How marketing analysts can make their CMO shine at the budget table

Isabelle Valette, Morten Estrup Madsen & Pontus Norberg

Om forfatteren

I will make you shine!

“The marketing budget is cut again.” How many times have you heard the Chief Marketing Officer (CMO) say that? Every time a company needs to save money, the marketing budget will be reduced. Guaranteed! Why is that? Let’s stop putting our heads in the sand and face the facts. Because many “budget cutters” are unaware of the consequences of reducing the marketing budget. And how could they know? If you ask the CMO, what will happen to sales if the marketing budget is reduced with 10%, very few CMOs can give a straight answer. According to McKinsey only 36% of CMOs have used analytics to quantify the Return On Marketing Investments (ROMI).

The perception that it is OK to cut the marketing budget will prevail until marketers start to do what it takes to demonstrate the effect of their marketing investments. Putting hard numbers behind marketing is challenging and not everything can be quantified. But quantifying 80% is better than nothing when it comes to budget discussions. Marketing Mix Modelling (MMM) is one option to quantify ROMI. MMM optimises the corporate media investments by quantifying the contribution of each marketing channel in terms of sales. In this article, we will present how MMM can be used and what it takes to build them.

On a strategic level, marketing mix models evaluates the different media groups’ profitability and helps to prioritise business investments. It secures that the most cost effective media mix is selected. On a more operational level, the models help define which campaigns can best support sales, when they should run in sequence and when they should run in parallel. MMM is also a tactical tool to measure campaign effects and evaluate how long these effects will last in the market. As such the models can be used in the daily media planning. MMM also provides insight other than ROMI. The model analyses all the forces and external parameters that could potentially affect the company’s business like price adjustments, reputation damage, competitor strategic decisions or any other factors. Analytical tools such as MMM provide decision makers with facts; hard core facts on the strategic, tactical and operational level, facts that quantify marketing decisions, facts that help decision makers increase the performance of their company.

The data

Before MMM can be built, we need data. Any good model requires good data. Anyone a little familiar with analysis knows that the outcome of the model is only as good as the input data. Data quality is therefore mandatory. At best, wrong data will create “noise”. Noise refers to all the inputs that are irrelevant for our models. Noise increases the number of hours used in modelling. Too much noise will give wrong results and wrong recommendations.

The first quality issue to deal with is to secure that all the needed inputs are included in the model. This can be done by brainstorming sessions with marketing and business experts. This will insure that significant inputs are not left outside the modelling exercise. Once input data has been selected, a data history of 3 years on weekly level will be required. Why do you need History? Well, history is necessary to separate seasonality variations from other kinds of effects, thus enabling the identification of the marketing’s consistent and significant contribution on sales over time. If you do not have access to 3 years history, don’t worry. There are workarounds and actionable insights can be also produced with less data.

Marketing and non-linear world

The modelling world is divided mainly into 2 kinds of tasks: grouping objects together (classification) and predicting a continuous outcome (regression). A typical regression task is to predict the marketing impact on the sales. Many remember (not always with a lot of affection) using linear regression at school. The main idea behind linear regression is that when the input increases, then the output (sales in this case) will increase as well with a certain factor. All this takes the form of a straight line.

The marketing world is far from behaving in a straight line however. Many factors affect an ad’s effectiveness in various ways. A new brand would have less impact than a well-known brand. A simple message will have a different impact than a more complex one. Some target groups are more easily engaged than others. Finally, let’s imagine the CMO’s dream scenario: There are no upper limits to the advertising budget. An infinite advertising budget will not have a similar infinite effect on sales. A natural level of “saturation” will be achieved after a while. The mathematical model that captures such effects takes the shape of an “S”. It is called the sigmoid function. This function is one of the most widely used functions in the modelling world because it is simple and yet very powerful (logistic regression). Combined in many layers, it is part of some of the most powerful algorithms that exist (backpropagation neural network) (Fig 1).

Another challenge for those of us that like “straight lines” is that a customer does not react in a “straight line” either. What we mean is that the customer does not immediately react to the advertisement. The customer reacts when he is considering buying your product, and not before. This could be the same week as the advertisement but also the weeks after. This delayed effect decreases over time however, until the customer is exposed again to the advertisement. In the modelling world, this effect is known as “advertising adstock” (a term introduced in 1979 by Broadbent).

The sigmoid and adstock functions (Fig 2) are the main tools for understanding the effect of marketing activities. They will be used to find the best models that show all the parameters that affect sales. These models will not be found directly however, more advanced data mining and computer science techniques will be used. One advantage of using this indirect approach is that we decrease the chances of fitting the “noise”. However, this approach is rather computationally expensive and you should have access to a lot of computer power to do this.

MMM has sometimes been criticized for not being able to measure properly the long term effect of Marketing. It is indeed true that ROMI modelling is primarily used for measuring the short term impact of marketing investment. However, if we want to look at the longterm perspective, it is possible to quantify the long-term effects of marketing by adding brand equity and/or brand awareness measurements into the model. The trick to do this is to have a two steps approach. First, a brand awareness model is created. This first model measures the effect of media on branding. The long-term effect is then calculated based on how much media changes branding over time. The second model measures how branding changes the sales. With this two steps approach, it is possible to quantify both the short term and some of the long term ROMI.

Competencies that you need when involved in a MMM-project:

  • A solid background in modelling, statistics or mathematics is a good starting point. So you are looking for someone with an analyst profile.
  • A fair knowledge about computer science, programming and machine learning is a definite advantage to understand how algorithms are built and what kind of computer power you need to run it.
  • Data visualisation, storytelling and communication skills are important as well because the findings from models are difficult to translate into value adding business insights.
  • Finally, you should be a domain expert for the area you are going to model on. And if you are not an expert, make sure that the expert is not far away and available to answer your questions.

Andre artikler du kan være interessert i

Se alle artikler

Segmenteringsutfordringer

- Det er få ting i markedsanalysefaget som er så lett i teorien og så vanskelig i praksis som segmentering, skriver…

fagartikkel, tema-artikkel

De unges partistemmegivning i valgomatens era

Valgomater er den enkeltfaktoren som er nest mest avgjørende for hvilket parti de yngste velgerne stemte på ved stortingsvalget i 2021.

tema-artikkel

Kjønn har ikke to streker under svaret

Hvordan spør vi om kjønn? Spørsmålet dukker opp stadig oftere i møter med oppdragsgivere, respondenter og kolleger. Tilbakemeldingene, og refleksjonene, varierer…

tema-artikkel