Decision Services

Decision services are specialized applications that automate business decisions using a set of predefined rules. In this chapter, we’ll walk through the process of implementing a complete decision service using Drools Rule Language (DRL), from requirements gathering to deployment.

Understanding Decision Services

Before diving into implementation, let’s understand what decision services are and their key characteristics:

What is a Decision Service?

A decision service is a standalone component that:

Accepts input data relevant to a business decision
Processes this data through business rules
Returns a decision or recommendation
Operates independently of other application components
Can be called synchronously or asynchronously

Common Decision Service Use Cases

Decision services excel in scenarios like:

Loan or credit approval: Determining loan eligibility, interest rates, and terms
Insurance underwriting: Calculating premiums, assessing risks, determining coverage
Regulatory compliance: Checking transactions against compliance rules
Pricing optimization: Setting dynamic prices based on market conditions and customer data
Customer segmentation: Categorizing customers for targeted marketing
Fraud detection: Identifying suspicious patterns in transactions
Tax calculation: Computing complex tax scenarios

Building a Loan Approval Decision Service

Let’s walk through designing and implementing a loan approval decision service step by step.

Step 1: Gather Requirements

First, we need to understand the business requirements for our loan approval process:

Functional Requirements

The system must evaluate loan applications based on credit score, income, loan amount, and purpose
Applications must be automatically approved, rejected, or flagged for manual review
The system must calculate a risk score for each application
The system must determine the appropriate interest rate for approved loans
The system must provide explanations for decisions

Business Rules (Simplified Example)

Applications with credit scores below 550 are automatically rejected
Applications with credit scores above 750 are automatically approved
The debt-to-income ratio must not exceed 43% for approval
The loan amount must not exceed 5 times the annual income
Interest rates range from 3.5% to 12.5% based on credit score and risk assessment
Applications that are not automatically approved or rejected require manual review

Step 2: Design the Domain Model

Based on the requirements, we’ll design our fact model:

package ai.aletyx.examples.loan;

public class LoanApplication {
    private String id;
    private String applicantName;
    private int creditScore;
    private double annualIncome;
    private double monthlyDebt;
    private double requestedAmount;
    private int requestedTermInMonths;
    private String purpose;

    // Decision fields - will be set by rules
    private String status; // "PENDING", "APPROVED", "REJECTED", "REVIEW"
    private String decisionReason;
    private double riskScore;
    private double approvedInterestRate;
    private double approvedAmount;
    private int approvedTermInMonths;
    private double calculatedMonthlyPayment;

    // Getters, setters, and calculated properties
    public double getDebtToIncomeRatio() {
        return (monthlyDebt * 12) / annualIncome;
    }

    public double getLoanToIncomeRatio() {
        return requestedAmount / annualIncome;
    }

    // ... other getters and setters
}

Step 3: Create the Rule Unit

Now we’ll create the rule unit to define the data sources:

package ai.aletyx.examples.loan;

import org.drools.ruleunits.api.RuleUnitData;
import org.drools.ruleunits.api.DataStore;
import org.drools.ruleunits.api.DataSource;

public class LoanDecisionUnit implements RuleUnitData {
    private final DataStore<LoanApplication> applications;

    public LoanDecisionUnit() {
        this.applications = DataSource.createStore();
    }

    public DataStore<LoanApplication> getApplications() {
        return applications;
    }
}

Step 4: Write the Rules

Now we’ll implement our business rules in DRL:

package ai.aletyx.examples.loan
unit LoanDecisionUnit;

import ai.aletyx.examples.loan.LoanApplication;
import org.drools.ruleunits.api.RuleUnitData;
import org.drools.ruleunits.api.DataStore;

declare LoanDecisionUnit extends RuleUnitData
  applications: DataStore<LoanApplication>
end

// Rule 1: Calculate Risk Score
rule "Calculate Loan Risk Score"
  salience 100 // Run early in the sequence
  when
    $application: /applications[ status == "PENDING" ]
  then
    // Complex risk calculation algorithm
    double creditFactor = 100 - ($application.getCreditScore() / 8.0);
    double dtiRiskFactor = $application.getDebtToIncomeRatio() * 100;
    double loanToIncomeFactor = $application.getLoanToIncomeRatio() * 20;

    // Adjust for loan purpose
    double purposeFactor = 0;
    if ("DEBT_CONSOLIDATION".equals($application.getPurpose())) {
        purposeFactor = 10;
    } else if ("HOME_IMPROVEMENT".equals($application.getPurpose())) {
        purposeFactor = 5;
    } else if ("EDUCATION".equals($application.getPurpose())) {
        purposeFactor = 15;
    } else if ("BUSINESS".equals($application.getPurpose())) {
        purposeFactor = 20;
    }

