How can you plan a store in a town where no comparable store has ever existed? You cannot observe its sales yet. You can, however, create data about what might happen. Synthetic data represents conditions or outcomes that have not been directly observed. It is generated from assumptions about the world, often informed by analysis of real data. section 7.2.4
Consider a retailer weighing new stores across Canada, the United States, and Mexico. Analysts gather information about the proposed towns, existing competitors’ stores, and the retailer’s stores elsewhere. They use those observations to construct figures for possible store sizes, product mixes, prices, sales, returns, and staffing costs. Those figures can help compare locations and decide where to build first. They are not records of stores that already exist. section 7.2.4
Synthetic data was once assembled by hand, then with rules-based systems; artificial intelligence can now generate large amounts that reproduce patterns such as seasonality. Simulation takes the next step: change a variable, rerun a scenario, and compare possible outcomes. A retailer might test different numbers, locations, and mixes of stores against its objectives. The results of those runs are simulated data, rather than observations of what actually happened. section 7.2.4 section 7.2.5
The distinction matters when someone asks you to trust a dataset. Raw data records activity; synthetic and simulated data extend what analysts can explore beyond direct observation. Ask what observations and assumptions informed generated figures, and whether a claimed result is an observation or a possible outcome. Manufactured data can also be used to support a false claim, so its origin and purpose matter as much as its apparent realism. section 7.7 section 6.4
