How Causal relationship can work in Marketing Science?
Introduction
As customer experiences become more complex due to the development of digital advertising, it is necessary to improve the efficiency of marketing activities by optimizing advertising budget allocation. For measuring the effects of marketing activities, it should be better to combine their short-term and long-term business goals and take a more holistic full-funnel approach so that it can continuously provide efficient marketing activities [1]
Once the integrated data regarding marketing activities has been given, it will be required to use the collected data to perform a holistic analysis of across these acitivities’ marketing ROI. There exists systematic approaches to such analysis, one of which is Marketing Mix Modeling (MMM).
MMM is based on statistical model to measure the effect of advertising spend on their KPI metric such as revenue or conversion.
According to [2], MMM helps to understand the impact of marketing tactics to then optimize the strategy and ensure that a business isn’t wasting marketing dollars.
In recent years, with the movement to tighten regulations on third-party cookies from the perspective of privacy protection, the tracking of marketing activities using aggregated data has been attracting renewed attention. In such a situation, MMM is rising their expectations as a representative of cross-channel and macro analysis approach in order to make hybrid use of personal data and aggregated data and promote increasingly complex marketing activities in a data-driven manner [3].
Since MMM often deals with the effectiveness of overall marketing activities, to assume and describe a causal structure for the effects of marketing measures handled by MMM and building a model that takes this into consideration leads to improve the interpretability of the model. In this article, it will be described the concepts of problems that can be considered and interpreted by assuming a structure of marketing activities in MMM.
Funnel effects
In marketing, the concept of “funnel” is often introduced. It describes the process from those who are not aware of the product or service to convert (purchase, contract, etc.).
Advertising channels may also have a structure that follows this funnel concept. For example, TV or OOH are likely to reach a wide range of users and contribute to increasing user awareness. On the other hand, search ads are more likely to be contacted by users who are in the funnel closer to CV. Therefore, it is possible to provide results by considering channel funnel effects in MMM, that are more in line with the actual situation.
For example, [4] expresses an image of a structure that leads to sales, using elements related to marketing other than media such as online/offline channels, promotional campaigns and time series, and actions on web pages as constituent elements.
Not only for validity of model building, it is expected to drive utilization of MMM to the decision-making in the marketing organization. By building a model that takes such structure into account, it is possible to visualize the impact of marketing activities that go beyond the direct factors on KPIs that leads to improve the interpretability of marketing activities as a whole. This will finally bring a clear and wide range of decision-making.
Endogeneity
It is considered that structuring the direct/indirect relationships between channels and KPIs mentioned above will lead to consideration of the endogeneity problem caused by marketing activities.
Generally speaking, endogeneity may occur in situations where an explanatory or dependent variable correlates with the error term and that may cause the inconsistency of model parameters which violates the reliability of the results [5].
Furthermore, as one of the characteristics of endogeneity, there is so-called “Simultaneity”. There often exists the difference between offline advertising that often needs to be fixed far in advance and is too be difficult to control on the way of delivering and online advertising that can control nearly instantaneously by watching their performances. In addition, given that it seems online advertising brings to higher sales that lead to higher profits thanks to the effect of the online advertising. In this case, it will be expected to have higher resources, or finally decide that might increase its online advertising. Now that the simultaneity concerns arise since online advertising and sales are assumed to mutually affected [6].
To deal with this situation, analyst have to control the data generation process e.g. to avoid determining advertising amount based on past sales. Describing causal relationship is expected to contribute to clarify the path of the effect to each variable and leads to solve the simultaneity situation.
Graphical Model as a solution
To describe the causal structure in MMM, introducing graphical model is one way so that it can express the dependencies between variables, both observed and unobserved and provide solution as raised in the above.
There is an existing research [7] uses structural equation modeling to clarify how brand search queries and online advertising in multi-channel retailers work across channels. The results of this research implies that there is significant structure between online or offline advertising, branded search queries and sales in which online or offline advertising is likely to contribute to sales through driving branded search queries. Based on that findings, marketers can clearly recognize the intermediate contribution of online or offline advertising and determine to continue advertising.
Therefore, it can conclude that it is possible to lead to clearer communication about the underlying causal structure of the media effect and it allows gaining a realistic and interpretable insights about how to improve their marketing activities [4].
Conclusion and Future expectation
As a conclusion, MMM allows for systematic analysis of a company’s marketing activities, but by modeling the structure of all marketing-related matters that can be converted into data, which are introduced as input to MMM, the situation surrounding a company’s marketing activities can be visualized. It is expected that this will improve the interpretability of marketing activities.
However, it is often argued that creating such a structure is highly difficult, and it is hoped that effective solutions will continue to be developed in the future.
Reference
[1] How to maximize a full-funnel media strategy with measurement
[2] Marketing Mix Modeling (MMM)- Concepts and Model Interpretation
[3] Measurement 360: A forward-thinking approach to measurement strategies
[4] Challenges and Opportunities in Media Mix Modeling
[5] マーケティング・サイエンスにおける個票データの課題と Marketing Mix Modeling の「再発見」
[6] Endogeneity and marketing strategy research: an overview
[7] The amplifying effect of branded queries on advertising in multi-channel retailing