Open-source MMM: transitional tool or permanent solution?

This article is co-written by Mathieu Lepoutre and Arnaud Parent.

‍

‍

For many organizations, open-source MMM libraries can be the core element of a long-term or even permanent solution. This is especially true when:

• Sufficient flexibility and power: The open-source tools meet the organization’s needs in terms of flexibility, customization, and performance. Many open-source MMM packages, such as those built on Bayesian frameworks, are highly sophisticated and capable of handling complex models effectively.
• Customization capabilities meet the needs: Organizations that require a high degree of customization or operate in niche markets may find that open-source tools provide the control and adaptability they need. These tools can be tailored to specific business contexts in ways that might be difficult with off-the-shelf commercial software.
• Cost sensitivity: Some organizations, particularly smaller ones, may find that open-source solutions offer a perfect balance between cost and functionality, making them a viable long-term option without the financial burden of expensive proprietary tools.
• Community support and sustainability: Open-source tools with strong, active communities often receive continuous updates, enhancements, and support, making them sustainable for long-term use. As long as the tool evolves with the industry’s needs and technological advancements, it can remain a valuable asset indefinitely.

For the vast majority of advertisers, open-source MMM provides an easy-to-implement solution that effectively and sustainably addresses their needs while delivering a quick return on investment. For more details, you can consult our previous article: Do you need an open-source MMM?  And for brands that invest tens of millions in advertising, open-source packages are a starting point or a stepping stone toward more complex, sophisticated models and solutions. Open-source libraries perform very well, but as advertisers become more sophisticated, they require additional analyses across more dimensions. For instance, this includes a distinction by product categories, by customer or audience segments, by distribution channels, or by differentiating between brand and branding campaigns. While open-source models can effectively generate insights for one dimension, there are still data science techniques to handle two dimensions in an open-source library. However, when it comes to three or more dimensions, open-source models tend to fall short.

It’s also the case for companies operating in complex markets - such as those with highly segmented customer bases, multiple sales channels, or significant external influences like regulatory constraints and economic fluctuations - often discover that the initial models developed using open-source packages do not fully capture the nuances of their market. In such cases, moving to more sophisticated and customized models still offers significant benefits and quick packback. Similarly, while open-source packages provide flexibility, they may not always accommodate the specific needs of certain markets. For example, companies in rapidly changing industries or those with unique customer behaviors might require advanced models that can integrate real-time data, manage complex interactions, offer granular insights, predict future trends with greater accuracy, or provide advanced scenario planning and simulation capabilities. Search could also be another good example detailed in this article How to accurately model Search Engine in MMM with Agent-Based Models.

Furthermore, for companies that are in the first phase of in-housing MMM, open-source tools can be instrumental in exploring the benefits of in-housing or creating proof-of-concept models in a low-risk, cost-effective manner. They allow advertisers to iteratively refine their models based on the insights and feedback they receive, test hypotheses, and experiment with different approaches, as well as build foundational models and gain initial insights.

Besides, these packages offer a very operational way to get up and running, optimize the learning phase, build internal expertise, and refine their understanding of MMM. This phase can provide valuable experience before transitioning to more advanced, possibly proprietary, solutions that offer higher performance or additional features.

Once a company identifies the specific gaps in the open-source models, it can begin transitioning to more complex solutions better suited to these unique demands. In this context, 55 gradually deployed various solutions, initially leveraging open-source algorithms as a foundation, which were then hybridized with other approaches, including agent-based modeling, adding functional enhancements that delivered significant decision-making advantages. Discover how to dive deeper in your marketing strategy simulations with Agent Based Models.

‍