Data Informed Decisions

24th March 2020 in Portfolio

Who Was This Built For?

TradeGecko is an inventory management platform that helps small and medium enterprises to build the business of their dreams.

What Did We Build?

Customer Engagement Scores

To reduce customer churn rates, we built a prediction model which gave us a customer engagement score. This allowed account managers to:

  1. Prioritize disengaged customers every day
  2. Understand the top reasons for customers to churn

It achieved an accuracy of 85% and was even used to determine team incentives.

Why Did We Build It?

At that time, our business was unsustainable. High churn of 4-5% meant less than 50% of customers stayed past a year.

A key bottleneck was overloaded account managers. Each had to manage hundreds of accounts and spend too much time firefighting.

How Did We Build It?

After preparing the data, we analyzed more than 110 hypothesis metrics and added them to our modeling dataset.

Sample Analyses

Then we ran the metrics through standard models in R (a programming language).

This included logistic regression, decision trees, and random forests (if this is alien to you, don't worry I skip over the technical details here).

Top Churn Reasons

Eventually, we found 7 metrics that contributed significantly to churn. These had an accuracy of 85% and a true positive rate of 86%.

For the customer engagement score to be successfully adopted, there were 2 other things we had to do in parallel.

Empower Teams First

I cannot stress more on how important one's comfort level with data is crucial to trust predictive models for prioritizing your work.

When the data squad was formed, our first goal was to empower everyone to create their own insights.

Classes On Data Analytics

By running basic classes on data analytics and SQL, we made it easy to ground conversations and decisions in numbers.

A Single Source Of Truth

Besides education, we also ensured that there was just one source of data within the company to prevent confusion.

AWS Postgres Data Warehouse

Before we began there were 5 different systems, each representing a part of the customer journey.

We built an automated extraction process which stored everything on a single database that was connected with our analytics tool.

Keen To Learn More?

This portfolio entry was meant to give a quick introduction to this project.

If you are curious about the insights, technical details, and more, drop me a message below๐Ÿ‘‡