Automating Complex Marketing Models for Insight That is Deep and Frequent

Speed versus depth of insight is a fundamental trade-off marketing strategists often face when working with analyst teammates. Both sides contribute to this reality. Strategists often ask nuanced or difficult questions (what is the ROI of a sports sponsorship?), and analysts or data scientists, reacting to a wide variety of these questions, tend to approach each new project from scratch.

This conundrum can be solved when convening agency and client strategists and data scientists. We believe the conversation should cover:

• What prior analytical work was most helpful to strategists?
• What new features would strategists like to see from analytical output, and what compromises are acceptable?
• Which data sources are reliable and are refreshed often?

Data scientists, now understanding the value of their past work, as well as having clear direction and an approved list of data sources, can then work on automated systems.

An example of what such a system could do is calculate and compare marketing ROI across detailed paid channels while accounting for the diminishing returns effects of increased marketing investment. These are sophisticated, non-linear models – the kinds marketing strategists associate with weeks or months of turnaround for each request. To be fair, it can take a few months to get this system up-and-running. But once that upfront work is completed, the outputs are automated and the insights valuable and frequent. A few thousand lines of code will import raw data, transform it as needed, test different models, pick the best one, and push the results to the strategist team.

Strategists can dig into the data and review fresh modeling outputs to inform thinking and next steps. Here are some use cases:

• The client wants to increase this week’s non-branded search budget by $10,000 to meet a business goal? OK! According to the outputs of the program, the first $2,000 should be in Exact Match and focus mostly on mobile in region A, the next $1,000 should be desktop Broad Match in region B, etc.
• How is our new display vendor doing? Easy! We are seeing great returns on low levels of investment but it really flattens out after that. There seem to be reach limitations and we should modify our spending accordingly.

An automated system like this allows strategists to deploy budgets fluidly and optimize across search engine or publisher, device type, region, and other dimensions. In contrast, most strategists still must pre-ordain budgets for each channel and try to optimize within them based on experience and simple reports. If they are lucky, they have access to marketing mix model results that optimize budgets by channel at a high level, and maybe only annually. Fast analysis is not limited to reporting, and sophisticated analysis does not have to be slow. But analysis, insights and reporting can make the difference – and should be a fundamental part of any marketing strategy and partnership.