Ecommerce Analysis Project – Case Study

The Client, a large Ecommerce player in the US, wanted to delve deeper into their data and understand the website behavior that is driving sales.

In this blog, I will be taking you through how we understood the work and analysis that was done for the clients, how we ran the project and ultimately delivered the results in the form of actionable insights to the clients.

In my humble opinion, these are some basic ingredients of running a data analytics project with your clients.

Note: All client sensitive information has been hidden/changed and images are from Google demo merchandise account only

1. Gathering The Key Ingredients

It’s essential to start these projects with three key aspects – Business Problem Statement & Details, Data Sources, and Expected Output.

Adding the exact same ingredient details from the project below as an example so that this can be fully understood.

 

Problem statement:  What Online user behavior is driving sales?

Uncover web behavior of users on deeper factors apart from basic campaign effects OR new vs old OR bounce rate

Problem Details:

I think the fundamental question in web analytics is what behavior is driving sales. It could be certain traffic patterns or user paths (and where is friction on those paths highest/lowest — and for predictive/prescriptive, can we dynamically show if friction increasing or decreasing at any points), attribution models to show how different channels and campaigns factor into different touch points, or any other underlying elements/cohorts that are showing strong correlation to revenue. Obviously, this needs to go deeper than the simple new vs existing visitor segment or saying Google PPC is driving X revenue, or these pages have a high bounce rate.

Data Sources:

The main data source was Google Analytics standard only. (Will be used as GA henceforth)

Even though they crossed a billion hits +, this was the only data source causing problems in terms of getting sampled data in various reports and especially the main user flow reports like Behavior flow and Event flow. But, we had to make use of GA standard only and use it

Expected Output:

Actionable insights that they can apply to their business and test to improve conversions.

2. Asking the Right Questions & Areas to Explore

In any type of project and more so in data analytics, asking the right questions is more important than jumping to answers/solutions.

The following is some of the very early initial Q&A that actually happened. This should give some basic context to the analysis done later and also on what type of questions should be asked in the early stages.

Q: What are the TOP traffic Acquisition channels currently?

A: Organic, paid search/shopping, paid social. We have a popup with 10% coupon offer to get them onto an email list if they don’t convert.

Q: What should the ideal website visitor do after coming to the website? How is the website facilitating that to happen?

A: They should ultimately make a purchase. However, how effectively the website is enabling that and where we see opportunities for optimization is what we want the analysis to uncover.

Q: How accurate are the campaign and conversion tracking setup done? Which ones are the most effective ones?

A: Ecommerce conversion tracking is accurate. However, we are not sure how accurate the attribution is or how to properly validate it. Currently, we rely on what the individual ad platforms tell us, e.g. 30 day click through ROAS attribution in Adwords and Facebook )28 day).

 

After these, we started exploring the GA datasets and reports that they had in relation to all of the above. The latest set of reports can be seen by logging in to GA and at the left bar.

We explored multiple Ecommerce Reports while working and doing the analysis, ETL etc, but only found the right insights in a few specific reports merged together.

There are multiple blogs that talk about the reports that GA has, but very few go deep and talk about the analysis that was done and how multiple reports were merged together to bring out insights.

This is the exact part of data analysis that I really love working on. To find something that no one really knows and is hidden in data!

Read on to know how you can do the same.

 

3. Data Exploration and ETL process

In this section, we will break down the analysis of reports that we worked on extensively.

Shopping Behavior Report

This report can answer a lot of good questions that we were looking to explore, like:

1. Which stage of your website is causing high drop-offs and leaking revenue?
2. What is the behavior pattern of New vs. returning users on your website?

.

The good part of this report is that you can actually create customer segments using this too. It can be seen in the image above. Wherever customers are dropping or engaging, each part can just be clicked at and a segment created.

You can then take these segments and look at behavior reports , Demographics etc to find out which type of audiences were leaving just before the purchases.

A few major segments we created were as follows:

1. Cart abandonment users
2. Checkout abandonment users
3. Sessions with Transaction

The first and second are drop-offs at final stages and the third is the converted lot. By comparing behaviors of all the three, we found some interesting insights that you will find in the next section.

This can be done for any Ecommerce business to find some insights. Showing the process below also in steps:

1. Create segment by clicking on the area of funnel whose users you want to analyze. Here, I clicked on Checkout abandonment and saved it for all views.

 

2. Now, go to any audience/behavior report and click on “All users” segment on top left and choose the segment you just created

 

 

3. Now just select Apply and see the results only for the particular customer segment.

Follow this process to explore the entire behavior of multiple customer segments in your funnel.

 

Remarketing using GA Segments

All the segments that were created in the previous report can now be retargeted using GA and Adwords here.

The multiple segments created can be of the main visitors dropping off at various stages and mainly just before purchases, so they can be re-engaged and then eventually, converted.

The entire process is very well laid out here and should be the best guide as it’s shown by the expert, Himanshu.

