From Pumpkin Spice Metrics to Year-Round Loyalty
Over the years the fall and winter have underlined the need for understanding seasonal sales. Customers are used to making seasonal purchases, with specialty items available for a limited time. Hence my name for metrics associated with seasonal sales – "pumpkin spice metrics."
Pumpkin spice metrics (PSM) are meant to represent the activity which occurs during a seasonal sales period. The metrics themselves are not too different tfrom the ones you've chosen as metrics related to your general business objectives.
The good news for understanding PSM is that the analytics are available. The bad news? Managers struggle to incorporate seasonal sales and ordering volatility into predictive models. But deriving insights from volatile data is possible. The key is highlighting the seasonal influences in a standard analytics dashboard, and then applying a time series statistical model to detect how sustainable the influences are.
Finding seasonal patterns
Starting with Google Analytics, marketers can annotate seasonal periods and lead-in times, then look for differences in traffic sources. Search patterns can change between a season and the time leading up to it. Annotation can be used to log when an event begins, making a starting point for an analysis. One good starting point, for example, is search patterns occurring a few months prior the season. The examination can raise the question "Are there any keywords or phrases that are appearing regularly before the season?” Those words may be opportunities to adjust paid search ad campaigns.
Site search reports can also be helpful. The results, examined within the seasonal period, should reveal which topics customers are typically looking for.
These steps in Google Analytics are certainly helpful. But you may need to determine if a trend is sustainable. To take that step, you should examine the data in R programming (or Python – both languages are popular for data mining and modeling). Using R will help provide a cleaner statistical view of cyclic data.
A deeper dive
In this case we can use time series, and apply a function called Decompose. Decompose provides a way to break time series data into three components:
- Trend - A smoothed version of the data, showing an overall trendline upward or downward
- Seasonal - A visual of spikes that represent the seasonality over a given time series. The graph displays the amount of spike that is occurring and how frequent is the cycle
- Random - the “noise” in the data that can be ignored overall.
These separate views of the data are graphed. Within the context of seasonality, a decomposition time series analysis can allow for a comparison of trends, and show if there is a general uptick or downward trend.
To run a "decompose," import data from a csv file downloaded from your analytics, or from a package that will import the data directly (such as RGA, a library for importing Google Analytics). As an aside, R programming has over 10,000 libraries published at the universal library repository CRAN.
Once loaded, the data must be placed into a times series object. To do this you run the ts function. The ts function should appear in the code like the example below, using the database AirPassengers (that comes with R) and frequency of 12 to indicate monthly periods:
Next, the decompose function is applied to the time series object. An object can be created using the decompose function, so it should look like this in the code:
One question that decomposition can answer is if the data is trending with the same size of spikes – this is called additive - or if the spikes themselves are increasing in value alongside the trend – which is called multiplicative.
You can indicate whether an additive or multiplicative approach is used with the data by adding a type parameter. Here's how the decompose looks like with that parameter.
You can then plot the object, which then provides the plot and the three supporting plots of seasonal, trends, and random. Here's how it should look:
This information can help your organization determine how seasonal spikes can potentially repeat. I say potentially, because sudden shifts in consumer tastes and marketplace conditions can certainly factor into the continuation of a trend. But with all things being equal, the analysis can certainly help confirm if the parameters associated with the data – product sales, website activity – merit further investment.
Overall, you just can't be prepared for a seasonal event without the right strategy. Identifying the PSM in your business is the right strategy of understanding sales seasonality, and drawing lessons from it which may be effective year round.