Leveraging Causal Graphs for Marketing Optimization

Traditional methods of measuring marketing effectiveness often rely on correlation, but correlation does not imply causation. Enter causal graphs, a powerful tool that helps marketers understand not just what is happening but why it is happening. In this post we will delve into how causal graphs can be leveraged for marketing optimization.

A Practical Example

This is what confusing correlation with causation looks like in real life and if you are a marketer you will have came across some version of this in the past.

Let's assume you are the owner of an online ice-cream delivery store. Many search-based advertisers offer the "opportunity" to you or competitors to advertise on your own brand name results. So your potential customers would go to a search engine, would type the name of your online store and then the top result would be an ad that is up for grabs by you or your competitors.

Let's add to the above scenario another very realistic assumption. You are using a last-click attribution model. What that means is that the credit for you new customers goes to the last source/advertiser that brought the customer at your doorstep.

The above setup is a variation of Prisoner's Dilemma.

• If you bid for the ad space, but your competitors don't, you have nothing to win as the traffic was already there for your product

• If your competitors bid for the ad space, but you don't, then your competitors can gain some of your traffic

• If both you and your competitor bid for the ad space, you get to keep traffic that would be yours anyway if the ad space wasn't there, your competitor will get some of that traffic (usually the low intent / low ltv share) but at extremely high cost and the search engine stands to profit the most.

Many times what ends up happening is you paying for traffic that would have been yours anyway and correlation tricks you into believing that this is actually a brilliant idea. Let's see why.

Let's say you paid $10k for the ad space. Your last-click attribution model gets these new (really organic) acquisitions and renames them into paid acquisitions. So if 10k customers clicked the ad instead of the organic result you end up observing:

1. Your bidding and acquisitions go up and down at the same time

2. The cost per acquisition is (suspiciously) low (1$/acquisition) but you are the champ right?

Now let's see why the above is misleading. When something causes a spike in your traffic, some of that traffic will visit your store through the paid ad (and so you get charged) and some will visit through the organic result. So you conclude that more ad spending leads to more traffic. What you are missing here is that the exogenous event caused both the spike in paid traffic and your spending to go up. If you wanted to measure the true causal effect you would have to play around with your bid and then you would observe that when you increase your bidding the only thing that happens is more or less organic traffic getting "renamed" to paid traffic while the total traffic oscillates around the same level.

Building your Causal Graph

The graph in the image above is a causal graph. Causal graphs, also known as causal diagrams or directed acyclic graphs (DAGs), are visual representations of cause-and-effect relationships between variables. Unlike correlation-based models, causal graphs aim to identify and illustrate the direct and indirect influences of different factors on each other. Using the Causalysis platform you can easily design causal graphs for your application of choice and state of the art algorithms will do the analysis for you. Let's see what are the steps involved to build such a causal graph with Causalysis.

1. Identify the short term and long term objective: In the example above, the short term objective would be to bring more customers and the long term objective would be the revenue generated by these customers over their lifetime.

2. Identify key variables: What are the variables that can have an impact on the outcome? This could be things within your control like marketing spend, activities from other teams within the business e.g. website traffic or things outside your control e.g. economic conditions.

3. Collect Data: Gather data on any potential variable you believe could have an impact on the application you wish to analyze. Ensure the data is comprehensive and covers all necessary aspects to test the causal relationships effectively e.g pick time periods with sufficient variation on key variables and outcomes.

4. Construct the causal graph: Develop a causal graph that maps out the hypothesized relationships between the variables. This involves determining which variables directly influence others and drawing directed edges (arrows) to represent these causal links.

5. Validate the Model: Causalysis uses statistical methods to validate the relationships in the causal graph. So even if some relationship you assumed are not really there, the algorithm will find it and support you in adjusting your causal graph.

6. Analysis: You can then click the button and let Causalysis take care of the rest. At the end you will get back the estimated model with the treatment effect of the variable of your choice.

Other Marketing Applications

Attribution Modeling

Attribution modeling is essential for understanding the contribution of different marketing channels to conversions. Traditional models, like last-click attribution, can be misleading. By using causal graphs, marketers can develop more accurate attribution models that account for the true causal impact of each touchpoint.

Budget Allocation

Causal graphs help marketers understand how changes in budget allocation across different channels affect overall performance. For instance, increasing the budget for social media campaigns might have a higher causal impact on sales than investing the same amount in display advertising.

Campaign Effectiveness

By analyzing causal relationships, marketers can pinpoint which elements of a campaign are most effective. This enables them to refine their messaging, creative assets, and targeting strategies to maximize impact.

Conclusion

Causal graphs provide a robust framework for understanding and optimizing marketing efforts. By focusing on causality rather than mere correlation, marketers can make more informed decisions, allocate resources more effectively, and ultimately drive better business outcomes. As data availability and analytical tools continue to improve, leveraging causal graphs will become an increasingly vital component of sophisticated marketing strategies.

Harnessing the power of causal analysis can transform marketing from a game of guesswork to a science of precision, unlocking new levels of efficiency and effectiveness.