Seasonality as a topic is often used as a substitute for holiday talk. As in we expect more sales due to seasonality in the winter holiday months. While helpful this concept can lead you to overlook the variances in performance across the year.
This will throw off any forecasting or analysis project you work on. But how do we find these trends? The answer is through basic visualization. It won’t solve all of your problems but it will point you in the right direction, getting you most of the way there.
Background
Recently I was talking to a coworker who needed estimates for multiple markets based on performance. After adjusting for performance metrics such as impression share, click through rates, conversion rates, and average returns, the coworker was stuck.
There was a big question mark on balancing each month and market. Should everything be treated the same in terms of months or would that hinder performance?
The coworker then stumbled across the answer when she realized goals changed every summer, meaning each month was inherently different beyond basic performance.
The same goes for modeling account activity. Recently Jacob Brown and Emma Franks have covered How To Use Historical AdWords Data To Create Your Own Bid Simulator and Using Excel Regressions To Better Understand KPIs. Both of which benefit by defining and adjusting for any trends.
Line Graphs
Line graphs are perfect for viewing the variance of a given metric by time. You’ll use your X axis for time and the Y axis for the corresponding value.
What to look for
Look for steep increases and declines and you’ll be on the right track. The key here is to select your periods. Some of the variability is noise. Other times it’s too specific to make use of.
How to improve
For example, the graph below is very noisy but only because each day of the week has very different performance. This isn’t wrong but makes it difficult to see larger trends.
One way to get around this is to either export your report in weekly or monthly segments. Another option is to build a rolling average.
A rolling average averages the last x number of values. This smooths the line and makes it easier to read. The benefit is retaining some of the variability while quieting what we assume is mostly noise. Your choice depends on the data itself. It might not make sense to build a rolling average for two days but it may make sense to average two weeks.
If you want to make comparisons, and the values are roughly in line, you may also plot multiple categories on the graph. You may put devices, campaign groups, campaign types, or any other break down to see how they compare over time.
Heat Maps
Heat maps are a popular way of showing the weight of a value in comparison to others. There are many examples of heat maps for ad scheduling.
Rather than a line or marker, the heat map relies on color to indicate value. This could be darkening or lightening to color or changing the color entirely depending on the value.
How to Improve
Heat maps can often be improved using an appropriate scale for mapping. For example, rather than mapping the color range to the min and max, which could be heavily influenced by outliers, you may map the middle range of the color to the average or mode of all values. This keeps everything in the middle from being muddled and pooled into the same bucket.
Scatter Plots/Box-Whisker Plots and More
Scatter plots allow you plot against two values. Unlike the heat map these exist as points rather than colored coordinates.
Box plots cover the distribution of values by category.
How to use them
Both of these are great for making comparisons between groups, such as campaigns. For example, set one of the axes to a time and plot.
Alternatively, you can get creative with your scatter plots and color code them by time of day or year. The goal here it not only to get a sense of seasonality but groupings as well. Are certain groups more similar than others. For example, in the graph below, we see the morning hours are surprisingly clustered compared to the afternoon and evening hours.
You can even get a little crazy with box and whisker plots and do a single plot per hour of the day.
This may be a clearer indicator than a simple value in a heat map. Similarly, we can use the box and whisker plot to see variation of performance within an hour.
Conclusion
There are many ways to explore the data in your accounts. By adjusting settings as needed and using these charts you can find your way to insights on seasonality, allowing you to make better decisions and create more accurate models. Each account will be different and require a different approach but these simple visualizations will make the process much easier than staring at a spreadsheet.
