By Sandro Catanzaro, co-founder & senior vice president, advanced analytics group
Knowing how to invest marketing dollars is the oldest question in marketing. It’s also a question that, with new advanced analytics and real-time modeling methodologies, is starting to be possible to answer.
Today, more than ever before, budget planning requires guidance that is accurate, timely and simple, yet, the traditional “bottom-up” and “top-down” approaches to marketing have failed to address these priorities, forcing trade-offs and choices that can negatively impact overall marketing performance.
Digital media attribution studies represent the bottom-up approach. They rely on the ability to assign an ID to consumers that are shown ads, and then link this ID to results achieved, on a consumer by consumer basis. The methodology has its advantages: as it is highly granular, digital media attribution studies can provide scores to individual inventory vendors, even when those are in the thousands. And, it can be executed quickly, through the use of computer technology and automated digital data collections.
But the complexity of digital media attribution studies makes it difficult to link different channels and platforms, and to de-duplicate consumers. Additionally, this approach does not easily link off-line advertising results. Digital media attribution studies are strongly affected by the data points chosen, with the simpler ones (last click, last view attribution) introducing extremely high distortion that favors vendors that flood the market with high volume of low impact impressions (impression bombing), or who are present in the very last minute (search). In this context, more involved and accurate methodologies may be unavoidable, yet, create complexity that few are able to stomach. As a consequence, most of the market relies on simpler methods that result in inaccurate budgeting guidance.
The top-down approach, also known as media mix modeling, relies on econometric analysis to identify correlation between changes in investment and changes in results. This methodology does not seek to identify consumers individually, but observes the behavior of the market as a whole, and this presents advantages. The method applies to cross-channel scenarios, including investments in non-digital inventory, as it relies on aggregated levels of investment as inputs. There is a rich history in creating this type of analysis, dating back to the 1960s, when it was created for justifying investments in TV. And these models are notionally simple, masking the complexity in seeking to compensate for external factors, such as competitor behavior, seasonality, distribution, events in connected market.
But media mix modeling has some weaknesses as well. These models predict a correlation but hardly causation; and as such, the bottom-up approach is more robust because it considers each consumer as an independent experiment thus enabling massive hypothesis contrasting. Media mix modeling can be slow to create and calibrate, since it requires long periods of data collection in order to acquire large enough datasets that enable reducing the model error range. This slowness makes it difficult to apply the model to decision making at the speed of digital advertising. Finally, this approach can be expensive, as it requires manual adjustments for reducing the effect of externalities.
What’s the answer? Convergence
Companies (like DataXu) that have the ability to do both model creation and model execution/testing are developing converged methodologies that close the gap between these two approaches, capturing key advantages of each.
This convergence approach incorporates bottom-up methodologies that use cross-channel IDs for consumers that provide actionable insight about multiple inventory vendors and across channels. Advanced mathematical models provide insight into what section of the conversion funnel is affected most by each channel, which results in a better appreciation for channels such as video that drive awareness.
Convergence also improves top-down methodologies by shrinking the size of the analysis cell, using time and space partitions. This is enabled by recent technology developments that make possible to accurately measure the causal effect of different online and off-line media investments. Patent pending methodologies that use a massive number of AB-Z testing experiments, combined with,faster data collection methodologies, including the use of POS data and store traffic data, are allowing top-down models with the pin-point accuracy of bottom-up methodologies.
The advertising community‘s interest in knowing where to invest marketing budgets is driving the development of these new methodologies. Companies with the ability to create advanced analytics and execute such models in real time are converging traditional digital methodologies and media mix modeling approaches, and as a result, are providing advertisers with insights that are not siloed and guidance that is free of distortions, creating models that are more accurate, timely and actionable–a scenario that is very appealing to CMOs and marketing teams everywhere!