INTRODUCTION

We are given a purchases/returns data set. Our primary objective is to find the return rate per month.

How do we define return rate per month?

The way I defined Return Rate is as follows:
Say June had a total of 10 orders.
Of the 10 orders, 1 was returned in July, and 2 were returned in August and there were no returns in the subsequent months.
So,
Total Orders = 10
Total Returns related to orders from June = 1 (July return) + 2 (August return) = 3
Return Rate % = (Total Returns from orders placed in Month M / Total number of orders in that Month M)*100

Thus,
Return Rate for June = 3/10 = 0.3
Return Rate % for June = (3/10)*100 = 30%
We shall achieve the same over the entire dataset at a later point in our analysis. First let us indulge in some Exploratory Data Analysis.

Note :
All the plots are interactive and can be zoomed by click dragging a square over an area of choice.
You can hover over the items in the charts to get the associated values.

Before diving into the analysis, let us take a look at the data:

dateordered datereturned orderstatus orders
2016-12-31 2017-01-24 returned 1
2016-12-31 NA complete 8
2016-12-30 NA complete 14
2016-12-29 NA complete 14
2016-12-28 2016-12-29 returned 1

So we have
* dateordered which is the date on which a product has been ordered
* datereturned which is the date on which a purchased product has been returned.
* orderstatus which tells us whether a particular order is complete or if it was a return.
* orders which tell us the number or quantity of the products in the picture.


Let us understand the date range over which the dateordered is present :

## [1] "Earliest Date :"
## [1] "2016-08-01 UTC"
## [1] "Latest Date :"
## [1] "2016-12-31 UTC"

So the dateordered varies from “2016-08-01” to “2016-12-31”

Now let us look at how the total number of orders varied over time


TIME SERIES ANALYSIS

As one would expect, there is a steady rise in the number of orders placed between November - December.
This could be attributed to the holiday season - thanksgiving, christmas new year etc. Also since it is winter, people would want to probably feel nice and warm on a nice king size Casper bed.

Now let us see how many complete orders and returned orders are there in all.


COMPLETE ORDERS vs RETURNED ORDERS


Based on the above chart (Hover for values), there are a total of 1033 complete orders and 60 returned orders.

Since there are 60 orders which have been returned, I will try to understand after how many days from the date of the purchase have these products been returned, and if there is a trend.

DURATION BETWEEN ORDER DATE AND RETURN DATE

dateordered datereturned orderstatus orders diff_in_days
2016-12-31 2017-01-24 returned 1 24
2016-12-28 2016-12-29 returned 1 1
2016-12-27 2017-01-28 returned 1 32
2016-12-27 2017-01-25 returned 1 29
2016-12-27 2017-01-15 returned 1 19
2016-12-27 2017-01-05 returned 1 9
2016-12-24 2017-01-20 returned 1 27
2016-12-24 2016-12-30 returned 1 6
2016-12-24 2016-12-29 returned 1 5
2016-12-23 2017-01-01 returned 1 9
2016-12-20 2016-12-25 returned 1 5
2016-12-20 2017-01-08 returned 1 19
2016-12-16 2017-01-23 returned 1 38
2016-12-16 2017-01-20 returned 1 35
2016-12-10 2017-01-17 returned 1 38
2016-12-10 2017-01-07 returned 1 28
2016-12-10 2016-12-15 returned 1 5
2016-12-10 2016-12-10 returned 1 0
2016-12-10 2016-12-20 returned 1 10
2016-11-30 2016-12-06 returned 1 6
2016-11-28 2016-12-15 returned 1 17
2016-11-19 2017-01-06 returned 1 48
2016-11-12 2016-12-10 returned 1 28
2016-11-12 2016-11-15 returned 1 3
2016-11-12 2016-11-16 returned 1 4
2016-11-05 2017-01-23 returned 1 79
2016-11-04 2017-01-07 returned 1 64
2016-11-04 2016-11-14 returned 1 10
2016-11-03 2016-12-11 returned 1 38
2016-11-02 2016-12-30 returned 1 58
2016-11-02 2016-11-20 returned 1 18
2016-10-29 2017-01-12 returned 1 75
2016-10-28 2016-12-11 returned 1 44
2016-10-20 2016-12-22 returned 1 63
2016-10-19 2016-11-10 returned 1 22
2016-10-16 2016-11-15 returned 1 30
2016-10-13 2016-11-05 returned 1 23
2016-10-08 2016-12-29 returned 1 82
2016-10-06 2016-10-24 returned 1 18
2016-10-06 2017-01-20 returned 1 106
2016-10-04 2016-10-17 returned 1 13
2016-09-29 2016-10-01 returned 1 2
2016-09-28 2016-10-19 returned 1 21
2016-09-23 2016-12-19 returned 1 87
2016-09-19 2016-10-06 returned 1 17
2016-09-12 2016-11-01 returned 1 50
2016-09-04 2016-12-09 returned 1 96
2016-09-03 2016-11-24 returned 1 82
2016-09-01 2016-11-27 returned 1 87
2016-09-01 2016-12-03 returned 1 93
2016-08-29 2016-09-23 returned 1 25
2016-08-18 2016-10-14 returned 1 57
2016-08-11 2016-09-10 returned 1 30
2016-08-11 2016-12-11 returned 1 122
2016-08-09 2016-10-27 returned 1 79
2016-08-09 2016-10-29 returned 1 81
2016-08-06 2016-11-13 returned 1 99
2016-08-06 2016-11-15 returned 1 101
2016-08-06 2016-11-25 returned 1 111
2016-08-03 2016-08-25 returned 1 22

