AB Analytics : Web Analytics and Optimisation

Far from average ecommerce conversion rate analysis - part two

Creating A Measurable Advantage

Like this article?  a) tell us  b) share it c) download it and d) keep updated

online shoppingCreating a measurable competitive advantage using data in Ecommerce  is nothing new. Companies like Amazon and eBay have built their entire business operations model around leveraging data to best affect and at the heart of optimisation strategy is precision.  

So what techniques can you start using to improve site conversion rate today?

Analytics Diplopia (Double Vision) 

Visitors are counted more than once when web analytics tools are poorly implemented. 

The good news is that this means conversion rate performance is most probably being under-reported by your business. Unique visitor (and visit) metrics used by many companies as the denominator for calculating conversion rate are most probably, being (artificially) inflated.  The standard ‘out-of-the-box’ configuration that comes with leading web analytics tools will double-count unique visitors who navigate between domains during their user journey. Why you ask?  Well imagine that a copy of the original visitor with slightly different DNA has to be created to ensure tracking continuity (using cookies).  

For example, if a visitor comes via Google organic search keyword “shoes” to an online store (http://www.myshop.com) and adds a product to their basket.  The visitor is transferred to a different domain for secure payment (https://www.mysecurepayments.com) for check out. The shopper will be counted as two separate “unique” visitors because different tracking cookies for each site (domain) were used.  Imagine what the actual conversion rate could be after these people are measured correctly!  

cross domain visitor tracking

Solution: Readers using Google Analytics should learn about how to improve visitor tracking and because every web analytics tool has a slightly a different method, it is best to ask your vendor or alternatively contact us for expert, impartial advice.    

Lore and Orders  

Traditional web analytics tools have been known to report purchases more than once. 

The bad news is orders measurement is most probably being over-reported too. Web tracking tools use a special piece of JavaScript code or “tag” that triggers when a successful purchase has been made and the “thank-you” confirmation page is loaded.  This method of data collection is prone to error because of the way customers behave onsite.  Order duplication arises when an online customer chooses to save, bookmark or refresh the confirmation page in their browser causing the page to reload and the web analytics tags to count another “successful” order.  Thankfully, some technologies automatically prevent the order confirmation page reload from being counted.  However, a number of platforms do not and if you are concerned, build a quick report and accurately determine the answer.

duplicate online orders  

Use the orders metric and group by the dimension that contains a (unique) purchase transaction id.  As a rule of thumb, where there is more than one order recorded per transaction id there is a problem.  Calculate the impact historically by pulling data for previous periods and remove these orders from future conversion rate reports. Reader using Google Analytics can use this pre-built duplicate orders report

Solution: The best long-term plan will require a web developer to adjust the confirmation page HTML code or template so that the ecommerce analytics tag only loads once per purchase so as not inflate the orders metric. 

Top Tip!
  While the developer is changing the ecommerce tag, ask them to rename reloaded confirmation pages (e.g. /purchase_confirmation/reload/). This will help to build more accurate purchase conversion funnel reports by using the authentic success page and bring purchase confirmation Pageview numbers into better alignment with the orders metric total.

Preaching To The Converted  

Fix cross-domain tracking in order to evaluate marketing activities properly.  

Now having read this far, you could be forgiven for thinking that if both components of the conversation rate formula are inflated, why not just leave them and the equation remains balanced?  The answer is not quite since there is probably more duplicate visitors than there are duplicate transactions.  However, the real reason to improve tracking is because of customer communication and the accurate evaluation of marketing performance.

bar chart ecommerce performance

Broken traffic source data originating from cross-domain tracking means that marketing campaigns don't get the recognition they deserve especially campaigns targeting existing customers. Channel  attribution modelling and ROI analysis are problematic to attempt also with the inflated revenue from duplicate order transactions.


Timely Promotion 

Get customers to more buy than once the easy way.

Part of Amazon’s success has been attributed to their recommendations engine that serves tailored site content to would-be customers based on site navigation behaviour and product combination that previous customers have made.  Whilst this technology infrastructure is not within reach for many sites, there is another option available to increase conversion and it is often underutilised by many online stores (including Amazon)…  

Recommend complimentary products to customers immediately after a successful purchase has been made.  Why does the immediacy of this promotional technique work you ask?   

Clickstream analysis of ecommerce sites regularly shows a small, but consistent percentage of visitors who immediately go back and purchase again after successfully completing their first order.  The Pareto principle at work perhaps?   
online shopping maze
Most likely people are just anxious about making a purchase online.

A. “Will my payment get accepted?”
B. “How big are the actual shipping charges?”
"Is the site really secure?” 

These are all familiar questions and psychological barriers that potentially deter visitor from becoming customers. 
On the other hand existing customers have laid to rest any doubts and are far more likely to buy again as a result.  How can the recommendation technique be implemented?  

Solution: Dedicate a highly visible area within the order confirmation page for ‘Recommendations’.  Insert products that are regularly purchased with the same item(s) that have just been ordered. Extend the recommendations approach and tailor order confirmation email content. You’ll be surprised at the impact of benefiting from favourable open rates. 

Top Tip!  Motivate and incentivise repeat purchase behaviour by adding time-based offers. 