    // Calculate final risk score (0-100)
    double riskScore = creditFactor + dtiRiskFactor + loanToIncomeFactor + purposeFactor;
    riskScore = Math.max(0, Math.min(100, riskScore)); // Clamp between 0-100

    // Set the risk score
    $application.setRiskScore(riskScore);
    applications.update($application);
end

// Rule 2: Auto-Reject Poor Credit
rule "Reject Application with Poor Credit"
  salience 90
  when
    $application: /applications[ status == "PENDING", creditScore < 550 ]
  then
    $application.setStatus("REJECTED");
    $application.setDecisionReason("Credit score below minimum threshold");
    applications.update($application);
end

// Rule 3: Reject High Debt-to-Income Ratio
rule "Reject Application with High Debt-to-Income Ratio"
  salience 90
  when
    $application: /applications[ status == "PENDING", debtToIncomeRatio > 0.43 ]
  then
    $application.setStatus("REJECTED");
    $application.setDecisionReason("Debt-to-income ratio exceeds maximum threshold");
    applications.update($application);
end

// Rule 4: Reject Excessive Loan Amount
rule "Reject Application with Excessive Loan Amount"
  salience 90
  when
    $application: /applications[ status == "PENDING", loanToIncomeRatio > 5 ]
  then
    $application.setStatus("REJECTED");
    $application.setDecisionReason("Requested loan amount exceeds 5x annual income");
    applications.update($application);
end

// Rule 5: Auto-Approve Excellent Credit with Low Risk
rule "Approve Application with Excellent Credit and Low Risk"
  salience 80
  when
    $application: /applications[
      status == "PENDING",
      creditScore >= 750,
      riskScore < 40
    ]
  then
    $application.setStatus("APPROVED");
    $application.setDecisionReason("Excellent credit profile with low risk assessment");
    $application.setApprovedAmount($application.getRequestedAmount());
    $application.setApprovedTermInMonths($application.getRequestedTermInMonths());
    applications.update($application);
end

// Rule 6: Set Interest Rate Based on Risk Score
rule "Set Interest Rate for Approved Loans"
  salience 70
  when
    $application: /applications[ status == "APPROVED", approvedInterestRate == 0 ]
  then
    // Base interest rate calculation
    double baseRate = 3.5; // Minimum rate
    double riskPremium = ($application.getRiskScore() / 10.0);
    double creditAdjustment = (800 - $application.getCreditScore()) / 40.0;

    // Calculate final rate (capped between 3.5% and 12.5%)
    double interestRate = baseRate + riskPremium + creditAdjustment;
    interestRate = Math.max(3.5, Math.min(12.5, interestRate));

    // Round to nearest 0.125%
    interestRate = Math.round(interestRate * 8) / 8.0;

    // Set the approved interest rate
    $application.setApprovedInterestRate(interestRate);
    applications.update($application);
end

// Rule 7: Calculate Monthly Payment for Approved Loans
rule "Calculate Monthly Payment for Approved Loans"
  salience 60
  when
    $application: /applications[
      status == "APPROVED",
      approvedInterestRate > 0,
      calculatedMonthlyPayment == 0
    ]
  then
    // Monthly interest rate (annual rate divided by 12)
    double monthlyRate = $application.getApprovedInterestRate() / 100 / 12;

    // Number of payments
    int payments = $application.getApprovedTermInMonths();

    // Loan amount
    double principal = $application.getApprovedAmount();

    // Monthly payment calculation using amortization formula
    double monthlyPayment = 0;
    if (monthlyRate > 0) {
        monthlyPayment = principal * monthlyRate *
                        Math.pow(1 + monthlyRate, payments) /
                        (Math.pow(1 + monthlyRate, payments) - 1);
    } else {
        // Simple division if interest rate is zero
        monthlyPayment = principal / payments;
    }

    // Round to nearest dollar
    monthlyPayment = Math.ceil(monthlyPayment);

    // Set the monthly payment
    $application.setCalculatedMonthlyPayment(monthlyPayment);
    applications.update($application);
end

// Rule 8: Route Remaining Applications to Manual Review
rule "Route to Manual Review"
  salience 50
  when
    $application: /applications[ status == "PENDING" ]
  then
    $application.setStatus("REVIEW");
    $application.setDecisionReason("Application requires manual underwriter review");
    applications.update($application);
end

Step 5: Implement the Service Layer

Next, we’ll create a service class to expose our decision logic:

package ai.aletyx.examples.loan;

import org.drools.ruleunits.api.RuleUnitInstance;
import org.drools.ruleunits.api.RuleUnitProvider;

public class LoanDecisionService {

    /**
     * Evaluates a loan application and makes a decision.
     *
     * @param application The loan application to evaluate
     * @return The updated loan application with decision results
     */
    public LoanApplication evaluateLoanApplication(LoanApplication application) {
        // Initialize the application status
        application.setStatus("PENDING");

        // Create the rule unit
        LoanDecisionUnit loanDecisionUnit = new LoanDecisionUnit();

        // Add the application to the rule unit's data store
        loanDecisionUnit.getApplications().add(application);

        // Create a rule unit instance
        RuleUnitInstance<LoanDecisionUnit> instance =
            RuleUnitProvider.get().createRuleUnitInstance(loanDecisionUnit);

        try {
            // Execute the rules
            instance.fire();

            // Return the updated application (rules have modified it)
            return application;
        } finally {
            // Always dispose of the instance when done
            instance.dispose();
        }
    }
}

Step 6: Create Unit Tests

Let’s create a comprehensive test case:

package ai.aletyx.examples.loan;

import org.junit.jupiter.api.Test;
import static org.junit.jupiter.api.Assertions.*;

public class LoanDecisionServiceTest {

    private LoanDecisionService service = new LoanDecisionService();

    @Test
    public void testAutoRejectionDueToPoorCredit() {
        // Create an application with poor credit
        LoanApplication application = new LoanApplication();
        application.setId("LOAN-001");
        application.setApplicantName("John Doe");
        application.setCreditScore(520); // Poor credit
        application.setAnnualIncome(75000);
        application.setMonthlyDebt(1500);
        application.setRequestedAmount(150000);
        application.setRequestedTermInMonths(360);
        application.setPurpose("HOME_IMPROVEMENT");

        // Evaluate the application
        LoanApplication result = service.evaluateLoanApplication(application);

        // Assert the application was rejected
        assertEquals("REJECTED", result.getStatus());
        assertEquals("Credit score below minimum threshold", result.getDecisionReason());
    }

    @Test
    public void testAutoApprovalDueToExcellentCredit() {
        // Create an application with excellent credit
        LoanApplication application = new LoanApplication();
        application.setId("LOAN-002");
        application.setApplicantName("Jane Smith");
        application.setCreditScore(780); // Excellent credit
        application.setAnnualIncome(100000);
        application.setMonthlyDebt(2000);
        application.setRequestedAmount(200000);
        application.setRequestedTermInMonths(360);
        application.setPurpose("HOME_IMPROVEMENT");

        // Evaluate the application
        LoanApplication result = service.evaluateLoanApplication(application);

        // Assert the application was approved
        assertEquals("APPROVED", result.getStatus());
        assertEquals("Excellent credit profile with low risk assessment", result.getDecisionReason());
        assertNotEquals(0, result.getApprovedInterestRate());
        assertNotEquals(0, result.getCalculatedMonthlyPayment());
    }

    @Test
    public void testManualReviewForBorderlineCase() {
        // Create a borderline application
        LoanApplication application = new LoanApplication();
        application.setId("LOAN-003");
        application.setApplicantName("Michael Johnson");
        application.setCreditScore(680); // Good but not excellent
        application.setAnnualIncome(60000);
        application.setMonthlyDebt(1800);
        application.setRequestedAmount(180000);
        application.setRequestedTermInMonths(360);
        application.setPurpose("DEBT_CONSOLIDATION");

        // Evaluate the application
        LoanApplication result = service.evaluateLoanApplication(application);

        // Assert the application was sent for review
        assertEquals("REVIEW", result.getStatus());
        assertEquals("Application requires manual underwriter review", result.getDecisionReason());
    }

    @Test
    public void testRejectionDueToHighDebtToIncomeRatio() {
        // Create an application with high debt-to-income ratio
        LoanApplication application = new LoanApplication();
        application.setId("LOAN-004");
        application.setApplicantName("Sarah Williams");
        application.setCreditScore(700); // Good credit
        application.setAnnualIncome(50000);
        application.setMonthlyDebt(2500); // Monthly debt too high
        application.setRequestedAmount(100000);
        application.setRequestedTermInMonths(240);
        application.setPurpose("EDUCATION");

        // Evaluate the application
        LoanApplication result = service.evaluateLoanApplication(application);

        // Assert the application was rejected
        assertEquals("REJECTED", result.getStatus());
        assertEquals("Debt-to-income ratio exceeds maximum threshold", result.getDecisionReason());
    }
}

Step 7: Create a REST Service (Optional)

For a complete solution, we can expose our decision service as a REST endpoint:

package ai.aletyx.examples.loan;

import javax.inject.Inject;
import javax.ws.rs.*;
import javax.ws.rs.core.MediaType;
import javax.ws.rs.core.Response;

@Path("/loan-decisions")
@Produces(MediaType.APPLICATION_JSON)
@Consumes(MediaType.APPLICATION_JSON)
public class LoanDecisionResource {

    @Inject
    LoanDecisionService loanDecisionService;

    @POST
    @Path("/evaluate")
    public Response evaluateLoanApplication(LoanApplication application) {
        try {
            LoanApplication result = loanDecisionService.evaluateLoanApplication(application);
            return Response.ok(result).build();
        } catch (Exception e) {
            return Response.status(Response.Status.INTERNAL_SERVER_ERROR)
                    .entity("Error evaluating loan application: " + e.getMessage())
                    .build();
        }
    }
}

Advanced Decision Service Implementation Patterns

As your decision services grow in complexity, consider these advanced implementation patterns.

Decision Service Orchestration

Complex decisions often involve multiple sub-decisions. For example, a loan approval might include:

Credit risk assessment
Fraud detection
Regulatory compliance check
Pricing determination

Each sub-decision can be implemented as a separate rule unit:

public class LoanOrchestrationService {

    @Inject
    CreditRiskService creditRiskService;

    @Inject
    FraudDetectionService fraudDetectionService;

    @Inject
    ComplianceService complianceService;

    @Inject
    PricingService pricingService;

    public LoanDecision processLoanApplication(LoanApplication application) {
        // Step 1: Assess credit risk
        RiskAssessment riskAssessment = creditRiskService.assessRisk(application);
        if (riskAssessment.getRiskLevel() == RiskLevel.EXTREME) {
            return createRejection("Credit risk too high", riskAssessment);
        }

        // Step 2: Check for fraud
        FraudCheckResult fraudCheck = fraudDetectionService.checkForFraud(application);
        if (fraudCheck.isFraudSuspected()) {
            return createRejection("Fraud suspicion", fraudCheck);
        }

        // Step 3: Check compliance
        ComplianceResult complianceCheck = complianceService.checkCompliance(application);
        if (!complianceCheck.isCompliant()) {
            return createRejection("Compliance issues", complianceCheck);
        }

        // Step 4: Determine pricing if all checks passed
        PricingResult pricing = pricingService.determineInterestRate(
            application, riskAssessment, fraudCheck, complianceCheck);

        // Create approval with pricing details
        return createApproval(pricing, riskAssessment);
    }

    private LoanDecision createRejection(String reason, Object... evidence) {
        // Create rejection response with details
    }

    private LoanDecision createApproval(PricingResult pricing, RiskAssessment risk) {
        // Create approval response with details
    }
}

This approach:

Separates concerns into distinct services
Makes complex decision logic more maintainable
Allows independent testing of each component
Provides clear traceability for decisions

Event-Driven Decision Services

For high-throughput environments, consider implementing event-driven decision services:

@ApplicationScoped
public class EventDrivenLoanDecisionService {

    @Inject
    LoanDecisionService decisionService;

    @Inject
    Event<LoanDecisionEvent> decisionEventEmitter;

    @Incoming("loan-applications")
    public CompletionStage<Void> processApplication(LoanApplication application) {
        return CompletableFuture.runAsync(() -> {
            // Process the application
            LoanApplication result = decisionService.evaluateLoanApplication(application);

            // Emit appropriate event based on decision
            LoanDecisionEvent event = new LoanDecisionEvent(result);
            decisionEventEmitter.fire(event);

            // Route to appropriate downstream queue
            String targetQueue = determineTargetQueue(result);
            routeToQueue(result, targetQueue);
        });
    }

    private String determineTargetQueue(LoanApplication application) {
        switch (application.getStatus()) {
            case "APPROVED": return "approved-loans";
            case "REJECTED": return "rejected-loans";
            case "REVIEW": return "manual-review-loans";
            default: return "error-loans";
        }
    }

    private void routeToQueue(LoanApplication application, String queue) {
        // Route the application to the appropriate downstream queue
    }
}

This pattern:

Decouples decision-making from request handling
Supports high throughput through asynchronous processing
Enables natural integration with event-driven architectures
Improves scalability and resilience

Decision Caching

For performance-critical decision services, implement caching:

@ApplicationScoped
public class CachedLoanDecisionService {

    @Inject
    LoanDecisionService decisionService;

    private Cache<String, LoanApplication> decisionCache;

    @PostConstruct
    void initialize() {
        decisionCache = CacheBuilder.newBuilder()
            .maximumSize(10000)
            .expireAfterWrite(24, TimeUnit.HOURS)
            .build();
    }

    public LoanApplication evaluateLoanApplication(LoanApplication application) {
        // Generate a cache key based on relevant application attributes
        String cacheKey = generateCacheKey(application);

        // Check if we have a cached decision
        LoanApplication cachedResult = decisionCache.getIfPresent(cacheKey);
        if (cachedResult != null) {
            // Copy decision data to the original application
            copyDecisionData(cachedResult, application);
            return application;
        }

        // No cache hit, evaluate with rules
        LoanApplication result = decisionService.evaluateLoanApplication(application);

        // Cache the result
        decisionCache.put(cacheKey, cloneApplication(result));

        return result;
    }

    private String generateCacheKey(LoanApplication application) {
        // Generate a key based on the inputs that affect the decision
        return String.format("%d:%d:%.2f:%.2f:%d:%s",
            application.getCreditScore(),
            (int)(application.getAnnualIncome() / 1000), // Round to nearest thousand
            application.getMonthlyDebt(),
            application.getRequestedAmount(),
            application.getRequestedTermInMonths(),
            application.getPurpose());
    }

    // Helper methods for cloning and copying application data
    private LoanApplication cloneApplication(LoanApplication original) { /* ... */ }
    private void copyDecisionData(LoanApplication from, LoanApplication to) { /* ... */ }
}

Caching is particularly effective for:

High-volume decisions with similar inputs
Decisions with expensive calculations
Decision services with external calls
Regulatory scenarios where consistency is critical

Decision Service Versioning

Business rules change over time. Implement versioning to maintain backward compatibility:

@ApplicationScoped
public class VersionedLoanDecisionService {

    @Inject
    @Named("v1")
    LoanDecisionService decisionServiceV1;

    @Inject
    @Named("v2")
    LoanDecisionService decisionServiceV2;

    @Inject
    @Named("v3")
    LoanDecisionService decisionServiceV3;

    public LoanApplication evaluateLoanApplication(LoanApplication application, String version) {
        switch (version) {
            case "v1":
                return decisionServiceV1.evaluateLoanApplication(application);
            case "v2":
                return decisionServiceV2.evaluateLoanApplication(application);
            case "v3":
            default:
                return decisionServiceV3.evaluateLoanApplication(application);
        }
    }
}

You can implement versioning at different levels:

API versioning: Different API endpoints for each version
Rule versioning: Different rule sets for each version
Runtime versioning: Dynamic loading of rules based on version
Decision output versioning: Consistent output structure across versions

Decision Service Testing Strategies

Thorough testing is critical for decision services. Implement these testing strategies:

Decision Tables for Test Cases

Create a spreadsheet or CSV file with test cases:

testId,creditScore,annualIncome,monthlyDebt,requestedAmount,term,purpose,expectedStatus,expectedReason
Test001,520,75000,1500,150000,360,HOME_IMPROVEMENT,REJECTED,Credit score below minimum threshold
Test002,780,100000,2000,200000,360,HOME_IMPROVEMENT,APPROVED,Excellent credit profile with low risk assessment
Test003,680,60000,1800,180000,360,DEBT_CONSOLIDATION,REVIEW,Application requires manual underwriter review
Test004,700,50000,2500,100000,240,EDUCATION,REJECTED,Debt-to-income ratio exceeds maximum threshold

Then create a test that loads this table:

@Test
public void testDecisionServiceWithTestCases() throws Exception {
    // Load test cases from CSV
    List<TestCase> testCases = loadTestCasesFromCsv("loan-test-cases.csv");

    for (TestCase testCase : testCases) {
        // Create application from test case data
        LoanApplication application = createApplicationFromTestCase(testCase);

        // Evaluate the application
        LoanApplication result = service.evaluateLoanApplication(application);

        // Assert the result matches expectations
        assertEquals(testCase.getExpectedStatus(), result.getStatus(),
                     "Test case " + testCase.getTestId() + " failed: Incorrect status");

        if (testCase.getExpectedReason() != null) {
            assertEquals(testCase.getExpectedReason(), result.getDecisionReason(),
                         "Test case " + testCase.getTestId() + " failed: Incorrect reason");
        }
    }
}

This approach:

Makes test cases easy to review with business stakeholders
Provides comprehensive coverage of decision paths
Simplifies adding new test cases
Creates a clear specification for the decision service

Decision Service Simulation

For complex decision services, create a simulation environment:

@Test
public void simulateLoanDecisionService() {
    // Create a distribution of applications
    List<LoanApplication> simulatedApplications = new ArrayList<>();

    // Add applications with various characteristics
    for (int i = 0; i < 1000; i++) {
        simulatedApplications.add(generateRandomApplication());
    }

    // Process all applications
    Map<String, Integer> decisionCounts = new HashMap<>();
    for (LoanApplication application : simulatedApplications) {
        LoanApplication result = service.evaluateLoanApplication(application);

        // Count decisions by status
        decisionCounts.merge(result.getStatus(), 1, Integer::sum);
    }

    // Analyze decision distribution
    for (Map.Entry<String, Integer> entry : decisionCounts.entrySet()) {
        System.out.println(entry.getKey() + ": " + entry.getValue() +
                          " (" + (entry.getValue() * 100.0 / simulatedApplications.size()) + "%)");
    }

    // Assert reasonable decision distribution
    assertTrue(decisionCounts.getOrDefault("APPROVED", 0) > 0, "Should have some approvals");
    assertTrue(decisionCounts.getOrDefault("REJECTED", 0) > 0, "Should have some rejections");
    assertTrue(decisionCounts.getOrDefault("REVIEW", 0) > 0, "Should have some reviews");
}

private LoanApplication generateRandomApplication() {
    LoanApplication application = new LoanApplication();

    // Generate realistic random data
    application.setId("SIM-" + ThreadLocalRandom.current().nextInt(10000, 99999));
    application.setCreditScore(ThreadLocalRandom.current().nextInt(500, 850));
    application.setAnnualIncome(ThreadLocalRandom.current().nextInt(30000, 150000));

    // Set other properties with realistic distributions
    // ...

    return application;
}

This simulation approach helps:

Validate the decision distribution
Identify edge cases and boundary conditions
Analyze performance under load
Test business KPIs like approval rates

Decision Tracing and Explanation

Implement decision tracing to understand why rules fired:

public LoanApplication evaluateLoanApplicationWithTracing(LoanApplication application) {
    // Initialize the application status
    application.setStatus("PENDING");

    // Create the rule unit
    LoanDecisionUnit loanDecisionUnit = new LoanDecisionUnit();

    // Add the application to the rule unit's data store
    loanDecisionUnit.getApplications().add(application);

    // Create a rule unit instance with debug listener
    RuleUnitInstance<LoanDecisionUnit> instance =
        RuleUnitProvider.get().createRuleUnitInstance(loanDecisionUnit);

    // Add a debug listener for tracing
    DebugRuleRuntimeEventListener listener = new DebugRuleRuntimeEventListener();
    instance.addRuleRuntimeEventListener(listener);

    try {
        // Execute the rules
        instance.fire();

        // Add tracing information to result
        application.setRuleExecutionTrace(listener.getExecutedRules());

        return application;
    } finally {
        instance.dispose();
    }
}

This tracing information can:

Provide transparency into decisions
Help debug rule execution
Support regulatory requirements for explainability
Enable better customer communications

Decision Service Performance Tuning

As your decision services handle increased load, consider these performance optimizations:

Rule Ordering

Organize rules to minimize evaluation time:

// Run rejection rules first, with highest salience
rule "Reject Application with Poor Credit"
  salience 100
  when
    $application: /applications[ creditScore < 550 ]
  then
    // Quick rejection
end

This pattern:

Rejects invalid applications quickly
Avoids unnecessary computations
Improves average response time

Rule Partitioning

Split large rule sets into logical groups:

public LoanDecision processApplication(LoanApplication application) {
    // Phase 1: Validation rules
    ValidationResult validation = validationRules.validate(application);
    if (!validation.isValid()) {
        return createRejection(validation.getReason());
    }

    // Phase 2: Risk assessment rules
    RiskAssessment risk = riskRules.assessRisk(application);
    if (risk.getRiskLevel() == RiskLevel.EXTREME) {
        return createRejection("Excessive risk");
    }

    // Phase 3: Decision rules
    DecisionResult decision = decisionRules.makeDecision(application, risk);

    // Phase 4: Pricing rules (only if approved)
    if (decision.isApproved()) {
        PricingResult pricing = pricingRules.calculatePricing(application, risk);
        return createApproval(pricing);
    } else {
        return createRejection(decision.getReason());
    }
}

This approach:

Reduces the number of rules evaluated per phase
Improves maintainability of rule groups
Allows specialized optimization per rule group
Enables better performance monitoring

Fact Indexing

For large fact bases, ensure proper indexing:

// Add @key annotation to indexed fields in fact declarations
declare LoanApplication
    @key id : String
    @key applicantId : String
    creditScore : int
    // Other fields
end

Properly indexed facts:

Speed up pattern matching
Improve join performance
Reduce memory usage
Scale better with large fact bases

Stateless vs. Stateful Sessions

Choose the appropriate session type:

// Stateless for simple, independent decisions
public LoanDecision evaluateStateless(LoanApplication application) {
    StatelessKieSession session = kieContainer.newStatelessKieSession();
    LoanDecision decision = new LoanDecision();

    session.execute(CommandFactory.newInsert(application),
                    CommandFactory.newInsert(decision));

    return decision;
}

// Stateful for complex, interdependent decisions
public List<LoanDecision> evaluateBatch(List<LoanApplication> applications) {
    KieSession session = kieContainer.newKieSession();
    List<LoanDecision> decisions = new ArrayList<>();

    try {
        // Insert applications
        for (LoanApplication app : applications) {
            session.insert(app);
            LoanDecision decision = new LoanDecision(app.getId());
            decisions.add(decision);
            session.insert(decision);
        }

        // Insert market data and other shared facts
        session.insert(marketDataService.getCurrentData());

        // Fire rules
        session.fireAllRules();

        return decisions;
    } finally {
        session.dispose();
    }
}

Choose based on your requirements:

Stateless: Use for simple, independent decisions with small fact sets
Stateful: Use for complex decisions with shared context or interdependencies

Operational Aspects of Decision Services

Consider these operational factors for production-ready decision services:

Monitoring and Metrics

Implement comprehensive monitoring:

@Interceptor
public class DecisionMetricsInterceptor {

    @Inject
    MeterRegistry registry;

    @AroundInvoke
    public Object recordMetrics(InvocationContext context) throws Exception {
        long startTime = System.nanoTime();

        try {
            // Execute the decision service
            Object result = context.proceed();

            // Record metrics
            if (result instanceof LoanApplication) {
                LoanApplication application = (LoanApplication) result;
                recordDecisionMetrics(application, startTime);
            }

            return result;
        } catch (Exception e) {
            // Record error metrics
            registry.counter("loan.decision.errors",
                            "error", e.getClass().getSimpleName()).increment();
            throw e;
        }
    }

    private void recordDecisionMetrics(LoanApplication application, long startTime) {
        long duration = System.nanoTime() - startTime;

        // Record decision outcome
        registry.counter("loan.decision.outcomes",
                        "status", application.getStatus()).increment();

        // Record decision time
        registry.timer("loan.decision.time").record(duration, TimeUnit.NANOSECONDS);

        // Record risk score distribution
        registry.summary("loan.decision.riskScore").record(application.getRiskScore());

        // Record interest rate for approved loans
        if ("APPROVED".equals(application.getStatus())) {
            registry.summary("loan.decision.interestRate")
                  .record(application.getApprovedInterestRate());
        }
    }
}

Key metrics to track:

Decision outcomes (approved, rejected, review)
Decision times
Rule execution counts
Business KPIs (average interest rate, risk scores)
Error rates

Rule Deployment and Versioning

Implement a robust deployment strategy:

@ApplicationScoped
public class VersionedRuleDeploymentService {

    private Map<String, KieContainer> versionedContainers = new ConcurrentHashMap<>();

    @PostConstruct
    void initialize() {
        // Load all available rule versions
        loadRuleVersion("v1", "com/aletyx/rules/loan/v1");
        loadRuleVersion("v2", "com/aletyx/rules/loan/v2");
        loadRuleVersion("current", "com/aletyx/rules/loan/current");
    }

    private void loadRuleVersion(String version, String packagePath) {
        KieServices services = KieServices.Factory.get();
        KieFileSystem fileSystem = services.newKieFileSystem();

        // Load rules from classpath
        fileSystem.write(ResourceFactory.newClassPathResource(packagePath));

        // Build the knowledge base
        KieBuilder builder = services.newKieBuilder(fileSystem);
        builder.buildAll();

        if (builder.getResults().hasMessages(Message.Level.ERROR)) {
            throw new RuntimeException("Rule compilation errors: " +
                                      builder.getResults().getMessages());
        }

        // Create container and cache it
        KieContainer container = services.newKieContainer(
                services.getRepository().getDefaultReleaseId());
        versionedContainers.put(version, container);
    }

    public KieContainer getRuleContainer(String version) {
        if (!versionedContainers.containsKey(version)) {
            throw new IllegalArgumentException("Unknown rule version: " + version);
        }
        return versionedContainers.get(version);
    }
}

This approach enables:

Progressive rollout of rule changes
A/B testing of rule versions
Safe rollback capabilities
Maintenance of backward compatibility

Decision Service Audit Logging

Implement comprehensive audit logging for regulatory compliance:

@Interceptor
public class DecisionAuditInterceptor {

    @Inject
    AuditLogger auditLogger;

    @AroundInvoke
    public Object auditDecision(InvocationContext context) throws Exception {
        // Extract the loan application from method parameters
        LoanApplication application = findLoanApplication(context.getParameters());
        if (application == null) {
            return context.proceed();
        }

        // Clone the application for before state
        LoanApplication before = cloneApplication(application);

        // Execute the decision service
        Object result = context.proceed();

        // Extract the after state
        LoanApplication after = (result instanceof LoanApplication) ?
                (LoanApplication) result : application;

        // Create audit entry
        DecisionAuditEntry entry = new DecisionAuditEntry();
        entry.setDecisionId(UUID.randomUUID().toString());
        entry.setApplicationId(application.getId());
        entry.setTimestamp(new Date());
        entry.setUserId(getCurrentUser());
        entry.setRuleVersion(getRuleVersion());
        entry.setInputState(serializeState(before));
        entry.setOutputState(serializeState(after));
        entry.setDecisionOutcome(after.getStatus());
        entry.setDecisionReason(after.getDecisionReason());

        // Log the audit entry
        auditLogger.logDecision(entry);

        return result;
    }

    // Helper methods
    private LoanApplication findLoanApplication(Object[] parameters) { /* ... */ }
    private LoanApplication cloneApplication(LoanApplication app) { /* ... */ }
    private String getCurrentUser() { /* ... */ }
    private String getRuleVersion() { /* ... */ }
    private String serializeState(LoanApplication app) { /* ... */ }
}

Audit logs should capture:

Input data (application details)
Output decisions
Rule versions used
Timestamps
User/system identifiers
Rule execution trace

Summary

Implementing decision services with DRL enables you to:

Capture complex business logic in a maintainable, declarative format
Create modular, reusable decision components that integrate with your applications
Adapt quickly to changing business requirements by updating rules instead of code
Scale decision processing to handle high volumes efficiently
Provide transparency and traceability for regulatory compliance

By following the patterns and best practices outlined in this chapter, you can build robust, high-performance decision services that deliver consistent, accurate business decisions. As your decision services mature, consider expanding into more sophisticated approaches like decision tables, complex event processing, and predictive analytics integration to further enhance your business decision automation capabilities.

​Understanding Decision Services

​What is a Decision Service?

​Common Decision Service Use Cases

​Building a Loan Approval Decision Service

​Step 1: Gather Requirements

​Functional Requirements

​Business Rules (Simplified Example)

​Step 2: Design the Domain Model

​Step 3: Create the Rule Unit

​Step 4: Write the Rules

​Step 5: Implement the Service Layer

​Step 6: Create Unit Tests

​Step 7: Create a REST Service (Optional)

​Advanced Decision Service Implementation Patterns

​Decision Service Orchestration

​Event-Driven Decision Services

​Decision Caching

​Decision Service Versioning

​Decision Service Testing Strategies

​Decision Tables for Test Cases

​Decision Service Simulation

​Decision Tracing and Explanation

​Decision Service Performance Tuning

​Rule Ordering

​Rule Partitioning

​Fact Indexing

​Stateless vs. Stateful Sessions

​Operational Aspects of Decision Services

​Monitoring and Metrics

​Rule Deployment and Versioning

​Decision Service Audit Logging

​Summary

Understanding Decision Services

What is a Decision Service?

Common Decision Service Use Cases

Building a Loan Approval Decision Service

Step 1: Gather Requirements

Functional Requirements

Business Rules (Simplified Example)

Step 2: Design the Domain Model

Step 3: Create the Rule Unit

Step 4: Write the Rules

Step 5: Implement the Service Layer

Step 6: Create Unit Tests

Step 7: Create a REST Service (Optional)

Advanced Decision Service Implementation Patterns

Decision Service Orchestration

Event-Driven Decision Services

Decision Caching

Decision Service Versioning

Decision Service Testing Strategies

Decision Tables for Test Cases

Decision Service Simulation

Decision Tracing and Explanation

Decision Service Performance Tuning

Rule Ordering

Rule Partitioning

Fact Indexing

Stateless vs. Stateful Sessions

Operational Aspects of Decision Services

Monitoring and Metrics

Rule Deployment and Versioning

Decision Service Audit Logging

Summary