Seasonal & Geographical Factors Affecting Online Retail Behavior
Entice your customers by showing merchandize relevant to them
Author: Adarsh Jaiswal, Senior Associate Data Analytics
If you are an online retailer focused majorly on clothing and related accessories, promoting seasonal product-types for each season may not be enough to elicit customer engagement on your site. The choice of products depends significantly on multiple other factors like geographic terrain, demography, local weather, etc. Consider winter clothing trend variations found for a US retail giant’s customer base:
· Demand for boots is expected to increase across US during autumn/winter. However, it was found that shoppers from Southern US regions were 25% more likely to order if the product-type involved is ‘Boots’. One of the possible reasons could be a high number of motorcycle registrations from Southern states[1], which may be directly correlated to the interest of southern region residents in riding boots.
· US Mountain states were found to prefer a balanced mix of formal and casual clothing like dress shirts and a pair of jeans. This is in line with what blogs have to say about Mountain states’ winter clothing trends[2].
· US Midwest states were found to buy hoodies, cardigans, sweatshirts, crew socks more than any other US state. This purchase behavior can be attributed to the presence of Big 10 Schools[3], which contributes to young, college-going population density in the region.
Similarly, you can easily get an indicative list of product-types preferred by your shoppers based on their geographic location to make your product-type recommendations more relevant to shoppers. Read on to know how!
1. Identify top perennial and seasonal product-types by order
In a full four-season cycle (Autumn, Winter, Spring, Summer), identify top N[4] product-types by order for each season. Split these product-types in two categories:
i. Perennial product-types: The ones which appear in the top N list for three or more seasons
ii. Seasonal product-types: The ones which appear in top N list for two or lesser seasons
For the above example, the information was fetched using Adobe Analytics Workspace Dashboard where product-type report was pulled with Orders metric, for each season time window. Adobe Analytics should have product-type (or equivalent) variable configured and implemented in order to carry out this analysis.
2. Classify US States into Regions[5]
Group all states into a few major regions as indicated in the map below:
3. Identify preferred product-types (perennial and seasonal) by region
For one full four-season cycle, find out:
i. Overall order share of each region (irrespective of product-type): The table below illustrates the overall order share calculated for each region[6].
ii. Order share of each region for each of the identified product-types: The table below shows each region’s order share (percent) for one[7] identified product-type.
US South-East Region contributes for 22% of total orders received by the retailer. However, the region’s order share increases to 26% when we consider orders only for T-Shirts.
iii. For each (Region, product-type) pair, calculate percentage difference of Order Share for current product-type with Overall Order Share for that region. In the above table, percentage difference for US South-East was calculated as shown below:
Likewise, delta values for all (Region, product-type) pairs were calculated and tabulated as shown below:
All pairs with delta values greater than alpha were identified; where alpha[8] is a threshold value used to adjust the number of identified product-types for each region. The greater the value of alpha, the lesser the number of product-types which are identified as preferred choice(s) of the region in question. The above table indicates that T-Shirts is a preferred product-type for US South East region.
By the end of the third step, you will have a list of preferred product-types for each region.
Making Impact
Once product-type preferences are known for each region, you can leverage this information to dynamically change product-types shown in seasonal campaign banners on homepage. This can be achieved using any experience personalization tool (ex: Adobe Target), where you can create perennial and seasonal collection of preferred product-types for each region and can set up a personalized homepage experience for each season and geographic location.
Before making the changes live for all visitors on site, it is advised to test the effectiveness of the new experience by carrying out A/B Test on site. In this test, the new experience is served to a small group of visitors on the website and conversion performance of this group is compared with the remaining group, which is shown the old experience.
Measuring Impact
In order to be sure that the new experience has led to a significant change in conversion performance, it is important to decide on how long should the test be run. This depends on multiple factors like:
· Avg. number of daily visitors on site
· Total number of experiences (including control) in the test
· Baseline KPI (Conversion Rate/Revenue per visitor)
· Confidence Level and Statistical Power[9]
Test run length estimation can be done with the help of online sample size calculators[10].
Impact measurement begins with choosing the right Key Performance Indicators (KPIs) which will facilitate in comparing the conversion performance of the new experience (variant) with the old experience (control). Here are a few KPIs that can be considered for impact measurement in this test:
Ø Conversion Rate
Ø Revenue per visitor
Ø Average Order Value
Ø Homepage Banner Click-Through Rate
Ø Product Views per user
After test completion, percentage changes in KPI values of variant from baseline values are calculated. This is used to decide whether the new experience has benefitted conversion on site.
As they say, the first impression often lasts forever…
Running such personalization activities may prove beneficial in eliciting engagement, especially from non-logged in visitors, for whom product-type preference is not known in the beginning. A good initial engagement with a new visitor may result in recurrent long-term benefits to your business!
[1] See Top US states by no. of motorcycle registrations
[2] Check out what fashion blogs have to say about US Mountains: fashion trends
[3] More on The Big 10 Schools
[4] Top 25 product-types for each season were considered in this demo
[5] Adobe hit-level segments were made to filter out data based on geolocation. ‘States’ variable was used with multiple value checks appended with ‘OR’ to create US Regions Segment.
[6] Numbers shown are for demo purposes only and are not actual figures.
[7] The actual table would contain additional columns to cover all the product-types identified in the first step.
[8] Alpha was set to +10% for demonstrated case
[9] 95% Confidence Level and 80% Statistical Power are widely accepted threshold levels
[10] Example: Adobe Target Sample Size Calculator