Causality refers to the relationship between cause and effect, where one event (the cause) directly produces or influences another event (the effect). Causality is an important concept in measuring marketing effectiveness: are ;marketing activities actually driving business outcomes, or are they simply correlated with them?
Understanding Causality vs. Correlation
The fundamental distinction between causality and correlation is critical. Correlation indicates that two variables move together, but causal relationship means that changes in one variable directly cause changes in another. The famous phrase "correlation does not imply causation" is particularly relevant in marketing, where many variables can appear related without having a true causal relationship.
As an example, branded search spend is typically highly correlated with sales. In the dataset below, correlation between sales and branded search is 62%. Branded search can be an effective driver of sales, but it is very unlikely to be as effective as the high correlation would imply. High correlation results from the fact that branded search is a demand capture channel that is influenced by other media, especially awareness media like Meta Awareness campaigns or TV. When a business in conducting awareness campaigns, more people are conducting searches for the brand, leading to increased branded search spend, unless it is capped. Takeaway: Understanding the true causal relationship between branded search and sales requires more advanced methods than looking at correlation.
General Methods for Establishing Causality
There are various methods for establishing causality, some of them providing stronger proof for causality than others.
Randomized Controlled Trials (RCTs)
The gold standard for proving causality, RCTs randomly assign subjects to treatment and control groups. They have broadly been used for example in clinical drug trials.
Natural Experiments
These leverage naturally occurring variations in treatment assignment to infer causality. For example, analyzing performance differences between regions where a campaign ran versus where it didn't due to budget constraints.
Quasi-Experimental Methods
Including techniques like difference-in-differences (DiD), regression discontinuity, and instrumental variables (IV), these methods help establish causality when true randomization isn't possible. "Causal inference in economics and marketing" provides an overview for In methods in marketing context.
Statistical Causal Inference
Advanced statistical techniques like propensity score matching, synthetic control methods, and causal machine learning help identify causal relationships from observational data.
Importance of Causality in Marketing Measurement
Establishing causal relationships between marketing activities and business outcomes (typically revenue) is the foundation for robust media measurement. And robust measurement is required for making informed decisions about budget allocation and campaign optimization. Without establishing proper causal links between revenue and media, marketers risk:
- Misattributing success to ineffective channels
- Continuing investments in activities that don't drive results
- Missing opportunities to scale truly effective initiatives
- Making decisions based on correlations that could be random or systematically misleading
Last-Click attribution is one example of a widely used media measurement approach that is not based on causal principles. Last-Click attribution assumes that the last clicked ad before conversion was driving the conversion, ignoring other advertising that happened earlier in the buying journey. A Sellforte study investigating Last-Click attribution found that Last-Click-reported effectiveness (ROAS) for campaigns was not correlated with effectiveness estimated by Sellforte Marketing Mix Model (MMM), which operates on the principles of causality and is built for estimating the true incremental sales driven by each campaign and channel.
Below is a summary of the study, showing a comparison between Last-Click reported ROAS and MMM-reported ROI, and highlighting the areas where the two methods differ the most.
Challenges in Establishing Causality in Marketing
While causality is critical in marketing measurement, there are factors that make establishing causality in marketing particularly challenging.
Confounding Variables
External factors like seasonality, competitive actions, or economic conditions can influence both marketing activities and outcomes, making it difficult to isolate true causal effects.
Selection Bias
When treatment assignment isn't random, systematic differences between groups can create apparent causal relationships that don't actually exist.
Measurement Windows
The time lag between marketing exposure and conversion can make it challenging to establish clear causal links, especially for longer customer journeys.
Complex Buying Journeys
In complex customer journeys with multiple touchpoints, determining which interactions truly caused the conversion becomes increasingly difficult.
Dynamic Effects
Effectiveness of marketing is not static. As an example, effectiveness tends to decrease the more you invest in a channel, because the channel starts to saturate.
