Call to Action

Webinar: Take a tour of Sparkling Logic's SMARTS Decision Manager Register Now
Home ยป Best Practices Series: Data-Centric Approach

Best Practices Series: Data-Centric Approach


Written by: Carole-Ann BerliozPublished on: Jul 24, 2019No comments

For decades, writing rules has been an abstract exercise. Business Analysts review requirements. They write the corresponding logic. If they are lucky, there is a testing infrastructure they can push the rules to. Often, they have together code for test cases, or wait for QA to catch possible issues. There is a better way that involves bringing data in early on.

Instant Feedback

The primary objective of a data-centric approach is to provide immediate feedback. As you look at one transaction at the time, you can see what decision and intermediate decisions are made. For example, an insurance application might be too aggressively turned down due to a somewhat-poor driving record. Reviewing this result, you can fine-tune your rule right-away. After adding the proper safe-guards, that same application might end up approved with higher premium, rightfully-so. This quick turn-around is key to quality rules in your decision management projects.

When data contains expected results, this data-centric approach delivers dividends. Not only do you get immediate feedback, but the system can filter out test cases for you. By doing such, your attention goes instantly to unexpected outcomes. You can focus on fixing this mismatched transactions. Consequently, your precious time becomes more productive. When your regression tests are fully passed, QA can take over. This process should, in theory, reduce significantly the number of iterations between QA and the rules writers.

Business Performance

Better quality rules mean greater compliance with requirements, but not only. As a trained underwriter or credit officer, access to your decision logic gives you the opportunity to experiment. In absence of data, tweaking score limits will not tell you much about the improvement you achieved in doing so. Alternatively, reports based on historical data can estimate whether the change looks like an improvement for your key performance indicators (KPIs), or a step-back.

Keeping track of these KPIs, in your sandbox or in Production, will allow your organization to deploy more competitive strategies. In your sandbox, you can prepare one or more candidate strategies that show promising value. Simulation in that fashion will set expectation processing all transactions in your sandbox. Eventually, Production time will establish a verdict. When experimenting using champion / challenger, you can actually compare their true value. One fraction of the live traffic will be routed to each candidate strategy. Tracking PKIs in this setup allows you to measure how well each one of them does for its own segment.

From Data to Rules

Ultimately, data can serve you as a source for business rules. There are two fundamental approaches for that. On one hand, you can expose the data to subject matter experts, and leverage data for rules elicitation. On the other hand, you can expose the data to machine learning algorithms that can turn you tagged data into rules.

For rules elicitation, experts review transactions. They highlight for each transaction what is outstanding. In one application, the applicant is under-aged. The expected action would be to mark him or her as ineligible. In another application, an applicant with multiple DUI offenses should be declined. This knowledge turns into business rules. The more comprehensive the data set, the more ‘gold’ SMEs can mine.

For rules induction, we need to have enough data that have been marked with the trait we want to predict. For example, as fraud occurs, account holders complaint about fraudulent payment transactions. Having a mix of fraudulent and legitimate transaction, machine learning algorithms can pinpoint a set of rules that will identify fraudulent activity. Fraud experts have the opportunity to massage these rules to improve their performance.

Regardless of the path you take, writing documented rules, eliciting or inducing rules, they will all end up equally in your decision services. Data will make these rules more accurate, more powerful.

To learn more about technical details associated with data integration, take a look at Carlos’s thoughts in his technical series.


Please share your thoughts on this post:

Your email address will not be published. Required fields are marked *

 2019 SparklingLogic. All Rights Reserved.