Authors: Drs. Hung-Ken Chien and Yu-Feng Wei, Co-Founders of Vizuro
Introduction
In the ever-evolving pursuit to measure promotional effectiveness, marketing mix modeling (MMM) has long been the norm to assess campaign lifts and help allocate their spending. However, the emergence of omnichannel marketing in recent years has imposed a great challenge to the conventional MMM methodology, due to its requirements for high tactic mix and segment granularity. Coincidentally, both Google and Vizuro started to tackle this problem in 2017 in parallel yet similar routes - using causal models to replace linear regression as the new foundation for MMM - Google’s Meridian and Vizuro’s Kairos.
Google just released Meridian as an open source toolkit so we are able to compare it to Kairos. In short, both aim to measure the true impact of marketing efforts using Bayesian framework and allocate their spends with constrained optimization. However, Meridian has its scope pegged on digital media (because it is designed for Google’s clientele) with some inconvenient assumptions and limitations for brands seeking true omnichannel MMM tools. On the other hand, as it was developed for large brands, Kairos takes a more flexible and granular approach. It is capable of handling 50+ marketing mix for over USD$100M budget, and near individual levels of micro segments, making it a superior choice for modern marketing strategies and beyond, such as influencer marketing. In this article, we will compare both platforms and highlight their key differences.
Understanding Meridian
The Meridian MMM framework consists of four main modules:
The Meridian Model – Uses a Bayesian regression approach with non-linear transformations, incorporating AdStock for carryover effects and the Hill function for saturation effects.
Causal Estimation – Based on Rubin's potential outcome framework, it estimates incremental outcomes by comparing historical media spend scenarios.
Budget Optimization – Solves constrained optimization problems based on fixed or flexible budget goals.
Market Segmentation – Recommends running models at the geo-level (e.g., DMA) and assumes that campaign-level modeling is infeasible due to AdStock memory constraints.
Limitations of Meridian
While Meridian provides a well-structured approach, it has significant limitations:
OmniChannel Blind Spots – Meridian is designed for channel-level analysis and does not support campaign-level insights. As a result, it lacks the flexibility to handle modern omnichannel marketing strategies.
Limited Granular Segmentation – The model assumes analysis at the geo-level and struggles with more granular segmentations such as demographic or behavioral targeting.
Hierarchical Modeling Constraints – Meridian does not naturally accommodate hierarchical modeling where different data sources exist at different levels (e.g., geo-level for some channels, national-level for others). This limits its ability to integrate a complex marketing mix.
Lack of Time-Varying Lifts – Meridian assumes that marketing performance is consistent over time, failing to account for seasonality and trending effects. Unlike Kairos, which dynamically models time-dependent effects, Meridian relies on static assumptions.
Lack of Pathway and Synergistic Interactions – Meridian cannot measure cross-channel interactions such as how TV impacts search engine marketing, a crucial aspect of modern digital strategies.
Lack of Constrained Optimization – Meridian does not support individual budget bounds across different tactics, a necessity for real-world media planning.
Why Kairos?
For its own good, Meridian primarily focuses on digital marketing, limiting its applications compared to Kairos, which is a true omnichannel MMM platform. The key differentiators include:
Multimodal Capabilities – Kairos processes omnichannel data with different data modalities, providing a more holistic analysis for the true omnichannel mix.
Causal AI for True Incrementality – with the capability to incorporate omnichannel mix, Kairos can accurately determine the true incremental impacts by minimizing unobserved confounders.
Generative AI for Hierarchical Modeling – Kairos uses generative AI to simulate response curves across the combos of market attributes (e.g., DMA, demographics, gender), so that every micro-segment can be analyzed and optimized.
Advanced Hierarchical Modeling – Unlike Meridian, which is limited to geographical segmentation, Kairos seamlessly handles hierarchical structures, such as:
Geographic: National -> DMA -> Zipcode
Demographic: Age Group -> Education -> Professional
Peer-to-Peer: Influencer Agency -> Creators
Dynamic Time-Series Modeling – Kairos does not require parameterized carryover effect settings, allowing for dynamic time-series adjustments such as trending and seasonality. This methodology proved invaluable during the COVID-19 pandemic when one of our clients successfully adjusted its marketing strategies by using the prior time-series pattern during the summer vacation for the home quarantine period.
Enhanced Cross-Channel Synergies – Kairos measures the interaction between channels, such as pathways and synergies, ensuring that marketing efforts are not analyzed in isolation but as part of an interconnected ecosystem.
Flexible Budget Constraints – Kairos allows realistic budget allocations that align with actual business constraints such as contractual budgets for channels.
Conclusion
While Google’s Meridian provides an open-source, revamped framework for marketing mix modeling, it still falls short in addressing the complexities of modern marketing. Vizuro’s Kairos offers a more powerful, granular, and flexible approach, making it the ideal choice for businesses looking to optimize omnichannel campaigns. As the marketing landscape continues to evolve, embracing more sophisticated and adaptive models like Kairos will be crucial for staying ahead in the data-driven economy.
For more insights on marketing analytics and AI-driven decision-making, stay tuned to The AI Frontier – Vizuro Knowledge Corner.
Kommentare