What are the types of predictive analytics?
Predictive analytics is often used to mean predictive models. Increasingly, people are using the term to describe related analytic disciplines used to improve customer decisions. Since different forms of predictive analytics tackle slightly different customer decisions, they are commonly used together. The models are built using a similar methodology but different mathematical techniques.
Predictive models analyze past performance to "predict" how likely a customer is to exhibit a specific behavior in the future. For example, an attrition model measures the likelihood of a consumer to attrite or churn. This category also encompasses models that "detect" subtle data patterns to answer questions about customer behavior, such as fraud detection models. Predictive models are often embedded in operational processes and activated during live transactions. The models analyze historical and transactional data to isolate patterns: what a fraudulent transaction looks like, what a risky customer looks like, what characterizes a customer likely to switch providers. These analyses weigh the relationship between hundreds of data elements to isolate each customer’s risk or potential, which guides the action on that customer
Unlike predictive models that predict a single customer behavior (such as attrition risk), descriptive models identify many different relationships between customers or products. Descriptive models "describe" relationships in data in a way that is often used to classify customers or prospects into groups. For example, a descriptive model may categorize customers into various groups with different buying patterns. This may be useful in applying marketing strategies or determining price sensitivity.
Decision models predict the outcomes of complex decisions in much the same way predictive models predict customer behavior. This is the most advanced level of predictive analytics. By mapping the relationships between all the elements of a decision—the known data (including results of predictive models), the decision and the forecast results of the decision—decision models predict what will happen if a given action is taken.
Unlike predictive models, decision models are generally used offline to develop decision strategies. These strategies can be deployed in real time. Optimization combined with decision modeling helps produce decision strategies that determine which actions to take on every customer or transaction, in order to mathematically optimize results and meet defined constraints. Before you roll out a new offer or strategy, decision modeling also allows you to "simulate" changes to volume, response and risk—for example, "What would happen if I lower my pricing by 5%? By 10%?" You can run hundreds of these simulations within a short period, exploring many more possibilities than would be practical with live testing.
What is a model?
A model is a mathematical equation that takes in data and produces a calculation, such as a score. Think of it as a very specific set of instructions on how to analyze data in order to deliver a particular kind of result.
A predictive model could measure how likely a doctor is to prescribe a specific brand drug (behavior). By contrast, a decision model could determine the appropriate number of drug samples to send to each physician that would result in a written prescription (action). A decision model considers economic and business drivers and constraints that a predictive model would not.
What is a score?
A score is the numerical value generated by a model or group of models when applied to a specific individual or account. Scores are common outputs of many types of predictive analytics—but not all models generate scores.
What is optimization?
While the term optimization is used loosely within the business community—sometimes simply to mean making a better decision— the technical definition is much more concrete. Optimization involves using analytic techniques to mathematically calculate the best action in a specific situation, given business goals and constraints. True optimization is highly complex, able to account for uncertainty and balance competing business objectives—such as growing your customer base without overexposing yourself to risk. Its effectiveness increases when combined with predictive and decision models.