1.1 What Is Marketing Science?
Context
In marketing, the gap between data science work and business value tends to show up in patterns like these:
- A coupon campaign is declared a success: recipients spent 20% more than non-recipients. Both the business and the data scientist agree the campaign worked — until it turns out the coupons were sent to customers who were already most likely to buy. The “lift” was selection, not effect.
- A data scientist delivers a fresh customer segmentation. Stakeholders nod, call it interesting, and keep using the segmentation they already had. The work never reaches a campaign.
- A demand forecast hits 2% MAPE at the brand level. But the business plans inventory at the SKU-by-region level, where the forecast is much noisier — leading to stockouts in some regions and overstock in others.
All three are different versions of the same problem. In the first, the analysis gave a clear number but missed the real cause. In the second, the model never reached the people who could use it. In the third, the analysis was done at a different level of detail than the business decision it was meant to support.
Definition
That gap is the starting point for the definition we use in this book:
Marketing Science is the applied discipline of using data, statistics, and machine learning to improve business decision-making in marketing.
The work is not done when the model is built. It is done when the analysis actually changes a decision, leads to a new action, and moves a business result.
In general data science, success is often measured by prediction accuracy. Marketing Science still cares about accuracy, but accuracy is a means to an end: the real measure of success is the business result the analysis enables.
Workflow
Figure 1 shows the central idea of this book: analysis creates value only when it changes a decision, an action, and ultimately a business result.
How Marketing Science Differs by Industry
The definition applies across industries, but how you do the work — what data you have, which methods you use, how easy it is to experiment — varies a lot.
| Industry | Data Examples | Data Characteristics | Key Methods and Themes | Ease of Experimentation |
|---|---|---|---|---|
| CPG (Consumer Packaged Goods) | POS (via retailers), panel data (NielsenIQ, Circana), media spend data, retail media reports | Companies often do not own purchase data directly (obtained via retailers or panel providers). Mostly aggregate-level | MMM, price elasticity analysis, planogram optimization, trade promotion analysis | Low (hard to control because distribution is intermediated) |
| Retail (Offline) | POS, loyalty program data (Kroger Plus, Target Circle, etc.), promotion and planogram data | Owns individual-level purchase history. Can analyze at SKU x store x day granularity. The rise of retail media networks is also expanding data use for advertisers | Promotion optimization, assortment optimization, customer segmentation, retail media ROI analysis | Medium (store-level and region-level experiments are possible) |
| E-commerce | Clicks, page views, cart additions, purchases, media spend data | Rich behavioral logs. Full-funnel tracking is possible end to end | A/B testing, funnel analysis, recommendations, CLV prediction | High (randomization is easy online) |
| Consumer App | Downloads, in-app behavior, purchases, media spend data | Retention and engagement are the key metrics. Freemium models are common | Retention analysis, ROAS optimization, cohort analysis | High (in-app A/B testing is easy) |
| SaaS | Product usage logs, contract and billing data, support tickets | Detailed usage logs are readily available. Easy to link with contract and billing data | Churn prediction, PLG analysis, product analytics | High (feature-level A/B testing is routine) |
You do not need to memorize the table. What matters is seeing that the same marketing question can call for very different methods, depending on who owns the data, how granular it is, and how easily the business can test ideas.
For example, MMM is a well-established method in CPG with decades of history, but it is rarely used in SaaS. Conversely, A/B testing is an everyday decision-making tool in app and e-commerce businesses, but running a rigorous A/B test in CPG is difficult.
In the U.S. market, the rapid growth of retail media networks (Amazon Ads, Walmart Connect, Instacart Ads, etc.) is blurring industry boundaries. CPG brands can now access near-user-level data through retail media. At the same time, this has created a new challenge for causal inference: “Is the retail media effect incremental, or are we just serving ads to customers who would have bought organically?”
This book focuses on concepts that work across industries, while touching on industry-specific considerations where relevant.