 

Conversion Paths Analysis – Behavior Flow/Users Flow

Another report we wanted to go deep and explore was the conversion paths shown directly in GA reports i.e. the behavior flow and user flow reports

Both these reports showcase the paths visitors take, but the behavior flow focuses more on the content with keeping users engaged and also on the content that has issues, whereas the users flow focussed more on the origins, the source of traffic and marketing vehicles being used

 

The problem we had here was that the data sampling issue, where both the reports had only a small portion of data (<1%) as they had GA standard version.

Thus, any final output could not be used as insights as it would not be accurate.

However, it can be very well used if you don’t have sampling issues as was done for another conversion analysis that I wrote about.

 

Site Search

The Site Search is a very interesting report to look at in Ecommerce as you always have high buyer intent users information in this section.

With this in mind, we went ahead exploring Site search usage, search terms, search pages and also simultaneously, analyzing live website search results. Site search reports can be found in the “Behavior” section of GA and is a must-check for any growing Ecommerce business.

We found that only 3.4% of users use site search, but they contribute to ~14% of Total revenue.

Also, They have a ~6% conversion rate (compared to 1.67% of non-search), stay on the website for 3X-4X longer (14+ minutes vs 3 mins)

The buying intent is very high on users who search than users who don’t and are a significant enough part of the revenue.

This led us to go deep and we found multiple things where we went and did the analysis. The questions we asked ourselves were as follows:

1. There were high exit rates on certain searches etc…why? why not? How to give them what they want?
2. How many of internal search exits contribute to no shopping behavior?
3. Which products do they finally buy after searching for terms that have high exit rates?
4. Does the arrangement of products after search matter? Should they be done as per maximum purchases, max units sold?

We then took all the search terms, their metrics and their counterpart website’s pricing and structure and compared them to do our analysis.

After looking at all the reports, we extracted our insights and shared it with the clients.

All the reports could have been used very well, but the effective ones used here were “Shopping behavior” and “Site search”.

4. Extracting Final Business Insights

After going through all the reports and analysis above, we had to present the final set of actionable insights to the business that can be used well for various aspects relevant to the business requirements.

Note: All client sensitive information has been hidden/changed and images are from Google demo merchandise account only

Insight 1: Avg Time on Site and Pages viewed – comparing different segments of users

What we found: The abandoners (cart & checkout) spend an average of 18-21 minutes and 19-20 pages over their entire time on the website, whereas the buyers spend around 25 minutes viewing ~30 pages overall.We cannot predict who will convert or not convert directly here. But, we can surely tell that people showing exit intent (leaving the page, closing browsers by cursor movements) will become abandoners and not end up purchasing.

What to do:For this segment of users who are showing exit intent at the 18-21 minute range or 19-20 page mark- Trigger an offer right when they are about to exit- with some positive message like:

“Couldn’t find what you want? Try our new range of ultra sexy lingerie instead, and get a 10% discount . Because you deserve it.”

Run a split test with an exit offer and no offer, measuring the lift and subtracting the subsidy cost of the discount. This split test and offer is only run for the visitors showing exit intent , only on desktop where it’s feasible.

Insight 2:  Search Terms Pricing Analysis 

What we found: Looking at the product performance reports, we found that the average of (Avg price/product) of ALL products bought was 35.52$. We then picked up Top 34 search terms in the analysis as it contributed to 20% of entire search volume (~150,000 searches), rest is a complete long tail with insignificant search volumes hence the data picked here is statistically significant.

Out of this search segment of users, we found that the Search exit rate is high in 8 out of 10 cases in no affordability bucket. Search exit rate is low in 16 out of 24 cases in affordability bucket.

What to do:Adding more products with value less than $35.52 to search results will reduce the search exit rate and help improve conversions. This can be done to start off with for at least Top 10 search terms that contribute to ~9% of overall searches and then moving on to Top 34 search terms that hold 20% of overall searches.

Insight 3:  Search Terms Analysis 2 – Product placements and arrangements 

What we found: For all the Top 5 search terms, which contribute to ~8% of search volume and have an average of ~80% search exit rate (Overall avg = 34%), the arrangement of products is not shown as per the end products that are actually bought by your customers. For the first three, there are very less number of products (15-20), but still the arrangement matters here in the initial look by new users. The data is statistically significant looking at the long tail of search volumes and the high search exit rate contribution by this 8% of searches only.

What to do: Search-pages product arrangement can be made as per product performance in the specific E & F patterns (which are the eye movements). All top products bought for each search segment is shown in the data below in boxes. This can be done as a start for Top 5 searches as they add up for a high search exit rate overall.

Note: Testing for #2 and #3 should be run in separate time frames to avoid cannibalization of their effects.

These were some of the final actionable insights delivered to the business that was eventually taken action on.

This is how we run multiple projects for the SMB and Enterprise markets in Ecommerce and multiple other verticals. Feel free to ask more questions in the comments.

You can also book a slot with me or get in touch with me if you need to talk about your business and data analytics. Always happy to help.

 

 

This post was exported in one click from Google Docs into WordPress with Wordable.