Home About

Enabling Smart Conversational Analytics

Role

Senior Product Design Consultant

Company

USA's Leading Telecom Company

Team Size

7

Duration

6 Months

Overview

Considering confidentiality with the company, i would not reveal the name of the product or any original details.

After creating a marketplace where teams across the company could collaborate and exchange reports, business wanted to give the power of reporting to fingertips of any user who wants a quick performance report.

In response to that, i have designed a smart analytics bot (let us consider it as SAB) that converses to the user in terms of metrics and reports.

The CHALLENGE

With nearly 50+ teams who use the analytics marketplace would have access to the SAB. Also, in addition to that C-level to upper leadership would have it integrated in their personalised dashboard to check quick metrics of the company. How can the data be retrieved and displayed as a response to small yet complicated requests human ask for? How does the natural laguage be responded in terms of metrics?

Users

Business Analysts

Marketing Managers

Leadership

The Approach

NLP Implementation

Before i designed the chatbot, i performed a discovery to understand what are the type of reports that users are primarily searching for. What are the data sets available? And what data sets needs to be requested from across the organisation?
The Datasets

I have first requested for the set of words that users used to search for reports in the marketplace. Then we created four focus groups from BI, BA, Marketing and Business teams. We have requested for a set of words that would meet their daily reporting goals. We have collected set of approximately 3000 words.

The Metrics

Later we have analysed the different types of metrics, the marketplace consists of and prepared a list of metrics. By sharing the metrics with the focus groups we also requested for additional metrics the teams daily track.

By obtaining this data i have designed different visualisations each metric would need based on conditions as below.

  • Visualisations for metrics with only measures
  • Visualisations for metrics with only dimensions
    • Visualisations for metrics with dimension by date range
    • Visualisations for metrics with dimension by another dimension
  • Visualisations for metrics with both dimensions and measures
    • The Design

    Along with designing the metrics, i have also created a responsive design system for the prompts in the chatbot. By following the brand guidelines set by the company, i have created a set of components for the product to dynamically use them.

    The Algorithm

    After setting up the natural language processing and design system, we have trained the algorithm to recommend metrics and actions based on the inputs given by the user. Such as,

  • Suggesting a relevant metric after displaying a metric.
  • Suggesting actions after displaying a metric
  • Scheduling and suggesting actions based on the inputs

    • An example shown below:

    The results

  • Integration to 70+ teams' dashboards across the globe, including C-level leadership.
  • 2-3X more footfall to the marketplace after 2nd month of integration.
  • Algorithm re-usage to search results in marketplace.