Utilising the Delphi Method to Construct Priors in Marketing Mix Modelling
Introduction
In the context of Bayesian Marketing Mix Modelling, and Bayesian methods in general, the building of priors is a key component. The concept of priors is occasionally regarded as controversial, with frequentists often arguing that the Bayesian approach is ‘too subjective’ and not data-driven. However, it can be comfortably countered that the frequentist approaches are also inherently subjective due to factors such as the choice of model and preconceived notions of what constitutes a good model. For example, you would never finalise a marketing mix model built using frequentist method or OLS which contained negative media impacts, even in the scenario of it being a ‘statistically good model’. Consequently, the imposition of priors in Bayesian Marketing Mix Modelling lends itself towards a transparent and clear methodology to incorporate domain knowledge across a business. One approach to achieve this can be the Delphi method.
The Delphi method is a formal and structured communication technique aimed at achieving consensus on a specific topic by engaging industry experts with substantial domain knowledge. This approach is particularly effective in situations characterised by high uncertainty, where a unified consensus is absolutely necessary. Originating from the Cold War era, the Delphi method, like other advances in economic theory such as game theory, was developed to assess the probabilities of various attack scenarios. More recently people will be familiar with the methodology during the COVID pandemic, where it facilitated the integration of multiple expert opinions into a single consensus to predict potential outcomes of the spread of the disease. By involving a diverse group of experts, the Delphi method accommodates a wide array of unique perspectives, ensuring the development of well-informed and robust estimates.
In the context of marketing mix modelling, domain industry expertise can be key to obtaining a precise and accurate model. Which necessitates both correct inference about past performance of campaigns and the ability to be able to the MMM for forward planning.
Objectives of Delphi
Develop a structured process to harness expert knowledge and opinions in constructing priors for marketing mix models, ultimately improving model performance and prediction
Methodology
The approach is broken down into key sections over a period of time to ensure a comprehensive result:
i. Select the experts:
- Gather a diverse panel of experts from different parts of the business who have extensive experience in marketing mix modelling, econometrics, and industry-specific knowledge. If you’re an agency delivering cross-industry, ensure it doesn’t just focus on one industry, likewise if you’re a global client, ensure a diverse range of markets are covered.
ii. First round questionnaire:
- Distribute a comprehensive, anonymous questionnaire to the panel, addressing key aspects of marketing mix modelling and prior construction of different marketing channels. Within the questionnaire focus on the range of believable marketing results, not just point estimates .
- Encourage experts to provide detailed reasoning behind their opinions.
iii. Feedback consolidation:
- Collect and summarise the responses, drawing attention to areas of agreement and disagreement. Identify any trends or patterns in expert opinions.
iv. Second round questionnaire:
- Distribute a revised questionnaire that incorporates the feedback from the first round, allowing experts to revise their opinions in light of the group’s input.
- Repeat the feedback consolidation process.
v. Convergence and consensus:
- Continue iterating the questionnaire and feedback process until a satisfactory level of consensus is reached.
- Analyse the resulting priors, potentially averaging the finalised set of distributions and integrate them into the marketing mix models.
Expected Outcomes
A successful Delphi method should deliver the following:
- A set of well-defined priors that reflect expert consensus and improve the accuracy & performance of marketing mix models.
- A formalised approach to gain consensus across all parts of the business of domain expertise, and more importantly how to directly incorporate this into Bayesian Marketing Mix Models
- A documented process for using the Delphi method in future marketing mix modelling projects.
Delphi itself can take some time to undertake, particularly to gain a final consensus. However once you have a formalised approach to undertake the concept, it can be re-visited at any time. Naturally, over time the performance of different media channels change, there are always different factors to consider, e.g. COVID, Recessions & changes in media penetration and so on. A documented process which can help in future creation of priors is key promoting a successful marketing effectiveness programme and alleviate and concerns or frustrations within the business of the concept of priors.
Author: Niall Oulton, Company: 1749
