A Social Media Chatbot's Case Study: Part 2

A Social Media Chatbot's Case Study: Part 2

Outlining the journey of a menu-driven chatbot and using the metrics derived from the communication to improve a home-appliance company's CRM system.

A Quick Recap!

In Part 1 of the series, we took a look at the problem statement provided. The first step we took was to make some basic assumptions about the foundation on which we would build our solution. We delivered a brief overview of the solution, sketched the system flowchart and user journeys, and that was about it. In this Part, we will be deriving the metrics that will contribute to improving the CRM systems of the company.


The metrics are divided between direct and derived. The direct metrics are the ones we can obtain from the communication between the bot and the user without any processing. The derived metrics are the ones that hold weighted proportionality to the direct metrics and provide the refined metrics that would help in making better business decisions and improving the chatbot in the future.

Direct Metrics (from the chatbot)

Direct Metrics (from platform engagements)

Derived Metrics (derived from direct metrics)

Legend for the above table:

  • Positive (green) - An increase in the value, adds to the overall value of the derived metric.

  • Negative (red) - An increase in the value, reduces the overall value of the derived metric.

  • Neutral (black) - Can be positive or negative depending on the value or does not affect the metric's value.


Leveraging social media platforms using the above-obtained metrics

From the metrics obtained, we will be using these social media platforms to increase user engagement and reach. We would be leveraging various social media platforms in ways described in the following diagram:

Working of the recommendation system

It uses the following derived metrics:

  • Location

  • Product demand (local) and rating

  • Price range (estimated from previous orders of the user)

  • Recent user search history

Generation of advertisements for a particular user

It uses the following derived metrics:

  • Recommender output

  • Peak hours

  • Platform specificity

    • Whatsapp/Telegram - Recommended advertisements and product surveys from the CRM are sent during peak active hours identified for the user, about the recently searched products within a week of the search. Feedback forms are sent as and when the order is fulfilled/returned.

    • Instagram/Facebook - Recommended advertisements from the CRM are displayed dynamically in the user’s feeds by the platform’s advertising algorithms.