Whether it's in the newspapers, magazines, airport media, or in barroom banter, Big Data has become a hot topic. Everyone is talking about this trendy topic, but so far only very few organizations are actually using this technology successfully! A major reason for this is the lack of insight into the key success factors for organizations to build actionable Big Data analytics models. Combining years of experience working with a number of global companies, we believe that in order to be successful, a Big Data analytics model needs to fulfill several requirements: (1) Business relevance. Business relevance is the first key requirement for an analytic model. Analytic models must be able to solve a specific business problem. Models that have superior performance, but do not solve the business problem are meaningless. Obviously, a thorough understanding of the business context and business problem is essential before model development. For example, in an insurance fraud detection problem, there must be a clear definition of how fraud will be defined, measured, and managed at the outset. (2) Statistical performance. Another important key factor affecting model success is model performance performance. In other words, the analytic model should significantly improve predictive or descriptive performance from a statistical significance perspective. Different types of performance evaluation metrics are often used depending on the type of problem being analyzed. In customer segmentation, the statistical evaluation metrics mainly evaluate the similarity within the comparison clusters versus the difference between the clusters; in customer churn prediction, the main evaluation is whether the model assigns higher scores to potentially churned customers. (3) Interpretability and rationality. Explanatory means that the analytical model is easily understood by decision makers, and rationality means that the model is consistent with experts' expectations and business knowledge. Both interpretability and reasonableness are subjective judgments that depend on the knowledge and experience of the decision maker. There is often a contradiction between these two factors and statistical performance analysis, e.g., complex neural networks and random forest models have better predictive performance but poorer explanatory power. Therefore, decision makers need to find a balance between the two. In application scenarios such as credit risk analysis, interpretability and rationality are very important factors, while in fraud detection and marketing response modeling, this factor is not so important. (4) Operational efficiency. Operational efficiency relates to the time required to be invested in the process of model evaluation, monitoring, testing and reconstruction. From this factor, it is clear that neural networks or random forests are less efficient, while regression models and decision trees, for example, are more efficient. In business scenarios such as credit card fraud detection, operational efficiency is very important because all decisions must be made within seconds of the start of a credit card transaction. (5) Economic Cost. Economic cost is the cost invested in the process of collecting the data needed for the model, running the model, and analyzing the model results, in addition to the cost of bringing in external data and models. All of these costs must be taken into account when analyzing the economic returns of the model and are usually not something that can be simply and directly calculated. (6) Compliance. Compliance is becoming increasingly important in many industries. Compliance refers to the extent to which a model adheres to existing systems and laws. In the area of credit risk, it is particularly important that analytical models comply with Basel II and III. In the insurance industry, models must comply with the EU Solvency II agreement. To summarize the above, we have briefly discussed the key factors for successfully building a data analytics model. As we have noted, the importance of each factor depends on the scenario in which the model is applied.