Campaign measurement: from attribution models to contribution

Like the rest of the disciplines of marketing and, in general, of the company, advertising advances at an unstoppable pace.

New channels, media, strategies appear … and this leads to a question: “Are we making progress in measuring and leaving attribution models behind ?” Obviously yes.

This article details how has evolved measuring the impact of advertising campaigns and how it has gone from attribution models to contribution models .

Marketing mix modeling

One of the great challenges of advertising has always been to be the most effective and efficient in advertising campaigns. The big agencies, Omnicom, Publicis and WPP, put the focus of their strategy on ROI. But how can you measure the ROI of a campaign? In a simple way, the return on investment is calculated using the following formula:

The cost of the campaign is still an accounting element that can be obtained in a trivial way, but the increase in sales is more complex. To calculate it, you need to know what part of the sales are due to said campaign and what to other factors. Among them, quality, product design, promotions and sales force.

This is where the marketing mix modeling or mixed marketing model comes into play . With it, the aim is to relate, through a model, all the marketing variables, and other related ones, with an objective variable. The latter is usually the one that indicates the sales, although it can also be another that can be directly related to them. The simplest model would be the linear one:

Once it’s clear how to calculate the impact of marketing variables on sales, it’s time to talk about attribution models and contribution models.

From attribution models to contribution models

Attribution models, as defined  in Google Analytics , are nothing more than rules or sets of rules that determine “how the value of sales and conversions is assigned to the contact points of the conversion routes”, that is, they identify all the events that have achieved the conversion. There are different ways to “attribute” this achievement of conversion. The most used are the following:  

  • Last click : in it, the click closest to the conversion is the one that assumes all the weight of the conversion.
  • First click : in this case, it is the first click made by the user that assumes all the weight in the conversion.
  • Linear : all events have the same weight for conversion.

This way of measuring raises considerable doubts as to whether or not it is correct. Many believe that it is not, because what if in the case of the last click there are two very close events? It may also be the case that the user does nothing when seeing an ad on the internet (impression), but after a while do a search for the product on Google and buy it. In this case, and according to the models we have mentioned, the display impression would not have any impact on the conversion.

Do you need to ask yourself to which event to attribute the success of a campaign , or to which event to attribute a conversion? The one to look for is how much each event has contributed to the success of a campaign or conversion .

The objective is to quantify the weight that each event has on its success . You have to think of all the events that take place as a great soccer team in which everyone contributes to winning the game. The one who scores the goal is not the only one who wins. They will have received a pass, which in turn will come from a defense kick-off due to a previous clearance by the goalkeeper. Advertising is a team game . Therefore, it is necessary to know the impact of each player. This is the essence that allows pass attribution models to contribution models .

How to quantify the weight of each event?

To easily quantify the weight of each event in success, it is necessary to use the concept of advertising souvenir . This is nothing more than what is remembered or forgotten about an advertising impact over time. Thus, in an advertising campaign made up of different events that take place over time, the most distant ones will have less influence at the present time.

In 1979, Broadbent put it simply, in the so-called adstock model :

t is the adstock (‘accumulated memory’) at time t. It will take into account not only the current event, T , but the memory of past events. The further away the previous event is, the lower its memory will be , since it is less than 1. Furthermore, it is constant for a given type of event (campaign). Therefore, one can be calculated per TV station, type of digital event (impression, click, video …), discipline, etc. The calculation of is not trivial, and there are different methodologies to obtain it.

Once the value of the different ones is known , it only remains to know what place in time each event has in relation to the conversion. Finally, the adstock of each event and its weighting are calculated.

This is how you can influence the advertising events prior to an online purchase:

As can be seen in the table, SEO and SEM influence the online purchase made by more than 40% each . Television is in third position, with 15%. It is true that television advertising continues to have an important weight in the last event, the purchase, but it is no longer the main channel.

In short, the models to measure the influence of advertising campaigns used mostly so far are changing. Generally, until not long ago, exclusively those based on identifying the variables responsible for the conversion in a purchase were used. But they are already giving way to contribution models, focused on identifying and quantifying the weight that each of the events that have intervened in the process has in a conversion.