Low Signal-to-Noise ratio, Low-Granularity Data
Marketing datasets tend to have random variation, which can be considerable compared to the effects that individual marketing investment can have on sales of a company. Combining this with a time-series dataset that could in the worst-case be on weekly level, can make estimation of marketing effectiveness of individual activities challenging.
Causal Marketing Measurement in Practice
Understanding causality in theory is just the beginning. Modern marketers have access to various tools and methodologies that enable practical implementation of causal measurement. Here are approaches used by many organizations.
1. Ad Platforms' Conversion Lift Tests
Ad Platforms' lift tests focus on estimating incrementality of one platform. Below are three examples.
Meta Conversion Lift Test: Meta's built-in incrementality testing tool randomly assigns users to test and control groups, measuring the true incremental impact of Facebook and Instagram campaigns. The platform automatically excludes the control group from seeing ads while tracking conversions across both groups.
Google Conversion Lift: Google's Campaign Experiments allow marketers to test campaign changes on a portion of auction traffic, measuring incremental performance versus a control group. This includes Search Lift studies and YouTube Brand Lift surveys.
TikTok Conversion Lift: TikTok's incrementality testing solution measures the true impact of TikTok campaigns by comparing exposed versus unexposed user groups, accounting for organic lift and view-through conversions.
While on the surface Conversion Lift Tests resemble RCTs, they have been critiqued for divergent delivery, meaning that audience is not selected randomly: the algorithm learns to show the test ads to people likely to convert, whereas the blank space is shown to random audience. However, conversion lift tests are still useful tools for causality and incrementality when used and interpreted properly. They can provide a snapshot of a platform's effectiveness, which can be used in validating and calibrating Marketing Mix Models.
Case study: Validating Marketing Mix Model with Meta Conversion Lift Studies.
Geo Testing (Geographic Experiments)
Geo testing involves selecting similar geographic markets and randomly assigning them to test and control conditions. For example, one could divide U.S. states or U.S. Designated Market Areas (DMAs), into test and control areas. The idea is to then apply a treatment to the test group, and compare the groups against each other, to make conclusions about incrementality of a marketing activity.
While conversion lift tests typically produce standardized outputs that include estimated value and its confidence intervals, geo tests can be more challenging to analyze, as the method lacks a similar level of standardization. Sellforte's Geolift module helps analyzing geo tests.
Similar to Ad Platforms' Conversion Lift tests, Geo tests only provide a snapshot into incrementality: they don't provide continuous measurement, saturation curves and other important outputs required for marketing optimization. However, they can still be useful in getting a snapshot of a channel's incrementality, which can be used in Marketing Mix Model calibration.
3. Causal Marketing Mix Models (MMM)
Causal Marketing Mix Models, provided for example by Sellforte, incorporate causal inference principles to provide more accurate measurement. Using methods like Bayesian Marketing Mix Modeling, Model calibration with causal experiments, and Causal Graphs, Causal MMMs provide granular and robust view into media effectiveness.
Advanced Concepts
Causal Graphs and DAGs
Directed Acyclic Graphs (DAGs) help visualize and analyze complex causal relationships in marketing systems, identifying potential confounders and mediating variables.
Causal Machine Learning
Emerging techniques that combine machine learning with causal inference principles to identify treatment effects and optimize marketing strategies.
Industry Best Practices
Leading marketing organizations establish causality through:
- Investments into advanced measurement tools and expertise
- Regular experimentation routine
- Continuous validation of measurement assumptions and methodologies
Conclusion
Understanding and establishing causality is fundamental to effective marketing measurement. While correlation can suggest relationships worth investigating, only causal analysis can provide the confidence needed to make marketing decisions. As the marketing landscape becomes increasingly complex, the ability to distinguish between correlation and causation becomes ever more critical for driving growth.
By investing in causal measurement capabilities, marketing teams can move beyond vanity metrics to focus on activities that truly drive business results, ultimately improving return on investment driving growth.