Reading time:  
A marketing attribution solution is essential to measure and analyze the contribution of each marketing lever to events on the advertiser's website.

As seen in our article “Marketing tools: attribution / contribution”, there are different marketing attribution models. What are the specifics of each model? Which tools and models are most appropriate for each advertiser's objectives?
Here is a list of the major attribution and contribution models:

  • First touch
  • Last touch
  • Linear weighted
  • Increasing time frame
  • Decreasing time frame
  • U-shaped
  • Algorithmic



A - The first touch model

This model assigns an event only to the lever that occurs first in the configured attribution window . If the marketing attribution tool only considers clicks, it is called a “first click” model.

When to use this model?
The “first touch” model makes it possible to measure the performance of the initiating levers (top of the funnel). It can be adapted to campaigns to win new customers, especially for brands / products with limited awareness.

Let's take the example of an advertiser in the tourism sector

The advertiser has set up a first lever attribution model in the Eulerian tool. The configuration is as follows: a 30-day post-click window and 48-hour post-impression window. It allows us to assess the performance of those initiator levers such as programmatic display. This lever is used here exclusively for the acquisition of new customers. The rate of new customers is monitored in the tool to ensure that the lever is meeting its objective. Conversions are therefore 100% attributed to the programmatic display lever when it is the first to intervene in the window (via click or display).

B - The last touch model

The “last touch” model or last lever is an attribution model that only values ​​the last touch before conversion. If the attribution tool only considers the click, we will speak of “last click” attribution. (the only model available in the free version of Google Analytics for example)

When to use this model?
This last touch model is beneficial to goal scorers levers. It means, those that will finalize the conversion by targeting captive audiences or those already engaged with the brand such as SEA, retargeting, affiliation, ..

Let's take the example of an advertiser in the banking sector

The advertiser deduplicates the performance of its acquisition levers via a last touch model. This is done within its Google Campaign Manager adserver. It allows us to test several DSPs on the programmatic display lever and to put their performance into perspective on the basis of this model to make trade-offs. It turns out that a mix between Xandr and Amazon DSPs is the most appropriate to promote conversion according to a last touch model for this advertiser.In particular, this enabled the CPL for opening an online bank account to be divided by almost 4 compared to other DSP combinations.
In short, these two models are the most used because they are easy to set up. However, they are increasingly challenged by fractional models because they don't really represent the facts . These models make it possible to have the contributory vision and to understand the potential of each lever without favouring any of them.


They allow the contribution of the levers to be assessed by considering all the keys within the attribution window.
The implementation of these models requires being equipped with complex attribution / contribution tools with more or less significant costs. They are, however, optimal for understanding the performance of an advanced digital mix in a granular manner.

A - The linear weighted model

Relatively basic, it consists in assigning the same weight of the conversion to all the players involved in the conversion window. Thus, if 10 keys occur in a window of 14 days before the conversion, each key will be allocated 10% of the conversion and of the corresponding turnover.

When to use this model?
This model is not widely used. It can be useful when the advertiser is in the exploratory phase of his mix and wishes to understand the potential of each lever without favoring any.

B - The increasing temporal model

It works on the same principle as the linear weighted model, except that it will overweight the keys closest to the conversion time.

When to use this model?
This model is designed to give priority to kicker levers. Therefore, it is useful to evaluate the performance of a hyper ROIste acquisition mix once brand awareness is well established and site traffic is consistent.

C - The decreasing time model

The decreasing time model will overweight the keys occurring at the beginning of the window. This model is intended to give priority to initiating levers.

When to use this model?
It will be relevant to evaluate the performance of a mix aimed at boosting the acquisition of new visitors and new customers for brands whose notoriety is still to be built.

D - The U-shaped model

This one will overweight the first and last keys considering that the initiators and strikers levers are the ones that have the most weight in the path of conversion. While the smuggling levers, although taken into account, are devalued.

When to use this model?
This model is well suited to advertisers with a mature digital acquisition strategy based on both market share conquest and direct conversion to an interested audience. Thus, it is the prerogative of pure players well established and well equipped.

E - The algorithmic model

Finally, the algorithmic model is a model that will assign a weight to each key in the conversion window based on a predefined rule specific to each advertiser. Generally, the algorithmic model is based on the observation of the different interactions between the levers as well as the autonomy rate of each (i.e., its ability to generate conversions alone) to assign a value to each lever according to the position it occupies in the conversion path. It is the model that contains the least bias (based on statistics) in contrast to other models, called “a priori”, which are based on feelings and human knowledge to choose the appropriate model.

Marketing attribution: algorithmic model

When to use this model?
In this sense, the implementation of an algorithmic model requires a good knowledge and a long experience of the different levers that composed the acquisition mix. It also requires that this mix be relatively stable over time. Thus, not all the contribution tools allow the implementation of an algorithmic model. It is essential to have full transparency on the methodology for creating the model in order to understand how each lever is valued - or not - through it.


To summarise on the different attribution / contribution models, attribution models are based on a single event. Very easily deployable, however, they do not allow effective analysis of actions carried out simultaneously through fundamentally different levers.
The contribution models, based on multiple events, make it possible to assess the role of each lever. And thus, rationalizing budget investment more effectively between them. However, their implementation requires a great deal of rigour. It is also essential to be vigilant about certain practices that promote the multiplication of keys within the window by the same lever to increase its contribution percentage.

We must therefore remember that there is no perfect model. The choice of model is closely linked to the marketing mix activated by each advertiser. But also, with the typology of the levers that compose it, with the sector of activity and with the types of marketed products that will induce more or less long attribution windows. And finally, brand awareness, site traffic and the objectives set.
Also, models valuing the first touches will be more favorable to the initiating levers such as display and social media in conquest. While, the models valuing the last touches will be more favorable to the scoring levers: brand or long tail search, retargeting or affiliation.

Based on these analyses, advertisers can adapt the budget allocation of their media mix. And therefore, generate the most positive ROI possible. These analyses can be done on a recurring basis (per month, per quarter, per semester).
For an actor like Gamned! having access to its analyzes makes it possible to adapt its programmatic activations in order to correspond to the needs of the customer and to ensure an always optimal distribution in its buying and targeting strategies.


The Return Of Contextual Targeting

Find out more


Your little guide to creating performant drive-to-store campaigns

Find out more



Tell us who you are, we will tell you which digital activations to adopt!