Despite the downturn caused by the coronavirus crisis, the digital advertising industry has maintained a strong growth for several years now and, according to Statista, in 2019 it reached a global turnover of more than 335 million dollars. With a complex advertising landscape made up of channels that are continually reinventing themselves and where new players are added every day, advanced attribution models become critical to accurately assess the return on advertising investment and website effectiveness.
The limitations of traditional attribution models
Attribution models are a set of rules companies use to assess the weight of each interaction with the brand in the final conversion (advertising impacts, but also each touchpoint on the page and even in the physical environment if we talk about omnichannel models). Traditionally, the most used attribution models are the following:
- Last-click: it gives the entire conversion to a single impact, the last click in the conversion flow. This model reveals a very limited reality: customer journeys nowadays are much more complex.
- Multi-channel: it incorporates more channels and phases, distributing a fixed percentage of the conversion between the different impressions. Their limitation is that they are closed models: the percentages are established and do not evolve, even if the user and their behavior changes.
- Data-driven attribution: these models adjust the percentages of attribution to each of the impacts in a variable way. Their main limitation is related to the fact that they analyze the route taken by the user outside the website (especially the different advertising campaigns) and, therefore, do not take into account the user’s behavior on the page itself.
Unlike these formulas, advanced attribution models not only evaluate the channels that have led the user to the web but also analyze the behavior that the user has shown on the site and cross-reference all the information to reflect the entire purchase cycle.
Advanced attribution models: the advantages of applying machine learning
The supervised learning algorithms work with already classified input and output data, meaning, the system is told the desired result and it derives rules from the data it receives.
Therefore, attribution models with machine learning are not pre-designed but are continuously adjusted and become more accurate over time. For example, every time the design of the website is changed, it is not necessary to rethink the attribution model, rather, it will incorporate this new feature into the algorithm readjusting itself.
In the initial information classification, it is essential to measure with events any opportunity that the user has to interact with the page and, moreover, to incorporate additional data to give context to the activity. This data has to answer questions such as: How much time has passed between events? How many times has each one been triggered? What devices has the user employed? From what geographical location does it access? and so on. The more information, the better the model will process it to determine what is quality traffic and what is not.
Álex Masip, Head of Data at Labelium Spain
Moreover, by relying on user behavior, attribution models can detect fraud with machine learning, given that if a bot is bringing in huge amounts of traffic but then their activity on the web is absent, the attribution model with machine learning will detect it and will not give it any weight in the conversion.
The way an attribution model works with machine learning
It can be summarized in three key steps:
1. Data collection: it can be exported through Google Analytics 360 or with a JavaScript code that saves the information. Other data sources can also be added such as Adservers, CRM or models that interpret the UTM.
2. Formulation of a data lake in which the information is grouped: it can be articulated with the help of platforms such as BigQuery, Amazon, Azure or Snowflake.
3. Implementation of the attribution model with machine learning on the data lake so that it assigns a value to each session and external impact received by the user. In this way, the probability of conversion is measured accurately. The algorithm conducts the calculation at an individual level for each visit and then groups it to obtain global percentages with the purpose of identifying the weight that each interaction has had on final sales.
Real-time and constantly evolving allocation
Attribution models with machine learning not only allow you to analyze the effects of completed advertising campaigns but also provide a global view of what is happening in real time. This makes it easier to improve the strategy on the go by making it possible to detect inefficiencies and correct them. Also, advanced attribution models can make future forecasts to raise possible scenarios and opt for the most profitable options.
If you want us to help you implement an advanced attribution model on your website or e-commerce, do not hesitate to contact us. We will study your case in a personalized way to find the most effective approach to implement it.
