1. Introduction
1.1 Challenges in Measurement
As digital marketing continues to evolve, traditional attribution methods, such as those reliant on cookies and tags, are rapidly losing effectiveness. Privacy regulations, browser restrictions, and platform fragmentation have introduced significant barriers, complicating the once straightforward task of tracking user behaviour across channels. While innovations such as Meta’s Conversion API, Google’s Enhanced Conversions, and server-to-server integrations help bolster signal resilience within individual platforms, they fall short of providing a comprehensive cross-channel view.
Advertisers now face an urgent challenge across channels: orchestrating performance across increasingly fragmented digital landscapes. With traditional methods no longer providing a “perfect” solution, marketers have three options:
- Wait and hope for a new universal tracking standard to emerge — a strategy often referred to as “waiting for biscuits.”
- Develop adaptive identity-tracking systems that remain compliant with evolving regulations.
- Embrace probabilistic measurement: a data-driven approach leveraging statistical methods to estimate performance, enabling continuous refinement and actionable insights.
The complexity of digital measurement extends beyond technical challenges. Marketers are increasingly overwhelmed by the sheer volume of available methodologies, tools, and practices. This complexity is compounded by:
- Platform proliferation: The growing number of media channels fragments data and strategies.
- Walled gardens: Platforms often restrict data sharing, offering increasingly opaque “black box” solutions.
- Divergent truths: Competing attribution models present conflicting views on performance.
- Platform bias: Each platform’s attribution systems tend to overvalue its own impact.
1.2 Modern Measurement
The evolving challenges in digital marketing measurement have spurred a shift towards what is now widely referred to as “Modern Measurement.”
This approach acknowledges that the era of relying on a single source of truth is over. Instead, the industry has embraced the concept of triangulation, where multiple methodologies and data sources are used together to validate and cross-check insights. Typically , this involves using:
Marketing Mix Modelling (MMM) to assess long-term, strategic impacts.
Attribution models, such as multi-touch or last-click, to capture more granular, immediate performance data.
Incrementality testing to measure causal relationships in controlled experiments directly.
Modern Measurement places significant emphasis on ongoing validation. Incrementality testing, for example, remains central to assessing the true impact of campaigns. Meanwhile, Marketing Mix Modelling (MMM) provides high-level, channel-specific ROI insights, and attribution offers tactical guidance for campaign optimisation. Together, these methods form a complementary suite of tools that enable marketers to make informed decisions despite data fragmentation.
At the heart of Modern Measurement lies a commitment to continual testing and validation, leveraging fresh data to refine strategies consistently. Advertisers who run at least 15 experiments per year achieve about 30% higher ad performance1 within the same year and 45% higher1 in the subsequent year, underscoring the importance of regular experimentation. Similarly, businesses that treat MMM as an “always-on” process, maintaining regular refresh cycles over multiple years, have observed an average 151% growth in return on advertising spend2 (ROAS), demonstrating the transformative potential of sustained measurement strategies.
In recent years, substantial investment from major industry players such as Google and Meta has further accelerated the adoption of advanced measurement technologies. Google’s Meridian solution, for instance, utilises Bayesian regression to model complex relationships and predict outcomes, while Meta’s Robyn framework employs Ridge Regression to optimise media allocation and measure incremental impact. These advancements highlight how cutting-edge statistical methods are shaping the future of privacy-first, data-driven marketing.
2. Lessons from Robyn: A Benchmark in Marketing Mix Modelling (MMM)
Robyn’s Key Contributions to Marketing Attribution
Meta’s Robyn, a ridge regression-based model, builds on econometrics techniques and offers key advantages for marketing attribution. As an open-source solution, it addresses several challenges:
- Privacy Compliance: Robyn’s privacy-first approach eliminates the need for sensitive personal data, such as cookies or personally identifiable information (PII), making it accessible even in highly regulated environments.
- Methodological Robustness: Using ridge regression, Robyn generates statistically reliable results while controlling for multicollinearity – the inter-relationship of variables found in digital media, offering robust insights across different data sets.
- Customisation: Its flexibility allows businesses to tailor the models to specific data types and marketing goals, ensuring precise analysis.
Robyn has proven to be highly effective for analysing historical data, providing businesses with a powerful, privacy-compliant framework for Marketing Mix Modelling (MMM). It offers a foundation for enabling economic opportunity and empowering organisations to succeed through its open-source nature. As a neutral platform, Robyn’s fully open code encourages innovation, allowing businesses to deploy it as-is or customise and extend its capabilities.
