Randomized Control Trials (RCTs) are the preferred method for assessing product effectiveness at Meta. They provide a clear way to establish causality, showing that a specific treatment leads to a particular effect. However, RCTs aren't always feasible. Sometimes, products must launch to the entire eligible audience due to demand, leaving no room for control groups. In these cases, causal inference methods become important, allowing teams to evaluate effectiveness post-launch based on available data, though this approach carries limitations and requires careful consideration against previous experiments. The article outlines a framework for deciding when to use experiments or causal inference methods, emphasizing the need for data scientists to communicate their findings and limitations clearly to cross-functional teams. The analysis should influence decisions throughout the product lifecycle, and data scientists must differentiate between causation and mere association. Strong associations can provide valuable insights, even if they don’t establish direct causality. The framework involves understanding project constraints, discussing the implications of analyses, proposing methodologies, and informing stakeholders about confidence levels in the results. An example illustrated is the development of the Omni channel Ads solution. Initially, the team relied on causal inference to gauge effectiveness based on existing features due to limited adoption. As more users adopted the product, they shifted to large-scale RCTs, boosting confidence in the product’s effectiveness before launch. This structured approach allows Meta to adapt its analysis methods based on project needs and ensures that product decisions are grounded in robust evidence, leading to better outcomes for users and the business.