A Case for the Data-Centric Approach
The data-centric approach to rules authoring is revolutionizing decision management. For decades, writing rules has been an abstract exercise. Typically, business analysts gather rules requirements and write the corresponding business rules. IT then takes over to translate rules into executable code. Testing, if any, is often handled by a QA team. It is only then when data is introduced. Even so, usually the test cases are more technical focused. This traditional approach may be effective in ensuring that the technical implementation aligns with the original logic. However, it provides no mechanism for ensuring that the original logic is the appropriate strategy to move forward with.
On the other hand, the data-centric approach incorporates data through out the development process. Rather that authoring business rules in a vacuum, business analysts write rules in context of test cases. These test cases can be a sample of historical records or fictitious records that cover different scenarios. If you’re using fictitious records, create them with your DBA team so that the records follow the correct form structure. Include expected decision outcomes in the test cases.
For example, if you’re working on a fraud detection project, you should know ahead of time which records should be flagged as fraudulent. Once you have your test cases, you can begin writing rules, referencing the data in one case and then testing the rules on all cases. Through this approach, businesses can not only produce better quality rules but also speed up the entire development process. You can also get a more technical perspective on data integration from Carlos.
Instant Feedback
One of the main benefits of a data-centric approach is providing immediate feedback. As you “run” the rules on each of the test cases, you can see whether the decision outcomes match your expectations. Then you can easily drill down on the cases where there’s a mismatch and do any necessary fine-tuning. This approach helps you quickly catch any errors in your logic or weaknesses in your strategy.
For example, for an auto insurance project, you might have a test case where the applicant has a less-than-perfect driving record but should still be approved for insurance at a higher premium. However, based on the rules you have written, the applicant is declined. You review your rules and find a typo in one of them: instead of 5 accidents, you wrote 4. You fix that error and run the rules on that test case again. The applicant is still declined. You review your rules again and find that they all align with the business requirements. At this point, your logic is “correct,” but the strategy is wrong. It’s time to set up a meeting with the product manager. By going through short cycles of writing rules and regression testing, you will be efficiently produce higher quality rules and make QA testing a breeze.
Transform Test Cases to Rules
There are three fundamental approaches for transforming data into rules: writing documented rules, rules elicitation, and rules induction. In practice, a project can use multiple approaches.
- Writing documented rules is the traditional approach I discussed in the beginning. Business analysts translate business requirements into rules (within or without the context of data). This is a more a top-down approach.
- Rules elicitation, on the other hand, is more a bottom-up approach. Subject matter experts apply their expert knowledge to data to write rules. You can think of it as combining requirements gathering with rules writing.
- Finally, rules induction involves exposing data to machine learning algorithms to “write” rules (usually in the form of a predictive or prescriptive model). In order for rules induction to be effective, you need to have a lot of data, enough for the algorithms to distinguish between different groups. For example, going back to the fraud detection project, you would need a mix of fraudulent and legitimate transactions that the algorithms can sift through to pinpoint which combination of behaviors and data points will likely indicate fraud.
Regardless which approach you use in your project, the more diverse test cases you have, the better your business rules will be. However, as the “old” computer science adage goes, garbage in, garbage out. Bad data will produce bad rules, and why data validation should also be part of your rules authoring and decision management strategy.
Experiment and Measure Business Performance
Having test cases on-hand also enables you to experiment in your sandbox environment. You can feed your test cases through different sets of rules based on different strategies to compare which strategy is the best. This ensures that you’re going into production with your best foot forward. Of course, in order to do effective business testing, you need to have first identified your objectives and KPIs and have a way to measure them.
This is why our clients are using Sparkling Logic’s SMARTS™ for decision management. SMARTS™ enables business analysts and other users to easily upload test cases, author rules from the data, run validation, regression, and QA tests, conduct experiments, and measure performance all in one place, through an intuitive user interface.
Continue reading my Rules Authoring Best Practices series:

