consulting. He has a Bachelor's Degree in Industrial and Organizational Psychology from the Utrecht University, and a Master's Degree in Statistical Science from the Leiden University in South Holland.

Comments are closed.