HISTOGRAM

What does above plot tell us ? (Hover over the bars for values)

This tells us that:

  • 21 orders were returned within 0-20 days of purchase.
  • 17 orders were returned within 21-40 days of purchase.
  • 5 orders were returned within 41-60 days of purchase.
  • 5 orders were returned within 61-80 days of purchase.
  • 8 orders were returned within 81-100 days of purchase.
  • 3 orders were returned within 100-120 days of purchase.
  • 1 order was returned within 81-100 days of purchase.

After a little bit of digging, what I found was that on the website: https://casper.com/faqs/the-dog-mattress/tell-me-more-about-the-100-night-trial

It states a mattress could be returned within 100 nights of use.
So I don’t know how to explain how/why 4 products were able to be returned after the 100 day limit.
This is something one should investigate. :)

RETURN RATE PER MONTH


As can be seen from above plot, the Return Rates are as follows:
* August = 6.79 %
* September = 6.29 %
* October = 5.55 %
* November = 5.30 %
* December = 4.81 %

So it can be concluded that the return rate has constantly decreased.
This is good news provided that number of completed orders have gone up while number of returns have gone down.

Let us confirm this by further exploring the data.

Since there is no data for the orders placed in January 2017, the metric of Return Rate for January 2017 doesn’t exist.

BUYS & RETURNS

Nomenclature :
OMonth = Month in which order was placed.
RMonth = Month in whhich order was returned.

Similar explanation follows for ODay and RDay.

HOW COMPLETED ORDERS PLACED CHANGED OVER TIME

So the number of completed orders placed from August to December have progressively increased.

Few of the possible explanations/reasons:
- People recieve holiday/end of year bonuses.
- Winter is Coming! , people need nice new warm comfortable bed to sleep in.
- Holiday season is approaching , so people might want to gift to their friends/family.

HOW RETURNED ORDERS CHANGED OVER TIME

i.e in which months are previously purchased products returned.

The number of returns from August 2016 to December 2016 have also increased and again falling slightly in Jan 2017.

However please note that, though the absolute values of returns are increasing, the return rate % is decreasing.

BUYS AND RETURNS BY WEEKDAY & MONTH


Let us examine if there is any trend in the purchase/return trend based on a weekday of a month.

COMPLETED ORDERS BY WEEKDAY & MONTH


RETURNED ORDERS BY WEEKDAY & MONTH

Another way to visualize the returns per month is using a Sankey Network Diagram.

This helps us understand the flow. I.E for products bought in August when are they returned, similarly for September until December.

SANKEY NETWORK FLOW

This is for understanding in which month were products returned based on the month in which the products were purchased

How to understand this plot? Hover over any flow originating from a node, say the August node, and the 1st flow. Hover over it.
It displays : Aug -> Aug 1 units .
This means of all the products bought in August, 1 was returned in the same month of August.

Now hover over the 2nd flow.
It displays: Aug -> Sep 2 units.
This means of all the products bought in August, 2 were returned in the month of September.

Similary you may explore for the remaining months in the Sankey Diagram.Other possible conclusions: most returns are around start of black friday and new year etc.i.e products being bought during non offers/ non seasons, are mostly being returned during or at the start of the offers seasons like Black Friday and New Year.

CONCLUSION

The return rate decreases from August -> December.

Future Prospects:

Identify why are the returns happening?

Returns reduce our bottomline, as the returned products are either recycled or donated to charity.(Acc. to Casper website)

Making note of the financial loss that product returns represent, it is not practical to accept the situation without acting. Businesses should take a hard look at their product return rate and explore ways to address this issue.

Companies are now challenged with taking a proactive approach and implementing a program that will help reduce returns, improve profitability and protect their reputation.

How to investigate returns?

  1. Before recycling or donating returned product, thouroughly investigate product quality.
  2. Understand consumer requirements based on location etc, so that we can tailor a product that would best suit their needs. 3. Take consumer feedback on products that are being returned most frequently.
  3. Run regular reports to identify trends

The End.

Thanks,
Vivek Mangipudi

https://www.linkedin.com/in/vivekmangipudi

https://www.kaggle.com/stansilas