Predictive Modeling Examples
Predictive models can be used to improve decision making throughout an organization’s daily operations and throughout their customer’s journey. Examples of predictive modeling applications include (but are not limited to):
- Customer segmentation
- Targeted messaging, advertising, and promotion
- Fraud detection and prevention
- Risk assessment
- Cross-selling and up-selling
- Forecasting and scheduling (in manufacturing)
- Contact center prioritization
- Churn mitigation
Explicit Decision Automation
In the early days of decision automation, the focus was on automating explicit business rules and decision logic. For example, in insurance, much of the policy rules such as eligibility criteria are driven by government regulations. Insurance firms leveraged business rules and decision management technologies like SMARTS™ Decision Manager to not only automate policy rules but also better manage those rules. These technologies separate the policy rules out from the systems and applications that use them. As a result, business analysts could easily review and make changes to policy rules with minimal IT involvement and avoid a full software development cycle. In other words, decision management technologies enabled firms to become more agile.
From Expert Knowledge to Predictive Modeling
Adapting eligibility criteria to new income requirements for a state is fairly straightforward. However, there are cases when decision making is not so black and white. For example, in consumer lending, government regulations may put restrictions on how much a consumer can borrow and at what cost, but it’s ultimately up to the financial institution to decide whether or not to approve an applicant and what to offer. While seasoned organizations may be able to easily identify qualities of their best and worst customers, assessing credit risk for everyone else is not so simple. In addition, other factors such as likelihood of acceptance must also be considered when making a credit decision.
This is where decision making and decision automation can benefit from predictive modeling. Predictive models apply sophisticated analytic techniques such as machine learning on historical data to make predictions about the future. Predictive modeling identifies patterns that human experts cannot. As a result, a financial institution could use predictive models to determine how likely a particular applicant is fraudulent, would accept a given offer, and would repay. Equipped with these probabilities, the financial institution would be able to make more informed credit decisions.
Deploying Predictive Models in Automated Decisions
While predictive models can provide probabilities (and/or scores), organizations still need to decide how to act on them. Modern decision management platforms like SMARTS™ enable firms to combine predictive models with explicit business rules to transform insights into action. In fact, SMARTS™ has over 17 predictive model engines to ensure model-driven decisions execute at high performance. Returning to the personal lending example, the financial institution could set up their decision logic to reject applicants with a high probability of being fraudulent and offer the approved applicants the “best offer” (the offer with highest probability that the applicant would accept and eventually repay).
Predictive Models Depend on Good Data
With predictive modeling, garbage in, garbage out is especially true. The accuracy and completeness of historical data will affect the predictive power of the model. In addition, predictive models assume that the past is a good predictor of the future. Since times are always changing, historical data can become stale fast, and therefore predictive models should be retrained on “new” historical data on a regular basis. This is why SMARTS™ provides the tools to not only deploy predictive models but also support closed loop, continuous learning of those models.
Learn more about Sparkling Logic’s SMARTS™ AI & ModelOps capabilities.