How are models developed?

While advanced modeling tools are essential, the real power in modeling comes from the analyst. Model development is equal parts "art" and "science." The analyst must know both what to look for in the data and what will work for a particular client in a particular industry. An analyst that doesn't understand your business goals will not produce the model you need. By and large, effective model development requires five key steps.

1. Define the business problem
2. Build the development sample with relevant historical data
3. Analyze the data for predictive patterns
4. Develop the model, fine-tune it and validate it
5. Deploy the model and evaluate it in practice

One note on analyzing data. Model development involves analyzing vast amounts of data, and in operation a model will also analyze data in order to guide action. In developing a model, an analyst identifies the data relationships that are important to the problem at hand and writes the equations that codify these relationships. In operation, new data runs through the model, which applies its equations to calculate a result that drives a decision. One is batch and involves large volumes of data. The other is often interactive and typically involves much less data.