Under the Hood Part 2: Slope, Standard error and Delayed effects

Christopher Engman, CMO/CRO at Proof, explains how slope, standard error and delayed effects are important terms you need to include in your marketing strategy.

Now we’ll go through the main components of multiple regression modeling or marketing mix modeling: slope/multiplier, delayed effect and standard error. In a simplified way, let’s look at one specific variable. How much we spend on for example retargeting. And then how that is influencing sales. Now bear in mind you cannot analyze these variables one by one because there’s too much biases introduced into the modeling. In reality you need to look at multiple variables at the same time. How they in a combined way are influencing sales, awareness or your Net Promoter Score etc. So we are looking at a simplified version.

Slope or multiplier

The Slope or multiplier tells you how much an activity is affecting sales. If you increase your spend for example on retargeting by one unit (at the x-axis) how many units does it increase the sales (on the y-axis)? Let’s say it’s 15, so if I go one level up in spend, then I go 15 levels in revenue. Then the multiplier or the slope of that variable is 15.

Standard deviation

Let’s say your data points are very spread out from the average curve or the slope. Then the distance between the curve and the dot is the standard deviation. With a higher spread or bigger volatility your standard error is higher, so it’s harder to predict each time. Over longer time periods it’s okay to have a pretty high volatility because you do enough attempts and over time you will end up pretty near the slope. So delayed effect is a very interesting term and I would say that most companies that are not using marketing mix modeling are completely missing delayed effects.

Delayed effects

Without the delayed effects the return on marketing looks pretty slim. It doesn’t look lucrative or profitable to do marketing in many cases. But when you quantify the effect over time, it comes out as very profitable in most cases. So in marketing mix modeling we look at delayed effects in two ways. First, let’s say that we put in $1 extra on retargeting. The multiplier, the slope is 15, so $1 in gives us $15 back. We can then look at the half-life, telling you when the half of the $15 will come back as additional revenue. In this case seven and a half dollars. This could be five weeks, it could be three months, it could be one year, could be a few days, it depends. The half-life is a very important indicator of when the effect will hit you. We also quantify the shower effect, because most marketing activities hava an effect that is spread out over time and often becomes weaker and weaker over time. We call it the distributed effect.