What your website says about your brick and mortar sales

By Erik Koto | 0 Comments | Posted in in Analytics | Permalink

 

What can your website activity tell you about your brick and mortar sales?  Ever wondered when your website metrics go up, if it means you’ll actually sell more products offline?  

While ecommerce activity continues to grow in all categories, for many companies the majority of online shopping activity still results in a purchase being made through an offline (brick and mortar) retailer.  As part of our Monetization Modeling practice we analyze this relationship so that we can better understand the role of online actions in the purchase process, and how to optimize the site experience to drive conversion. 

Following is an example of this research, what we learned, and the actions taken as a result.  The products in the example below are high consideration consumer product traditionally sold through a network of retailers (for confidentiality, we’ve anonymized the data).

Methodology and data:

Our research methodologies were correlation analysis and linear regression.  In order for these techniques to be meaningful we needed at least 30 months of both website activity and offline retail sales. For both datasets, we had ‘rolled up’ data (e.g. total site visits, total retail sales) and also product level data (e.g. product ‘A’ web visits, product ‘A’ retail sales). 

HIGH LEVEL SALES ANALYSIS:

With the datasets assembled our first step was analyzing high level relationships.  We ran a series of correlations between online KPIs and offline sales. We started by analyzing site visits, as this represents one of the highest level indicators of consumer interest.  Visits to the website show a fair, but not excellent correlation with offline sales:

Correlation between Website Visits and Offline Sales:  .70

The story becomes more interesting when we begin to look at visits to specific research tools on the site.  In this next graph we are looking at total visits to a group of 5 ‘upper funnel’ product information pages on the site.  These pages contain basic information about the products that a first or second time visitor might rely on to better understand the features of the product.  In this case we see the correlation with offline sales has improved over basic site visits.

Correlation between Upper funnel actions and Offline Sales:  .80

Lastly, we looked at the correlation between a group of 5 ‘lower funnel’ pages and offline sales.  The ‘lower funnel’ pages are actions a website visitor would typically engage in once they are close to the purchase decision, this actions include; finding a retailer, financing options, online inventory search, and requests to be contacted by a sales consultant. The correlation between these actions and offline sales proved to be very strong.

Correlation between Lower funnel actions and Offline Sales:  .90

 

 

High level conclusions and findings:

This analysis immediately yielded insights into shopping behavior.  Clearly, ‘lower funnel’ actions on the web site are a very important part of the purchase process. Their correlation of .90 with offline sales is proof of a very strong relationship between these actions and the purchase decision. 

Another important insight gained from this analysis is the lead time between online research and offline sales.  Prior to this analysis, the client believed users engaged in detailed online research 45 – 60 days prior to product purchase.  In our analysis all three web metrics (visit, upper funnel visits, and lower funnel visits) all show their strongest correlation with same month offline sales. This clearly demonstrated that online research takes place within 30 days of purchase, much closer than previously believed.

PRODUCT SPECIFIC ANALYSIS:

Our next step was to drill down into the data and analyze more specific relationships.  We started by looking the relationship between website actions and offline sales for specific products.  The company had 4 distinct product families.  Within each product family there were 2 to 4 unique products (or SKUs).  We knew from the analysis above that ‘lower funnel’ visits had the best correlations with offline sales, we began by rerunning the same analysis at a product level.  Once again we observed very distinct relationships.

The following table shows the results of product level analysis.  In this case we ran correlation of product specific sales against product specific web activity.  Here we see a distinct difference in correlations between Product Category A, and Product Category B.  Category A shows very strong correlations with offline sales, indicating that online research is particularly important and highly related to the purchase event for these products.  Category B shows the opposite effect, these products show very little relationship between web visits and offline sales.  Indicating these customers are not highly engaged in the web for researching their purchase.

Product Category	Product	Correlation with offline sales
Category A	Product 1	0.79
	Product 2	0.91
	Product 4	0.84
Category B	Product 1	0.40
	Product 2	-0.10
	Product 3	0.52

More specifically, this information can be used to target visitors who have completed specific research activities in Product Category A with timely offers to incent offline purchase.  Since we know these visitors are in market and likely to purchase in the next 30 days, this is a key opportunity to convert sales.

To further illustrate the strength of the correlation between online research and offline sales, we analyzed a single product through a promotional period.  In the chart below, you can clearly see the spike in both online research activity and offline sales during this period.  Knowing the strength of the relationship and the temporal proximity of online research to offline purchase is instrumental in developing timely targeting strategies.

Product specific correlation between Lower funnel actions and Offline Sales:  .91

REGRESSION ANALYSIS:

Our final analysis was to determine which specific online actions were the best predictors of offline sales.  It is important to note, our objective was not to predict product sales on an ongoing basis, but rather understand what actions were the best predictors of actual sales.  Understanding the predictors of offline sales gives us a key insight into the importance and influence of specific online actions.

We built a series of multiple regression models to determine what group of online actions could best forecast actual sales. This was a highly iterative process where we analyzed multiple groups of different KPIs and assessed their R-squared values.  The R-squared metric tells us how well a group of online actions explains the variability of offline sales.  The higher the R-squared, the better that group of online actions are at predicting sales. 

After a series of regression models, we narrowed in on a group of online actions that predicted offline sales with an R-squared of .71 (in this particular application, a very good fit).  Back-testing this model showed a very strong ability to forecast sales based on web site actions.

With the regression model in place and back tested, we had identified a group of 6 online research actions that predict offline sales with a high degree of accuracy.  The completion of these actions signal that a highly qualified visitor is on the site, and should be targeted with specific messaging.  Additionally, since we now had quantitative insight into the most important site actions, modifications to site architecture, on site advertising, and even offsite display advertising, were made to promote these key activities.

ONGOING ANALYSIS

The findings shown above provided the first quantitative insight into the relationship between online actions and offline sales.  These findings provided the rational for changes to content strategy (both onsite and bought media), promotional offerings, and targeting strategies.  The findings also helped to change commonly held notions about how and when consumers interacted with the digital channel. To ensure this was not a “one and done” analysis, we built the capability for easily updated, ongoing correlation analysis into our standard monthly performance reporting.  As the site evolves we can now monitor the correlations between online actions and offline sales, providing us with a new and powerful way to continually optimize the digital channel.