In an era of increasing number of communication channels and brand touch points, proper identification of the importance and impact of each channel is becoming increasingly important. Correctly answering the question: to what extent did the use of a given message and channel affect the achievement of a goal is crucial for optimizing activities and maximizing the return on the invested marketing budget. The problem is as important as it is difficult. However, attribution modeling and data science methods come to the rescue.
What is attribution modeling?
Attribution modeling is the process of building a model to assign value to each of the touchpoints along a customer’s conversion path. It aims to understand which marketing channels and activities contribute to achieving business goals, such as making a sale, acquiring a new customer, activating dormant customers, recruiting new loyalty program participants or increasing brand awareness. Under the term attribution model, there can be many different constructs, from very (too) simple to very complex. In general, models can be divided into: single-point, rule-based multi-point and algorithmic multi-point models.
Single-point models allocate the entire value of a conversion (or, more broadly, goal achievement) to only one point of contact. Typical approaches are first-click or last-click. These are simplistic models. They do not take into account the entire customer conversion path. They don’t take into account the interactions between different points of contact and the context in which it takes place. Their advantage is simplicity and ease of application. However, in the complex world of today’s marketing, they are too simple to reliably reflect reality.
Rule-based multi-point models
Multi-point models distribute value among different touch points along the customer path. At the same time, they are divided into rule-based and algorithmic models. The former assign value to individual contacts based on predefined rules. For example:
- linear model – assigns equal value to each contact point encountered by the consumer on his path to conversion;
- U-shaped model – assigns the greatest value to the first and last points of contact, intermediate points are of lesser (though non-zero) importance in this model;
- model based on conversion time – assigns the greater value the closer the point was to the moment of conversion. In this model, the greatest weight is assigned to the last contact immediately preceding the conversion.
The advantage of rule-based models is their clarity and relative simplicity. Also that they do not omit any touch points on the path to conversion. However, their weights are given based on arbitrary rules. Justification can be found for each of them. However, it is impossible to say which one is the best. As with single-point models, their disadvantage is also that they do not take into account interactions between different points of contact and do not take context into account.
Algorithmic multipoint models
Algorithmic models, like rule-based models, assign a weight to each touchpoint along the customer path. However, instead of arbitrary rules, they use sophisticated statistical methods to determine these weights. So instead of adopting predefined rules, these models “learn rules” from real data (using machine learning methods). Such models take into account the order of contact points and interactions between them. For example, the impact of an email on conversion may be greater when it was preceded by a banner display. They also take into account context, e.g. time of year, weather, media activity of competitors, pricing. They can operate at a very detailed level, e.g. distinguish the impact of individual creative variants or where and when they are displayed.
It is hard not to agree with the statement that today these types of models are the “gold standard”. Only they make it possible to take into account the entire complexity of consumer-brand contact paths. However, behind the accuracy and benefits of algorithmic models, there are associated challenges. In particular, as to the quantity, quality and scope of the data and the analytical competence required to create them. They also have the disadvantage of limited transparency due to the complexity of the rules that govern reality and are identified by the model. Algorithmic models, however, allow for advanced simulation (what if?) of various scenarios, e.g. what if we dropped channel A altogether? what if we reduced the budget for channel B? what if we switched the order of messages in the sequence? This in turn allows you to optimize your budget and activities. The investment in this type of model can therefore more than pay for itself.
Marketing attribution models have undergone a long evolution from simple single-point models to multi-point models based on complex machine learning algorithms and statistical methods (including those based on deep artificial neural networks). In doing so, it is still an area of intensive research and experimentation both in the scientific community and among practitioners. Despite the complexities and challenges of their creation and application, they are increasingly accessible thanks to the falling costs of data collection and processing. Thus, we are entering an era where we should not ask “whether” they are worth using, but “how” to build and use them effectively.