Beyond Attribution: Causality & Incrementality with Professor Kenneth Wilbur

The word “incrementality” has become one of the most fashionable—and misused – terms in modern marketing. Vendors frequently promise incremental measurement, but often deliver dressed-up correlational data that takes credit for sales that would have happened anyway.

In this episode, Rajiv Nayer is joined by Professor Kenneth Wilbur, Professor of Marketing and Analytics at the Rady School of Management (UC San Diego) and the newest member of Lifesight’s Scientific Advisory Council. Professor Wilbur breaks down the fundamental difference between correlational and causal measurement, explains why the world’s most sophisticated AI models cannot fix bad data, and explores the critical need for alignment between marketing, finance, and data science teams.

Key Takeaways

• Incrementality is Causal Lift: True incrementality isolates the specific impact of the ad itself, removing the effects of targeting, timing, and brand momentum.
• Incrementality Washing: Many tools claim to measure incrementality but are fundamentally correlational. Without a valid control group or quasi-experiment, you are measuring targeting efficiency, not ad effectiveness.
• Causality is a Data Problem, Not a Modeling Problem: A machine learning model or neural network cannot create causality out of thin air. You must have a strict identification strategy (comparing apples to apples) before you ever select a model.
• MMMs Are Not Magic: Marketing Mix Models (MMMs) are incredible tools, but without a causal identification strategy (like holdout experiments) to calibrate them, an MMM is just a correlational tool.
• Measurement Must Drive Decisions: Measurement is only valuable to the extent that it changes or improves a business decision.

Core Discussion Topics

1. Bridging the Gap Between Economics and Marketing

Professor Wilbur’s journey spans software engineering, theoretical economics, and marketing analytics. Early in his academic career, theoretical economists assumed companies only bought advertising if they possessed strict proof of its profitability. In reality, the statistical challenges of proving ad effectiveness are immense. Wilbur shifted into marketing science to get closer to the actual data, understand how managers make decisions in the real world, and collaborate directly with industry practitioners.

2. The Open-Access Measurement Deck

Professor Wilbur recently published a revised, open-access deck titled Introduction to Advertising Measurement, giving it away to the industry for free. He designed it to serve three distinct audiences:

Technical Practitioners: Data scientists and analysts coming from physics, economics, or computer science who need unbiased information to understand the exact data-generating processes behind marketing numbers.

Marketing & Finance Executives: To successfully allocate budgets, marketing and finance must speak the same language. Finance often underestimates the complexity of ad measurement, while marketing often lacks the econometric background to prove their impact.

The Broader Industry: By publishing his work openly, Wilbur invites practitioners to correct him, ask questions, and continuously push the industry forward.

3. Causal Lift vs. Correlational Measurement

It is incredibly easy to measure correlational advertising. You just need data. However, correlational measurement confounds two distinct things: the impact of the ad itself, and the agency’s ability to put that ad in front of someone who was highly likely to buy anyway.

True incrementality means causal lift. To evaluate ad effectiveness, brands must isolate the ad’s impact by finding a valid control group. If a vendor claims to measure incrementality without establishing a strict counterfactual, they are engaging in “incrementality washing.”

4. Why Models Cannot Fix Bad Data

A major talking point in the episode revolves around a profound realization in econometrics over the last 40 years, championed by 2021 Nobel laureate Guido Imbens: An identification strategy must precede a model.

If you want to identify incremental effects, you first need a sound strategy to ensure you are comparing apples to apples (exogenous randomization, treatment vs. control groups). Once you have that clean data, you can choose any model you want – MMM, difference-in-differences, or machine learning. However, if your underlying data is correlational (apples to oranges), the most sophisticated AI model on earth cannot magically turn that orange into an apple.

5. AdSimulator: Teaching the Value of Decisions

To help his students understand these complex dynamics, Professor Wilbur created AdSimulator. In this simulation game, students manage ad budgets across 12 instruments, running MMMs and live experiments simultaneously. The game forces users to experience the visceral trade-offs of deploying limited resources. It teaches them how to spend money on costly experiments to productively learn the truth, while maximizing the daily utility of correlational models.

Quote of the Episode

“Once you have that sound identification strategy, you have a choice of many models – MMM, difference-in-differences, regressions, machine learning, or neural networks. But the model is not magic; it cannot turn an orange into an apple. The logic comparing apples to apples must precede model selection.

— Professor Kenneth Wilbur

Actionable Steps for Marketing Leaders

1. Demand an Identification Strategy: Before purchasing a new measurement tool or software, ask the vendor to explicitly explain their causal identification strategy. If they cannot explain how they establish a valid control group, they are selling correlational data.
2. Calibrate Your MMM: Do not rely on Marketing Mix Modeling in a vacuum. Use historical MMM data to generate hypotheses, but run live, geographic holdout experiments to calibrate those models with ground-truth causality.
3. Align with Finance: Share resources like Professor Wilbur’s Introduction to Advertising Measurement with your finance counterparts. Establish a shared vocabulary around what “proof of impact” actually looks like before budget planning begins.

About the Guest

Professor Kenneth Wilbur is a Professor of Marketing and Analytics at the Rady School of Management, University of California, San Diego, and a member of the Scientific Advisory Council at Lifesight. With a Ph.D. in Economics, he has spent over two decades studying television advertising markets, the adtech supply chain, and the causal measurement of ad effectiveness. He has collaborated on published research with more than ten major companies to bridge the gap between academic theory and practical marketing management.