Marketing Science, what is it and what could it be? (Part 2)
Continuing the series of blogs on Marketing Science, let’s once again reflect on John Wanamaker’s words:
To discern the effective spending on advertising, Marketing Science must engage in the measurement and attribution of marketing spend, eventually progressing towards budget optimisation.
The strategy’s focus hinges on the company and the composition of the marketing mix. If the majority of your budget is concentrated in one channel, a bottom-up strategy is advisable — starting with a channel-specific approach to thoroughly understand your primary marketing source before expanding to cover the entire marketing spend.
Conversely, if your channel mix is diverse and no channel stands out, a top-down strategy is recommended — initiating with an overarching focus on understanding channel interactions, assessing how the investment in each contributes to the company’s growth objectives, and subsequently delving into channel-specific refinements.
Consider the example of working with five paid marketing channels: Google Shopping, Google Display, Social Media (Instagram for example), TV and Radio ads.
- If your budget is allocated 75% to Shopping, 5% to Display, 5% to Instagram, and 10% to TV and 5% Radio ads, your initial efforts should concentrate on the largest channel.
- If your budget distribution is 40% to Shopping, 15% to Display, 15% to Instagram, and 25% to TV and 5% Radio ads, understanding the mix interaction becomes crucial.
Note that these approaches are not mutually exclusive; rather, they define a starting point for the quick wins on your budget. In the long run, implementing both is essential. In many companies, the scenario aligns with the second one, and Marketing Science plays a pivotal role in understanding the marketing mix.
Measuring and attributing marketing spend involves tools and business decisions, such as the Media Mix Model, Attribution Model, and Lift Tests. Each has its strengths and limitations.
The Media Mix Model provides a unified view of all channels but sacrifices detail due to aggregated data. It’s commonly used in monthly or quarterly budget planning sessions, allowing inclusion of business knowledge or external factors in the model for better predictions.
The Attribution Model facilitates day-to-day optimisation, particularly for online channels, relying on user data touch points. It assesses the quality of daily traffic generated with your marketing budget, offering various model options based on available data. Building an attribution model is complex and demands cookie-level data, potentially involving the development of probabilistic graph models.
Something intriguing to consider is that the focus of the Media Mix Model is on the marketing channels, while Attribution Models are oriented towards understanding the customer’s journey.
The integration of the Media Mix Model and the Attribution Model is achieved through Lift Testing, a channel or campaign-specific measurement of incremental impact — the perceived “source of truth” for marketing impact. While requiring thorough planning, time, and resources, Lift Tests enable calibration of the Attribution Model and the Media Mix Model.
It’s crucial to emphasise that marketing teams should not completely delegate the development of this trio of measurement tools to the Marketing Science team. Despite the considerable expertise that your Marketing Science team may possess, Marketing should remain cognizant that they are the domain experts and owners of the budget, channels, and strategy. A key principle to remember is that while Marketing delegates or assigns the construction of these data products to Marketing Science, the responsibility of managing the marketing budget remains squarely with the Marketing Team.
When it comes to measurement, marketing teams must define a crucial parameter, known by different names depending on the component under development: lag effect for the Media Mix Model or attribution window for Attribution Models. This concept signifies the maximum time you expect your advertisement to have an effect on your customer. Although it can be inferred from data, relying solely on this approach carries the risk of adopting a data-driven stance, placing more dependence on the model and fitting rather than the inherent nature of marketing. Personally, I would define these concepts as inputs to the models — instances of business knowledge — rather than outputs of the model.
For the Media Mix Model, another essential marketing concept to define is channel saturation. Introducing an estimated saturation per channel as input to the model prevents it from attributing everything to the most correlated channel. However, caution is needed at this stage, as incorrect model restrictions could lead to inaccurate measurements.
A distinctive aspect of the Media Mix Model, compared to the Attribution Model, is the historical data requirement. The Media Mix Model, with its aggregated data, necessitates an additional input: seasonality. This input can be obtained by collecting a sufficient amount of historical data. In terms of data time frame, Attribution Models demand a smaller amount of data over a specific period (the attribution window). However, it’s essential to note that strictly speaking, the volume of data within that time frame could be massive.
The Media Mix Model, Attribution Model, and Lift Tests predominantly center around measuring and comprehending specific calls to action: customer acquisition, customer purchases, app installs.
However, certain marketing channels, often referred to as upper funnel channels, wield most of their influence through indirect metrics, such as brand awareness. To gauge these indirect metrics, Marketing Science can provide support by developing causal inference-based models. These models are well-suited for measuring offline channels as well.
I hope you enjoyed the reading of the second article of this series. In following publications, I will try to keep covering some challenges associated with paid media optimisation, exploring how marketing science can support teams for a better customer understanding and how to integrate both to maximice the outcomes of your marketing efforts. Please leave comments about your experience with data and marketing to enrich our collective knowledge!
