2026-08-19

ADR-006: Security Architecture for a Retail Loyalty Platform

A Zero Trust security approach for a cloud-native retail loyalty platform, including the PCI-DSS scope decision and loyalty-specific fraud controls, with the critique and fixes that followed.

Author: Lanre — Cloud Security / Enterprise Architect
Certifications: CISSP, CCSP, CISM, ISSAP, GCP-PCA
Date: 2026-08-20


Overview

ADR-006 addresses a retailer building a cloud-native loyalty platform. It decides on Zero Trust architecture, SAQ-A PCI scope minimization via a hosted payment page, and loyalty-specific fraud controls.

The companion record is ADR-005: Regional Data Residency and Hierarchical Encryption .


Contents


FieldValue
TitleSecurity Architecture Approach for Cloud-Native Retail Loyalty Platform
StatusProposed
Date2026-08-19
AuthorLanre
StakeholdersCISO, Enterprise Architecture, Platform Engineering, Compliance (PCI-DSS), Privacy Officer, Fraud Operations

1. Context

A retailer is building a new customer loyalty platform on public cloud with the following components:

  • Public web and mobile front end
  • API gateway / layer
  • Microservices back end
  • Customer-profile data store (personal data)
  • Third-party payment processor integration
  • Third-party email provider integration

The architect must establish the security architecture approach before detailed design begins. This ADR records the decision on frameworks, threat-modeling methodology, and control strategy.

Business Context: Loyalty points are a stored-value currency inside the retailer’s ecosystem. Points balances, earn/burn transactions, referral bonuses, and promotional credits are direct financial liability and fraud targets. The primary loyalty-specific threat vectors are account takeover for points theft, points laundering across accounts, insider point issuance, and business-logic manipulation of earn/burn rules.

2. Decision

We will adopt a defense-in-depth, Zero Trust security architecture grounded in inherited enterprise frameworks, not greenfield invention.

2.1 Framework Inheritance

LayerFramework / StandardRole
Enterprise ArchitectureTOGAF ADMSecurity work woven into relevant ADM phases
Risk & TraceabilitySABSAMap every control to a business driver and documented requirement
Cloud SecurityCSA Enterprise ArchitectureCloud-native reference architecture
Design PrincipleZero TrustNo implicit trust; authenticate/authorize every inter-service call; microsegmentation; pervasive monitoring

2.2 Threat-Modeling Stack

MethodPurposeScope
STRIDESystematic per-element, per-boundary threat enumerationSpoofing, Tampering, Repudiation, Information Disclosure, DoS, Elevation of Privilege
LINDDUNPrivacy threat analysisLinkability, Identifiability, Non-repudiation, Detectability, Disclosure of information, Unawareness, Non-compliance
MITRE ATT&CKThreat realism & prioritizationRetail-sector threat intelligence — three techniques cited in §3.2

CVSS Exclusion: CVSS scoring belongs to post-implementation vulnerability management (operational phase). It is out of scope for this pre-design architecture decision.

2.3 Trust Boundaries

The data flow diagram establishes the following trust boundaries:

  1. Internet → Front End
  2. Front End → API Layer
  3. API Layer → Microservices
  4. Microservices → Customer-Profile Data Store
  5. Microservices → Third-Party Payment Processor
  6. Microservices → Third-Party Email Provider

2.4 Rejected Alternatives

Alternative A: Greenfield Custom Security Framework

AspectAssessment
DescriptionBuild a bespoke security framework and control taxonomy specific to the loyalty platform, without inheriting enterprise standards
Why RejectedReinvents validated patterns; creates audit friction (auditors cannot map controls to recognized standards); no vendor or community support; knowledge walks out the door when architects leave. The enterprise has already invested in TOGAF, SABSA, and CSA — abandoning that investment for one platform is unjustified.
Residual UseNone — all security work must inherit enterprise frameworks

