Do you need an open-source MMM?

In a previous article, we reviewed How open-source MMM frameworks democratize MMM usage. To summarize, cost-effectiveness, flexibility, transparency, and the ability to fully adopt and manage the solution are the main factors driving interest in open-source MMM packages.

‍

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

‍

Which advertisers are the open-source models aimed at?

Based on fifty-five’s experience, three types of advertiser contexts/expectations stand out:

Advertisers new to MMM. Adopting open-source packages lowers entry barriers by eliminating high costs, making advanced marketing analytics accessible to businesses of all sizes, including SMEs. Additionally, it offers tailored solutions for specific industries or marketing contexts that generic SaaS platforms often overlook.

Advertisers who have previously conducted occasional MMM analysis with an external provider and now seek more frequent updates. Due to budget constraints, these studies were often performed annually before budget allocation. However, the approach is increasingly outdated in today’s fast-paced media ecosystem, which requires continuous optimization and more frequent decision-making.

Advertisers who only use MMMs for a small number of their brands. A common scenario occurs in CPG companies that apply MMM only to a few power brands, despite managing substantial investments across a broader portfolio. Open-source packages make it financially feasible to extend MMMs across all brands, enabling new use cases such as optimizing marketing resource allocation across the portfolio and measuring halo effects.

Advertisers who want to internalize an MMM-based process. In most cases, turning to open-source libraries serves as an ideal entry point for organizations looking to evaluate opportunities or begin the in-housing process. This approach is typically motivated by several factors: the desire for customization and flexibility, the need to delve deeper into the advertisers' first-party data or to get faster / more frequent insights, a requirement for transparency and control, the drive to cultivate a data-driven culture within the organization, and considerations of cost sensitivity and budget constraints.

‍

How to succeed in open-source MMM deployment?

Open-source packages are undeniably transformative, providing advertisers with a practical catalyst to improve their approach to MMM projects and to partially or fully bring decision-support solutions in-house.

Although data science – particularly the core ‘modeling’ work - is central, focusing solely on this aspect without considering the broader requirements for successful implementation poses a significant risk of underachievement. Even if it may seem obvious, we must emphasize the importance of:

> Skilled team with a broad range of expertise: Assemble a diverse group, including data scientists with expertise in marketing models, data and business analysts, senior media experts (across both online and offline channels), and marketing professionals. This combination ensures that the models are not only technically robust but also aligned with business insights and relevant to your marketing objectives. It's also advantageous if the team has experience with - or can be guided by experts in - the specific open-source tools to be used, as this can greatly streamline the implementation process.

>High-quality data: An MMM project is, first and foremost, a data project. Invest time in carefully selecting, cleaning, and organizing data from the outset. Having the right level of granularity in each data source is really a key success factor (learn more in “Granularity, a key to measure and optimize marketing effectiveness”). This should encompass all relevant business drivers and proxies, from sales figures and marketing expenditures to external factors like economic indicators. This foundational step is indispensable for any MMM project and will undoubtedly pay off in the long run by enhancing the accuracy, robustness, and reliability of the models.

> Stepwise / iterative process towards customization: Start with pilot projects by applying open-source packages to select geographies or specific product lines and embrace a mindset of continuous learning and improvement. Open-source tools provide the flexibility to fine-tune and refine models as needed—whether that involves integrating unique data sources, experimenting with different methodologies, or adapting to new market conditions. A key corollary of this approach is the integration of test results and marketing effectiveness experiments into the framework, which not only enhances the explanatory potential of the models but also promotes internal adoption.

> Sustainability and scalability: Adopt an ongoing approach, as an MMM project is not a one-time effort. While quick wins are important and necessary, it’s essential to plan from the outset for the long-term sustainability of the solution. This includes regular updates, model maintenance, and adaptation to new data or business requirements, ensuring that the solution drives lasting change rather than serving as a short-term fix. Additionally, designing the solution with scalability in mind allows it to be expanded to other markets, countries, and broader business lines. This approach ensures that the solution can grow alongside the business, accommodating increasing data volumes and complexity over time.

‍