Miller spends considerable space on what he considers the most underestimated prerequisite for successful AI: data quality and data infrastructure. The model — the algorithm that makes predictions — typically receives the most attention. But the model is only as good as the data it trains on. Garbage in, garbage out, at scale and at speed. In practice, most AI deployments fail not because the model architecture was wrong but because the data was incomplete, biased, inconsistently formatted, or not representative of the real distribution of cases the model would encounter. For builders in emerging markets specifically: the data infrastructure problem is acute because the data that exists is often inconsistently collected, not digitised, or reflective of historical patterns that should not be perpetuated. Building good data infrastructure is the most important AI investment most organisations can make before they build a model.