Alternative B: Perimeter-Based (Castle-and-Moat) Architecture

AspectAssessment
DescriptionImplicit trust inside the network perimeter; strong border controls (firewall, WAF) with weak internal segmentation
Why RejectedIncompatible with microservices architecture — implicit trust between services allows lateral movement on compromise. The loyalty platform’s microservices communicate east-west extensively; a perimeter model would leave the points-ledger service implicitly trusted by the profile service, creating a direct path from front-end compromise to balance manipulation. Also fails to address insider threat and supply-chain compromise.
Residual UseNone — perimeter controls (WAF, DDoS protection) are layered under Zero Trust, not replacing it

Alternative C: Framework Inheritance Without Zero Trust Principles

AspectAssessment
DescriptionUse TOGAF + SABSA + CSA for structure and documentation, but do not enforce per-service authentication, microsegmentation, or least-privilege access
Why RejectedFrameworks provide governance and traceability, but without Zero Trust execution principles, the architecture retains implicit trust between services. The result is a well-documented design that still allows lateral movement and fails to limit blast radius. Zero Trust is the operational layer that makes the frameworks meaningful.
Residual UseNone — Zero Trust principles are mandatory for this platform

3. Threat-to-Control Mapping

3.1 Security Controls (STRIDE)

STRIDE CategoryThreatControlRequirement Traceability
SpoofingUser impersonation; service-to-service spoofingStrong authentication (OAuth 2.0 / mTLS); mutual service authenticationREQ-001: Identity & Access Management
TamperingProfile data or API request modification; loyalty points balance manipulationInput validation; integrity controls; request signing; business-logic validation layer for earn/burn transactions with anomaly detectionREQ-002: Data Integrity; REQ-010: Fraud Prevention
RepudiationDenial of profile changes or transactionsImmutable audit logging; non-repudiation mechanismsREQ-003: Accountability
Information DisclosurePersonal and payment data exposure in transit/restEncryption (TLS 1.3, AES-256-GCM); access control; tokenization of payment dataREQ-004: Confidentiality; REQ-005: PCI-DSS Scope Minimization
Denial of ServicePublic front-end overloadRate limiting; WAF; auto-scaling; circuit breakersREQ-006: Availability
Elevation of PrivilegeLateral movement within microservicesLeast privilege (RBAC/ABAC); authorization checks; workload isolation (containers / VPC-SC)REQ-007: Authorization

3.2 Loyalty-Specific Threats (Fraud & Abuse)

ThreatAttack VectorControlRequirement Traceability
Points Theft via ATOCredential stuffing, password reuse, phishing → account takeover → points transfer or redemptionMFA on high-value actions; velocity checks on points transfers; device fingerprinting; account lockout after failed attemptsREQ-010: Fraud Prevention
Points LaunderingStolen points moved through mule accounts; secondary market resaleTransfer limits; cooling-off periods; recipient account age verification; machine-learning anomaly detection on transfer patternsREQ-010: Fraud Prevention
Referral / Promo AbuseSynthetic account creation to harvest referral bonuses; repeated use of one-time promo codesIdentity verification (KYC-lite) for referral payouts; promo-code rate limiting; device/IP clusteringREQ-010: Fraud Prevention
Insider Point IssuanceAuthorized staff manually crediting points to colluding accountsDual-control for manual point adjustments; immutable audit trail with tamper-evident hashing; SOX-style segregation of dutiesREQ-003: Accountability; REQ-011: Insider Threat
Earn/Burn Logic ManipulationAPI parameter tampering to inflate earn rates or bypass burn validationServer-side validation of all earn/burn calculations; immutable transaction ledger; reconciliation engine comparing expected vs. actual points flowREQ-002: Data Integrity; REQ-010: Fraud Prevention

3.3 MITRE ATT&CK Mapping (Retail Sector)

Technique IDTechnique NameLoyalty Platform Relevance
T1589Gather Victim Identity InformationHarvesting loyalty account credentials from breach dumps for account takeover
T1110.004Credential StuffingAutomated credential stuffing against loyalty login APIs
T1078Valid AccountsAbuse of compromised legitimate accounts to transfer or redeem stolen points

