Measuring marketing performance has become a high-stakes balancing act. Today’s marketing leaders are expected to deliver accurate, data-driven insights at speed, without compromising the rigor required to guide critical budget decisions.
“Traditional measurement forces a trade-off: either we are meticulous but slow and rigid with models that struggle to keep up, or we get fast insights but oversimplify reality in our results.”
Kimberly Leung
AI/ML Group Product Manager, Adobe Mix Modeler
Adobe is no stranger to this challenge. Nearly a decade ago, the Adobe marketing team embarked on a journey to transform how they measured performance. While the landscape has changed dramatically since then, today’s marketing leaders continue to face the same core challenges:
What began as an internal initiative to unify disparate methodologies and accelerate speed to insight ultimately evolved into a commercially available product, Adobe Mix Modeler. While Adobe Mix Modeler applies a wide range of advanced techniques, the sections below highlight key capabilities in AI/ML and statistical modeling that power its measurement and planning workflows.
“Each measurement method has its own blind spots. But when layered together, they start to fill in the gaps left by one another. It’s like stacking slices of Swiss cheese — each piece has holes, but when you stack the pieces, the holes close.”
Matt Scharf
Vice President, Growth Marketing Performance & Global Media Center of Excellence, Adobe
No single measurement method provides a complete picture of marketing impact. Marketing Mix Modeling (MMM) offers strategic, long-term insights; Multi-Touch Attribution (MTA) captures user-level behavior across digital channels; and experimentation delivers direct evidence of causality.
Each method has distinct strengths — but also has clear limitations when used in isolation. MMM uses aggregate data, limiting its ability to track individual behavior. MTA relies on digital-only data that is increasingly constrained by privacy regulations and data deprecation. Experimentation is often resource-intensive, time-consuming, and prone to error.
Recognizing these trade-offs, the Adobe marketing team set out to build a solution that harnessed the strengths of each method, while minimizing their weaknesses. Adobe Mix Modeler brings this philosophy to life — combining MMM, MTA and experimentation into a unified measurement framework within a single UI.
While it doesn’t execute experiments directly, Adobe Mix Modeler ingests and incorporates experimental results — using them within its bi-directional transfer learning capability to improve model accuracy and strengthen causal inference. Beyond just measurement, Mix Modeler also connects planning and forecasting workflows in the same user interface, enabling marketers to simulate scenarios, optimize budgets, and make impactful investment decisions with confidence. This integrated measurement and planning approach empowers marketers to make smarter, faster decisions that drive ROI, achieve revenue targets, and support strategic business goals.
Marketing Mix Modeling (MMM) analyzes performance at the aggregate level, typically across daily or weekly time periods. In the sample dataset shown, each row corresponds to a week, while columns capture variables such as media spend and factors data (e.g., inflation rate, promotions, seasonality). The model evaluates how these inputs, taken together, drive business outcomes such as sales, ROI or conversions over time.
The MMM engine in Adobe Mix Modeler uses a multiplicative nonlinear regression model which closely mirrors the complexity of real-world marketing dynamics. Instead of summing the effect of each channel, as additive models do, the model treats weekly conversions as the product of baseline demand and the combined influence of marketing channels. This structure better accounts for key marketing dynamics such as media synergies, time-varying effects, and budget reallocation across time — none of which are well-handled by additive models.
Unlike MMM, which relies on aggregate-level data, Multi-Touch Attribution (MTA) is built on event-level data and uses stitched customer journeys to assess the relative impact of different online marketing touchpoints. For the MTA engine, Adobe Mix Modeler uses a discrete-time survival model, which more accurately reflects customer behavior by capturing the delayed impact of marketing (ad stock) in flexible, time-specific windows, without relying on restrictive continuous-time assumptions.
“Both types of paths — converting and non-converting — are important in building multi-touch attribution models. The contrast between the two tells us if a marketing touchpoint is effective in driving a conversion… or not.”
Bowan Wang
AI/ML Engineering Manager, Adobe Mix Modeler
Unlike rule-based or naïve algorithmic models, the MTA engine in Adobe Mix Modeler trains its machine learning model with both converting and non-converting paths to deliver a more precise understanding of incremental performance.
This contrast is key — when a touchpoint occurs as frequently in non-converting journeys as in converting ones, its incremental impact is likely overstated if you analyze only converting paths. But, if it consistently shows up only in paths that lead to conversion, the model flags it as a stronger driver of performance. By analyzing these patterns at scale, MTA isolates the true incremental contribution of each channel across the conversion journey.
Bi-directional transfer learning: bridging MMM and MTA results for unified, actionable insights. Adobe Mix Modeler uses bi-directional transfer learning to automatically connect its two core models — MTA and MMM — so they can learn from each other and produce consistent, comprehensive insights.
By training both models on overlapping time periods, MTA scores feed into MMM to enrich it with incremental impact of channels and campaigns with user-level detail, while MMM outputs loop back into MTA to calibrate how it evaluates the impact of touchpoints on user paths. This behind-the-scenes, two-way exchange ensures alignment across measurement layers and improves downstream outputs like budget recommendations.
A unified measurement framework across MMM and MTA supported by bi-directional transfer learning offers several key benefits:
It enables marketers to move beyond siloed models that often produce conflicting insights and toward a single, integrated system built for speed, scale, and strategic planning1.
Looking ahead, Mix Modeler will support budget optimization across multiple conversion goals. For example, weighting between in-store and online purchases, or optimizing for both lower-funnel purchases and upper-funnel awareness, in an upcoming roadmap feature, Portfolio Planning.
When the Adobe marketing team set out to transform how they measure marketing effectiveness, they asked themselves: How can we keep pace with today’s fast-moving marketing landscape where speed and accuracy both matter?
The result? A powerful, AI-driven solution built for accuracy, speed and scale, enabling marketers everywhere to make smarter decisions today and plan with confidence for tomorrow.
Read this guide to learn more about the Adobe Mix Modeler measurement methodology.
Hear from Kim and Bowen as they explore these and additional AI/ML techniques powering the intelligence behind Adobe Mix Modeler.
Sources
1 While this unified approach is recommended and used internally at Adobe, the framework is designed to be flexible in its implementation across customers — recognizing that not all teams are ready or looking to adopt both MMM and MTA at once.
Kimberly Leung is a Group Product Manager for Adobe Mix Modeler where she leads AI-driven solutions that empower brands to make practical, data-informed marketing decisions. She previously worked on AI/ML applications in Adobe’s Intelligent Services team and led TV planning in Adobe Advertising Cloud. Innovation in marketing measurement and planning. With a degree in Electrical Engineering from the University of Waterloo, Kimberly brings a unique ability to bridge technical depth with business strategy to drive innovation in marketing measurement and planning.
Bowen Wang leads AI/ML engineering for Adobe Mix Modeler where he focuses on building scalable solutions that power marketing measurement and optimization. Over the past seven years at Adobe, he has held various roles across data science and engineering, driving innovation at the intersection of machine learning and marketing analytics. Prior to Adobe, Bowen earned a PhD in Statistics from the University of Washington and a Master’s in Statistics from Yale University.
