Alexa Feature Recommendations

This is a project I did in school. I always find it so delightful to find stories and build solutions around stories. Well, in statistics we call it inference.

' Data is your quest '

In this project, I had the following data. The goal is to find interesting user behaviors and provide recommendations to improve the user experience.

Shopping orders of households that has placed at least one order on Amazon through Alexa over a period of two months.
Data contains details about the product, category, order time
Data also contains cancel date, return date if any

Segmenting:

First of all, I noticed that ‘Toys and Games’ category orders that are placed by Alexa have a significantly higher cancellation rate than other types of orders. The normal order cancellation rate is in the range of 0% to 5%. However, ‘Toys and Games’ orders placed by Alexa have a cancellation rate of 54%.

There’s also an unusually high volume of toys and games orders placed by Alexa between 4PM and 9PM. Over 70% of these orders got cancelled afterwards.

Let me further explain why I found the analysis helpful.

According to the USA education guide website, the school day in elementary schools can vary but usually runs from 8 am to 3 pm or 3.30 pm. Isn't our order cancelation rate spikes after the school hour? I thus came up with the hypothesis that kids were utilizing Alexa to order toys that were subsequently rejected by their parents or caregivers in a later time.

I recommend an authentication feature that adds an additional step to place order for toys and games products during 4PM and 9PM using Alexa. This would add the chance that parents finding it out before the order went through so that they don't need to cancel the order afterwards.

I also recommend setting up an internal triggering system to alert abnormal increase of product orders or product order cancellation rates. This will allow the Alexa team to take timely actions to minimize waste and create a better user experience.

Takeaway: Sometimes looking at day over day data or slicing down to smaller groups is a good way to identify patterns and spot something that might be unusual.