Causal inference

1. Randomized Controlled Trials (RCTs)

• What It Is: RCTs (like A/B testing) are the gold standard for causal inference. In this, you randomly assign customers to either a treatment group (receiving the marketing action) or a control group (no action).
• Application: Test the impact of a new email campaign by sending it to only a subset of your target list and comparing the sales against a control group. If the treatment group shows a significant uplift, you can infer that the campaign caused the increase.

2. Difference-in-Differences (DiD)

• What It Is: This approach measures the difference in outcomes (like sales) between groups before and after a marketing intervention.
• Application: Say you’re running a campaign in two regions where one receives the promotion (treatment) and the other does not (control). You can analyze how sales changed in both regions and infer causality based on the difference.

3. Instrumental Variables (IV)

• What It Is: IVs help establish causality when there's an external variable (instrument) that influences the treatment but doesn’t directly impact the outcome.
• Application: If you’re unsure whether a rise in sales is due to a marketing push or seasonal demand, use weather as an instrument. For instance, sunny days might influence outdoor events and thus indirectly drive demand for products like sunglasses, allowing you to parse out the effect of your ad campaign.

4. Regression Discontinuity Design (RDD)

• What It Is: RDD exploits cutoffs to assess causality. It compares observations around a threshold where treatment changes.
• Application: Imagine a loyalty program where customers get discounts once they spend over $100. By comparing customers who spend just above and just below $100, you can determine if the discount causes repeat purchases.

5. Propensity Score Matching (PSM)

• What It Is: PSM matches treated and control units based on their likelihood (propensity) of receiving the treatment, aiming to mimic randomization in observational studies.
• Application: If you want to assess the impact of a past ad campaign, you can use PSM to match customers who saw the ad with those who didn’t, based on similarities like demographics or previous purchases. This allows a clearer view of the ad’s effect on sales.

6. Synthetic Control Method

• What It Is: Creates a "synthetic" control group by combining data from multiple untreated units that match the treated unit before intervention.
• Application: If you launch a new product in a specific region and want to know its sales impact, you can create a synthetic control group based on similar regions without the product.

Practical Tips:

• Use Multiple Methods: Combining methods helps validate results.
• Analyze Time Windows: The timing of sales lift can indicate causality if it aligns with campaign rollout.
• Track Customer Journey: Identifying points where customers interact with marketing campaigns gives insight into potential causal impacts.

Causal inference techniques help you determine which marketing efforts actually drive sales, enabling you to focus resources on strategies with proven impact.