Robyn’s role remains distinct from that of third-party MMM providers, which deliver more comprehensive services. These include integrating diverse data sources, offering market expertise, and providing tailored support, consultation, and analytics—capabilities that extend beyond Robyn’s out-of-the-box functionality.
Robyn offers exciting potential for future advancements, including the ability to automate attribution and accelerate the optimisation of marketing strategies. With several models available, it provides an invaluable resource for technical teams to uncover actionable insights. By simplifying access and streamlining model selection without compromising analytical power, Robyn can make these insights accessible to a broader audience, empowering more organisations to harness its value.
3. PRBA: Evolving the Concept for Modern Attribution Needs
While Robyn offers a comprehensive Marketing Mix Modelling (MMM) approach, Publicis Regression-Based Attribution (PRBA) addresses the evolving need for a cookieless measurement solution that can be put in the hands of media practitioners. By automating data aggregation, navigating evolving regulations, and delivering cross-channel insights, PRBA transforms open-source inspiration into a robust, scalable solution for today’s complex digital ecosystem.
Rather than replacing MMM, PRBA serves a complementary role, offering the precision that traditional methods often cannot consistently deliver. Through frequent and granular insights, PRBA helps advertisers orchestrate performance across platforms, identifying actionable opportunities and seamlessly integrating with broader data science and measurement initiatives. Whether deployed as a standalone tool or as a complement to existing systems, PRBA empowers advertisers to navigate the measurement landscape with clarity and confidence.
3.1 Bringing RBA to Life
To make regression based attribution a core component of any digital measurement framework, Publicis have productised PRBA—an automated system with an interactive interface—to get the data and insights into the hands of the decision-makers.
The system has three features to ensure the products meets the needs of the end-users:
- Real-Time Data Integration: PRBA integrates real-time data sources to support dynamic decision-making. This feature ensures marketers can track the performance of their campaigns as they unfold, allowing for immediate adjustments and more efficient budget allocation.
- Ease of Use: Designed with accessibility in mind, PRBA provides a user-friendly interface that simplifies the reporting process. Marketers can quickly generate insights without needing in-depth technical knowledge, making it more approachable for teams of all skill levels.
- Actionable Recommendations: PRBA provides data and offers actionable insights, guiding marketers on where to optimise their efforts. This information helps them focus on high-impact strategies and use their resources effectively to drive better results.
3.2 How This Fits into Modern Measurement
PRBA fits into the broader framework of Modern Measurement, complementing other tools such as traditional MMM, multi-touch attribution, and incrementality testing.
- Fitting into the Triangle: PRBA is part of the triangulation model in modern measurement, working alongside other techniques to validate and refine data insights. It focuses on the digital realm, while traditional MMMs still excel in analysing the contribution of both online and offline marketing efforts.
- Focus on Frequency and Granularity: Unlike traditional MMMs, which provide high-level strategic insights on an infrequent cadence, PRBA excels at offering detailed, granular views of digital marketing performance each week. It allows marketers to zoom in on specific campaign elements and audience segments, offering a deeper level of understanding.
- Validation through Incrementality: PRBA should be validated through incremental lift tests to ensure its accuracy. These experiments help assess the true causal impact of marketing efforts, feeding back into PRBA to refine its data and improve decision-making.
PRBA works in tandem with MMM and Incrementality testing to form a comprehensive measurement ecosystem where each technique adds value to the overall process.
4. Key Findings from PRBA
PRBA provides transformative insights into digital marketing attribution by addressing common challenges such as data collinearity and the limitations of ID-based attribution methods. Employing hierarchical ensemble modelling reveals deeper relationships between marketing channels, enabling marketers to optimise performance and refine strategies.
4.1 Solving for Multicollinearity
Multicollinearity—a key challenge in digital media measurement—occurs when variables are highly correlated. For instance, broader media activity often influences Brand PPC and Affiliates, making it difficult to disentangle the contributions of each channel. A typical example is the launch of awareness campaigns, which can drive increased search activity and inadvertently inflate the perceived effectiveness of paid search channels.
To address these challenges, PRBA employs a range of techniques:
- Variance Inflation Factor (VIF): This metric quantifies the extent to which a variable is explained by other variables, helping to identify problematic correlations.
