Long-term effects of marketing and brand models

Introduction

Since the advent of marketing advertisers have wondered about the impact it has on their brand. Unsurprisingly, this has led to a lot of terminology being invented. Everybody is talking about brand equity, brand value, brand strength, etc. There are whole departments that work with nothing but brand-related topics. Still, when it comes to measuring the effect of brand on tangible business outcomes we are usually left with an incomplete picture. In this blog post, we’re going to dive deeper into how we can capture not only the true and longer-term brand effect on business outcomes but also the connection between different brand metrics.

Measurement of brand-related effects today

A marketing mix model (MMM) is currently the go-to tool for quantifying the immediate effect brand has on sales. While MMMs are good at explaining and quantifying these short-term effects, the question about the long-term effects of marketing is left in a black box, commonly represented as “the base”.

The most common and simple approach MMMs use to enable this quantification is including different brand metrics as explanatory variables in a sales model. Some approaches take this one step further and model the indirect effect as a proxy for long-term effects. This could be for example by integrating brand metrics as a hierarchical step in a sales model where marketing affects the brand and brand affects the sales. This get’s the causality right but still fails to capture the long-term effects.

As such, MMMs today don’t deal with the question of actual long-term effects. Also, while brand tracking can monitor the development of brand health, the unsolved question is still: How does a brand point convert to sales? How much is an increase of 1 percentage point worth in terms of sales? This may seem like a mouthful to keep track of but it all comes down to avoiding the 5 pitfalls of brand measurements and optimization.

Five common limitations in traditional solutions to be mindful of

1. Fixed base: MMMs typically implement sales models with a single static variable representing the base effect. The assumption here is often that the long-term marketing effects are baseline sales, but how can this be interpreted if the baseline is static? The answer is that it cannot, and the interpretation of base can only be “we don’t know”. Anything else is speculation.

2. Incorrect attribution of marketing effects: By including brand as an explanatory variable in a sales model, the hierarchical structure of effects is often not considered. Consequently, marketing effects that affect the brand are double counted, i.e., effect on brand and effect on sales. This creates, in linear models, multicollinearity which is just a fancy term for saying that the model won’t be able to identify and attribute the correct effects to the correct cause. Marketing effects have to be modeled hierarchically. Period.

3. Brand is not modeled as a hierarchy: Brand is not one thing, as it works more like a funnel. It starts with awareness and moves all the way down to purchase intent. Not all brand metrics have the same impact on sales, but they do have a hierarchical connection. For example, there is no brand consideration without brand awareness. Identifying the correct hierarchy between your brand measurements is key to understanding how you can affect them and make them grow.

4. Long-term vs indirect effect: Long-term effects are not really long-term in MMMs (more like a few months). Indirect effects are used synonymously for long-term effects but a real quantification is not done. This is a fallacy that often underestimates the effect of your investments and consequently drives you to make suboptimal decisions.

5. Brand measurements are noisy: Brand is typically measured using questionnaires and often there is a group of people that are asked over and over again, and not everyone answers every time. Regardless, brand metrics are often included as they were measured and therefore have a lot of measurement errors present. These measurement errors typically arise due to small or non-representative samples which are typically not handled in the model. But there is no reason for this as we today know how to model this correctly and therefore should be taken care of. The noise in the measurements then propagates to your outputs which allows you to identify and mitigate the risks associated with your brand investments.

A better way of doing it

While there are indeed many pitfalls, there is also a solution to all of them. That solution is of course to do our due diligence regarding our modeling dynamics such that we can extract the information we need. This way we can get our model to work for us instead of against us. So what are we talking about?

First off, we need to model the short-term effects and long-term effects seperately since they operate at completely different timescales. Mixing them together into one single module is like trying to measure the speed of a sports car and a photon using the same device.

Second, modeling the hierarchy correctly such that indirect and direct effects are identifiable is key as it provides us with the inductive bias we need to extract the information from the data that we require. An example of the high-level architecture of such a model is given in the illustration below.

Using this setup we are equipped to continuously measure and optimize our brand and the effect it has on our business growth. Not only the immediate effect but also the long-term one. This allows us to plan for and execute strategies that enables us to sustainably grow over the long term.