Martech: Martech is Marketing Logo
  • Topics
    Digital Transformation
    Marketing Operations
    Data
    Customer & Digital Experience
    Performance Marketing
    Marketing Management
    Special Reports
    MarTech Topics
  • Conference
  • Webinars
  • Intelligence Reports
  • White Papers
  • What is MarTech
    Mission
    Staff
    Newsletter
    Search Engine Land
    Third Door Media

Processing...Please wait.

MarTech » Data » ‘Tis The Seasonality Of Your Metrics

‘Tis The Seasonality Of Your Metrics

A few posts back, I examined a simple technique for using an exponential moving average (EMA) on your time-series metrics. This has the advantage of smoothing out the metrics while at the same time keeping a “memory” of all previous values of the metric that came before. It also has the side benefit of being […]

John Quarto-vonTivadar on July 10, 2013 at 8:45 am

A few posts back, I examined a simple technique for using an exponential moving average (EMA) on your time-series metrics. This has the advantage of smoothing out the metrics while at the same time keeping a “memory” of all previous values of the metric that came before. It also has the side benefit of being easier to update as new values to your metrics become available.

This time, I want to show you a technique to fix another common time-series problem: seasonality. Yeah, your metrics are down in January, but is that the usual post-holiday sales slump? Or is it the start of a true downtrend that you need to keep an eye on? The article will illustrate a fast and simple way to de-seasonalize your data.

Let’s work through an example step by step:

Screen-Shot-2013-07-08-at-1.50.15-AM

Imagine we’re working at a software company where the enterprise-level product has a fairly long sales cycle, and our metric is actual raw conversions during a quarter. If you were looking at your conversion sales metrics in the above chart, how are you doing lately? Better than 2010? As good as 2011? Here’s the raw data:

Screen-Shot-2013-07-08-at-1.33.32-AM-300x115

Step 1: Collect Metrics Data Going Back At Least 3 Full-Cycle Periods

For most folks, this means quarterly or monthly data going back three years. Yes, you can use weekly or daily data, although you’ll usually want to smooth this data out (cha-ching! another great use for the exponential moving average!).

And, it needn’t be over the course of an entire year if the “seasons” your data go through isn’t measured by the calendar months — so if you’re looking at day-of-the-week data, say, to compare how your customers act on Monday versus Thursday, then a good 8-12 weeks of data would be a smart minimum.

Step 2: Compare Like Time-Periods To Like Time-Periods

For example, look at all the Januarys, or all the Tuesdays, and calculate an average.

Screen-Shot-2013-07-08-at-1.33.52-AM-300x82

Here, I use a simple average rather than the EMA. Because the EMA is designed to be useful for the time-series period compared sequentially — comparing a February to the January that came before it — and we’re not doing that here; rather, we’re treating the data as pure data, and our end goal is to extract the seasonal time-series.

Step 3: Normalization

Compare all these averages to each other, and divide each of the averages to the average of averages, yielding a seasonal adjusted factor for that time period, on average, compared to the normal value, referred to as “normalization.”

This is how we compare “apples to apples” across multiple years and in the context of the entire seasonal effect.

Step 4: Divide Each Original Data Point By Its Seasonally Adjusted Factor

This gives you an effective value for that metric with the seasonal component removed.

Screen-Shot-2013-07-08-at-1.34.07-AM-1024x187

Step 5: Draw Conclusions

Look at this new de-seasonalized data and extract conclusions, if any, from it.

Screen-Shot-2013-07-08-at-1.50.26-AM

Now that we’ve subtracted out the seasonality of the sales, are your conclusions any different? Looking at the red, de-seasoned data, it sure looks like the 2012 low is even lower than the end of 2010, and the 2012 isn’t nearly as high as in 2011. This should cause some consternation in the next sales meeting!

Of course, there are a zillion caveats here. Is your data even seasonal in the first place? Looking at the blue line, the best we can say is “maybe” — it’d be great to have more data. Perhaps a monthly breakdown of data instead of quarterly.

Another caveat might be cyclic — if the economic cycle dominates your sales cycle, then it can easily swamp the seasonal component — but if that’s the case, why the lower low in 2012 sales?

So, my question this month: when you look at your metrics, do you account for seasonality? Or do you just look to see if the numbers are “up” in the most recent report? You might be missing a key insight.


Opinions expressed in this article are those of the guest author and not necessarily MarTech. Staff authors are listed here.


New on MarTech

    Worsening economy has more shoppers getting online info before making in-store purchases
    Getting Started with the Agile Marketing Navigator: Building a Marketing Backlog
    3 ways to dominate with Google Auction Insights and search intelligence
    Webinar: Work smarter, not harder, to give customers what they want
    B2B customer journeys that begin at review sites are significantly shorter

About The Author

John Quarto-vonTivadar
John Quarto-vonTivadar is one of the inventors of Persuasion Architecture and regularly combats innumeracy among marketers in his popular "Math for Marketers" series. John's 2008 best-seller, "Always Be Testing", written with business partner Bryan Eisenberg, has been the standard reference for conversion optimization through testing since its release and has been used for the basis of both academic coursework as well as corporate training.

Related Topics

Data

Get the daily newsletter digital marketers rely on.

Processing...Please wait.

See terms.

ATTEND OUR EVENTS The MarTech Conference logo.

September 28-29, 2022: Fall

Start Training Now: Master Classes

Start Discovering Now: Spring



The SMX Conference logo.

Start Training Now:: SMX Advanced

November 14-15, 2022: SMX Next

March 8-9, 2022: Master Classes

Webinars

Agencies: Grow Revenue Streams Through Web Accessibility & Compliance

Protect Your Paid Advertising Spend Against Ad Fraud and Invalid Traffic

Build an Integrated Search Strategy Across Google, Amazon and YouTube

See More Webinars
Intelligence Reports

Enterprise SEO Platforms: A Marketer’s Guide

Enterprise Identity Resolution Platforms

Email Marketing Platforms: A Marketer’s Guide

See More Intelligence Reports
Featured White Paper

5 Best Practices for Bringing Together All Your Marketing Data

See More Whitepapers

Receive daily marketing news & analysis.

Processing...Please wait.

Topics

  • Transformation
  • Operations
  • Data
  • Experience
  • Performance
  • Management
  • All Topics
  • Home

Our Events

  • MarTech
  • Search Marketing Expo - SMX

About

  • What is MarTech
  • Contact
  • Privacy
  • Terms Of Use
  • Marketing Opportunities
  • Staff

Follow Us

  • Facebook
  • Twitter
  • LinkedIn
  • Newsletters
  • RSS

© 2022 Third Door Media, Inc. All rights reserved.