- Pairwise Correlation Analysis: This approach measures the strength of linear relationships between variables, enabling informed decisions on channel segmentation.
- Channel Aggregation: In cases where two channels are highly correlated, they can be aggregated, and credit for conversions apportioned based on spend or delivery.
- Incrementality Testing: This technique validates the independent contribution of each channel, ensuring that credit is distributed accurately.
Social media activity presents similar challenges due to collinearity across platforms such as Meta, TikTok, Snap, and Pinterest. These platforms often display correlated performance patterns because campaigns are typically planned and executed in bursts. PRBA addresses these complexities by automating key modelling decisions. By dynamically adjusting model parameters and enabling rapid, scalable remodelling, PRBA ensures accurate attribution while maintaining its value proposition.
In summary, these solutions mitigate the effects of multicollinearity and enhance marketers’ ability to evaluate channel performance, enabling better decision-making and maximising return on investment.
4.2 Common Applications
PRBA has delivered significant insights into improving marketing attribution and campaign strategies. Below are the key takeaways based on its application:
- Budget reallocation testing: PRBA often revealsan increase in reported ROI for Impression-based and audience-targeted media. For example, Meta’s performance is frequently undervalued by a factor of 1.5x when relying on traditional analytics like GA4. With automated data extraction at the coreof PRBA, it delivers more frequent digital attribution reads per annum compared to other non-MTA modern measurement solutions. Through regular remodelling, advertisers can validate their testing decisions over time, ensuring confidence in maximising digital returns.
- Econometric validation: PRBA has a 6-week onboarding process that provides advertisers with immediate insights to validate existing attribution models, unlike the multi-year timelines often required for traditional MMM setups. By leveraging automated data pipelines, PRBA delivers a consistent view over time and integrates seamlessly existing attribution frameworks, enabling rapid and reliable decision-making.
- Incrementality prioritisation: Due to the collinear nature of PPC tactics (commonly Brand, Generics & PMax), PRBA often highlights the need for tactic-level conversion lift tests within platforms. These tests help differentiate the individual contribution of each tactic, which can then be used to calibrate and improve the models over time. PRBA supports the development of systematic testing roadmaps, allowing advertisers to validate the incremental impact of marketing activities. Findings indicate that regular testing significantly improves ROI, equipping marketers with actionable insights for ongoing budget optimisation.
5. Conclusion: Future Directions for Modern Measurement
Agency-Led Measurement and the Closed-Loop System
PRBA drives a broader movement towards agency-led measurement, where teams generate insights and actively apply them within marketing strategies. This approach creates a closed-loop system, ensuring teams swiftly implement recommendations rather than letting them remain theoretical. By closing the loop between insights and execution, Performics can rapidly refine campaigns and adapt as the market evolves.
Shaping the Future of Performance-Driven Marketing
As the marketing landscape continues to evolve, Performics embraces Performance Orchestration to coordinate marketing efforts better and drive improved performance. This methodology unifies all aspects of a campaign, enabling teams to leverage insights from multiple channels and data points to craft a cohesive strategy. By breaking down traditional silos, Performance Orchestration provides a unified view of campaign performance, making it easier for marketers to optimise across platforms.
Embracing New Opportunities for Smarter Decisions
Modern measurement will focus on iterative improvements rather than unattainable precision, using available tools to drive smarter decisions. Marketers must strike a balance between forward-looking solutions and practical, on-the-ground strategies. This challenge is both operational and methodological. Successfully navigating the shift from ID-based measurement requires proactive stakeholder management and a commitment to modern solutions.
“At Performics, we believe the future of digital media lies in performance orchestration. The agencies that will lead the way are those who truly understand the levers at their disposal and can expertly activate the insights within the various algorithm-driven platforms to maximise growth for their clients. Publicis Regression Based Attribution (PRBA) provides advertisers with powerful tools and consistent measurement to understand the true impact of their digital investments. This consistency empowers smarter, data-driven decision-making, ensuring businesses can optimise strategies and allocate resources more effectively.”
Paul Kasamias, Managing Director, Performics UK
Sources
- Harvard Business Review, Marketers Underuse Ad Experiments. That’s a Big Mistake. Oct. 2020 [link]
- Deloitte, AI Marketing Transformation & MMM, September 2024, [link]
With thanks to Meta’s Igor Skokan and Camille Bernard for comments.