Test It: Which of the order confirmation pages below would perform best?

confirmation page redesign contest

Send your answer and reasons via email: info@ab-analytics.com 

Analysing Basket Cases

Discover cross-sell and up-sell opportunities through data mining and increase conversion 

From a Merchandiser’s perspective, the hardest part about cross-sell is deciding what products go well together and displaying them on site.  The starting point for anyone not wishing to rely on gut feel alone is usually the analysis of existing customer order records as the process can yield surprising results.  And in case anyone reading this article has heard of a famous supermarket data mining example, here are the real beer and diaper facts via Daniel Power. After completing the next challenge you’ll be left with an actionable report for cross-sell analysis to customise.  

Download the Product Affinity Excel Template

product affinity analysis

How to get started with product affinity (or market basket) analysis. 

  1. Use a customer database to obtain order level data for transactions with more than one product purchased (Item Qty>1).  

    Use a manageable but representative time period for analysis and include the following fields:  Transaction Id, Product Name and Quantity [optional]

    This task can be done using web analytics tools but most reporting interfaces come with data query and export limitations that are time consuming to work around.  

    Use an Excel based reporting plugin or API where available as an alternative.  Readers using Google Analytics can try
    Excellent Analytics for Excel or the Core Reporting API.  Product Category analysis is also possible using the same technique.
  2. Export the transactions data from the platform and save the file somewhere safe on your local machine (e.g. desktop) in .CSV format.
  3. Open the file and prepare the dataset by standardising any product names or transaction id anomalies.  For example, altered text strings (“&” instead of “&”) and encoded characters that might cause a problem when matching records.

  4. Download a software package called, “R” from www.r-project.org. The site cosmetics are less important than the powerful and free software it holds but be sure to find the download files that are appropriate to your operating system from the mirror country locations.

  5. After the download has completed, run the .EXE file install ‘R’ using the menu wizard provided and open the software package.  R console version 2.15.3 (below).

    r-statistics startup console

  6. Using the ‘R’ console command prompt > copy & paste the following string then press [enter]: 


  7. Select a mirror location from which to install the ‘arules’ package and click [ok]

    arules package installation

  8. After successfully installing the “arules” package at the ‘R’ command prompt > copy & paste the following string then press [enter]: 


  9. Find the .CSV file prepared earlier and note the location: 

    C:\Users\[Insert Name]\Desktop\Transactions.csv

  10. Change the original folder location path back-slashes “\”to double-back-slash (or alternatively, use a single forward-slash “/” as preferred): 

    C:\\Users\\[Insert Name]\\Desktop\\Transactions.csv
  11. At the ‘R’ command prompt > copy & paste the following string then press [enter]: 

    txn = read.transactions(file=" C:\\Users\\[Insert Name]\\Desktop\\Transactions.csv", rm.duplicates= FALSE, format="single",sep=",",cols =c(1,2)); 

  12. At the next ‘R’ command prompt > copy & paste the following string and hit [enter]: 

    basket_rules <- apriori(txn,parameter = list(sup = 0.002, conf =0.5,target="rules"), appearance = list(default = "both"));inspect(basket_rules); 

    Adjust the sup and conf values to find the best rule criteria for your dataset

  13. Copy results to the Product Affinity Excel Template and customise. 
  • Support is the proportion of transactions which contain these items (frequency)

  • Confidence is the estimated probability of the event occurrence (likelihood)

  • Lift is predicted performance when applying rules to the total population (potential)  

After the data has been properly mined and analysed, then comes the interesting bit of finding out whether or not the cross-sell works in real life.

Happy Optimising!

What other techniques are you using to improve site conversion? 

We would love to hear your opinion in the comments section.  

About the Author

Alex Brown is a Digital Analytics and Site Optimisation expert who works as an independent freelance consultant.  The opinions shared in this blog are based on personal experiences gathered over a decade of data crunching and technology evaluation.  The author makes no attempt to be grammatically, politically (or otherwise) correct.  Spelling was never a strong point and for practical reasons, not all vendors in the market are referenced in the article - no hard feelings.

Related Products & Services

With in-depth knowledge of both the market place and major technology vendors, allow us to help select, negotiate, support and manage web analytics implementation for your business.  Contact us for a confidential and informal discussion to see how our solutions can improve your business today.

(In Alphabetical Order)
Amazon Retail Site
Daniel Power Beer and Diapers
EbayAuction Site
Google Analytics (1) Cross Domain Tracking
Google Analytics (2) Duplicate Orders Report
Google Analytics for ExcelExcellent Analytics
Google Analytics Query BuilderReporting API
Pareto Principle80/20 Rule
Statistical Software PackageR-Project
Recommendation TechnologyEngine
Wikipedia (1)Cross Sell
Wikipedia (2)
Market Basket Analysis

Like this article?  a) tell us  b) share it c) download it and d) keep updated

Credit card image courtesy of Naypong / Freedigitalphotos.net
Graph image courtesy of Winnond / Freedigitalphotos.net
Maze image courtesy of Pakorn / Freedigitalphotos.net

Comments: 1

  • meena Nov 16

    Please mention http://www.hirewpfreelancer.com/, it is the best freelancing website also its free and user friendly.

Add a comment

Email again: