Test Plan & Acceptance Criteria

ARIA Agentic AI | SkyWave Airlines IROPS Deployment

Test PhasesPhase 1: Unit & Integration | Phase 2: UAT | Phase 3: Load | Phase 4: Shadow
Total Test Scenarios48 functional + 12 performance + 8 security + 6 chaos scenarios
Go/No-Go Threshold>=95% pass rate on P1 scenarios; 0 open P0 defects; all acceptance criteria met

1. Acceptance Criteria Summary

Acceptance CriterionThresholdMeasurementTest Method
Intent classification accuracy (top 20 intents)>= 92%Automated NLU testing2,000 labeled utterances
End-to-end rebooking success (no human)>= 90%E2E test suite200 scenario executions
Compensation issuance accuracy100% per policy rulesRule engine validationAll 12 loyalty × disruption combos
Escalation trigger accuracy>= 97%Manual + automated review50 escalation test cases
Agent response latency P95< 2.5 secondsLoad testing toolJMeter at 10K concurrent sessions
PII data not exposed in logs0 violationsLog audit scanAutomated PII pattern scanner
Circuit breaker activates on GDS failure100% within 30sChaos testKill GDS mock mid-session
Proactive outreach sent within 5 min of IROPS>= 98%Event timestamp comparison20 IROPS event simulations

2. Functional Test Scenario Matrix

2.1 IROPS Rebooking Scenarios

TC#ScenarioPreconditionsExpected ResultPriority
TC-001Flight cancelled — 1 alt availableIROPS declared; 1 rebooking optionARIA offers option; confirms; reissues BPP0
TC-002Flight cancelled — 3 alts availableIROPS declared; 3 options availableARIA presents all 3; books customer choiceP0
TC-003Flight cancelled — 0 alts same dayNo same-day availabilityARIA offers next-day options + hotel voucherP0
TC-004Delay > 4 hours — rebookingDelay confirmed; customer requests rebookARIA processes as IROPS; same flowP1
TC-005Platinum member — priority rebookingLoyalty: Platinum; multiple optionsPlatinum options shown first; $300 comp autoP0
TC-006Basic member — compensation ceilingLoyalty: Basic; cancellationARIA issues $75 max; does not exceed policyP0
TC-007Customer refuses all options3 options offered; all declinedARIA offers full refund option + escalationP1
TC-008Multi-leg itinerary disruptionConnection cancelled; downstream legs affectedARIA identifies all affected legs; rebooks allP0

2.2 Escalation & Handoff Scenarios

TC#ScenarioTriggerExpected ResultPriority
TC-021Customer requests agent explicitly'I want to speak to a human'Immediate transfer; full context in Zendesk ticketP0
TC-0223 consecutive disambiguation failures3 unclear inputsMandatory escalation; empathetic messageP0
TC-023Confidence < 0.72 on 2nd attemptLow NLU confidence scoreSoft escalation offer; customer choosesP1
TC-024Unaccompanied minor rebookingMinor flag on PNRImmediate escalation; no autonomous actionP0
TC-025Context successfully passed on escalationAny escalation scenarioHuman agent sees full transcript; no re-askP0

2.3 Compensation & Voucher Scenarios

TC#ScenarioLoyalty TierExpected CompensationPriority
TC-0314+ hour delay compensation requestPlatinum$300 voucher + 5,000 miles auto-issuedP0
TC-032Flight cancellation compensationGold$150 voucher issued; no miles (policy)P0
TC-033Meal voucher — delay > 3 hoursAny$15 meal voucher code sent to emailP1
TC-034Overnight cancellation — hotelAnyHotel voucher + $50 transportation issuedP1
TC-035Duplicate compensation attemptAnyARIA detects prior claim; blocks duplicateP0

3. Performance Test Plan

3.1 Load Testing Configuration

Test TypeParametersSuccess Criteria
Baseline Load500 concurrent users; 10-minute ramp< 1.5s P95 latency; 0 errors
Peak IROPS Load10,000 concurrent sessions; 5-minute ramp< 2.5s P95; < 0.1% error rate
Spike Test0 to 8,000 sessions in 60 secondsSystem recovers within 90s; no data loss
Soak Test3,000 sessions for 4 hours continuousNo memory leaks; stable latency throughout
GDS Failure TestKill GDS mock at 5,000 session peakCircuit breaker in < 30s; cache fallback active

4. UAT Sign-Off Process

User Acceptance Testing will be conducted over a 2-week period in a staging environment mirroring production:

  • Week 1: Business stakeholders execute 30 scripted scenarios from the functional matrix above
  • Week 1: CX team evaluates agent tone, response quality, and escalation behavior on 20 additional exploratory tests
  • Week 2: Regression cycle after defect fixes; QA sign-off on all P0 and P1 scenarios
  • Week 2: Performance test execution and review with Engineering team
  • Final: Go/No-Go meeting with VP CX, Head of Digital Transformation, Engineering Lead, and Netomi CSM
Sign-Off StakeholderSign-Off GateCriteria
QA LeadAll phases0 open P0 defects; >= 95% P1 pass rate
VP Customer ExperienceUATAgent tone and resolution quality approved
CISOSecurity0 PII exposures; encryption verified
Engineering LeadPerformanceAll latency and availability thresholds met
Netomi CSMFinalPlatform health confirmed; monitoring live

Document prepared as part of an Agentic AI solution design portfolio case study for enterprise customer experience transformation.