Correlation versus causation is the difference between two things moving together and one actually making the other happen. In marketing, a channel can sit right next to your sales in every report without causing many of them, and budgets get wasted when the two are confused.
How correlation and causation differ
A correlation is a pattern: when one number goes up, another tends to go up (or down) too. Causation means changing the first number changes the second. Patterns appear without causation for a few common reasons.
- A third factor drives both. A garden centre spends more on ads in April and sells more in April. Spring causes both. The ads may add something, but the chart cannot tell you how much. Seasonality is the most common hidden factor in marketing data.
- The direction is reversed. Retargeting ads go to people who have already visited, and people who have already visited were more likely to buy anyway. The ad follows the intent rather than creating it.
- Selection. Customers who open every email spend more, but they were your keenest customers before the emails began.
- Chance. With small numbers, two unrelated lines will often move together for a few weeks.
The way to show causation is an experiment where you control who sees what. An A/B test randomly splits visitors between two versions of a page. A holdout test keeps a random group of customers out of a campaign. A geographic test runs ads in some matched areas of the UK and not others, then compares sales.
Why it matters
Most marketing reports show correlation. Attribution in ad platforms gives credit to ads that appeared before a sale, which is not the same as proving the ad caused it. That is why platform-reported conversions from brand search ads and retargeting so often look excellent: they sit right beside people who were already on their way to buy. A business that moves budget towards whatever correlates best with sales tends to pour money into the end of the journey and starve the work that created the demand.
The same trap catches website changes. A new site launches in March, enquiries rise in March, and the redesign gets the credit, when March is simply busier than February for that business every year.
Common mistakes
- Before-and-after comparisons with nothing to compare against. Always ask what else changed in the same period: prices, season, competitors, press coverage.
- Reading statistical significance as proof of cause. Statistical significance says a pattern is unlikely to be chance. Only a controlled experiment says what caused it.
- Stopping tests early because the first week looked good.
- Letting the platform mark its own homework. Each ad platform’s own reports are designed to show its value.
How to act on it
Before acting on any trend, write down the other explanations and check them: same period last year, other channels, price changes. Compare against a control wherever you can, even a rough one such as a region where nothing changed.
For decisions with real money behind them, test for incrementality. Pause a campaign in a few matched postcode areas for four to six weeks and compare sales with the areas where it kept running. Decide before the test what result would change your mind, so you are not tempted to explain away an awkward answer. Designing tests like these into a paid media plan is part of how I run performance marketing.