3.4 Privacy Controls (LINDDUN)

LINDDUN CategoryPrivacy ThreatControlSTRIDE Collision Resolution
LinkabilityBehavioral profiling across sessionsData minimization; separate data stores for identity vs. behavioral data; session isolationNone — STRIDE does not address linkability
IdentifiabilityRe-identification from pseudonymous dataPseudonymization of customer profiles; k-anonymity checks on analytics exportsNone — STRIDE information disclosure covers unauthorized access, not re-identification risk
Non-repudiation (Privacy)Data subject cannot prove privacy violation or incorrect processingDivergence from STRIDE: STRIDE non-repudiation protects the system (user cannot deny a transaction). LINDDUN non-repudiation protects the data subject (user cannot prove a privacy breach). Resolution: Implement both — immutable audit logs for STRIDE (transaction non-repudiation) AND a data-subject access log / privacy dashboard for LINDDUN (privacy non-repudiation).Explicitly resolved in architecture
DetectabilityAdversary can infer user membership in loyalty programMinimize public profile enumeration; return generic errors; rate-limit lookup APIsNone
Disclosure of Information (Privacy)Unauthorized access to PII beyond loyalty dataEncryption at rest/transit; access control; DLP monitoringAligns with STRIDE Information Disclosure — unified control
UnawarenessCustomer unaware of profiling, data retention, or third-party sharingGranular consent management; just-in-time disclosures at earn/burn; clear retention policies; privacy dashboardNone — STRIDE does not address awareness
Non-complianceRetention beyond legal limit; processing without legal basisAutomated data-retention enforcement; TTL on behavioral data; periodic compliance scans; consent expiry checksNone — STRIDE does not address compliance

4. PCI-DSS Scope Decision

The most consequential security decision for this platform is the payment integration architecture, as it determines PCI-DSS scope and audit burden.

4.1 Selected Approach: Hosted Payment Page (SAQ A)

AspectDecision
Integration ModelRedirect to payment processor’s hosted payment page (HPP) or embedded iframe (iframe-based SAQ A)
MobileProcessor’s certified mobile SDK (SAQ A)
TokenizationProcessor vaults card data; platform stores only non-sensitive tokens and last-4 digits
PCI ScopeSAQ A — lowest scope; platform does not touch, process, or store CHD
Network SegmentationPayment token returned to platform; no CHD enters platform network boundary

4.2 Rejected Alternative: Direct Post / API Integration (SAQ A-EP / SAQ D)

AspectAssessment
DescriptionPlatform front end posts card data directly to platform API, which then forwards to processor; or platform API accepts card data directly
Why RejectedExpands PCI-DSS scope to the entire platform (SAQ D or A-EP), requiring network segmentation, ASV scans, penetration testing, and annual QSA audit of the full environment. Rough order-of-magnitude cost increase: ~$150K–$300K/year versus SAQ A, based on additional QSA hours, ASV scanning of platform infrastructure, and broader penetration-test scope. The business does not require direct card handling; tokenization via HPP satisfies all functional requirements.
Residual UseOnly if processor HPP does not support required payment methods (e.g., certain BNPL providers); requires explicit CISO and QSA sign-off

5. Trust Boundary Controls

5.1 Boundaries 1–4 (Internet → Front End → API → Microservices → Data Store)

Covered in §3.1 (STRIDE table) and §3.2 (loyalty-specific threats).

5.2 Boundary 5: Microservices → Third-Party Payment Processor

ThreatControlRequirement Traceability
Man-in-the-middleTLS 1.3 with certificate pinning; mutual TLS where supported by processorREQ-004: Confidentiality
Token replay / substitutionToken validation with processor on each transaction; token expiry; token-to-account bindingREQ-010: Fraud Prevention
Processor breach expanding platform liabilityContractual BAA/processor DPA; right to audit; breach-notification SLA (24 hours); sub-processor disclosureREQ-008: Third-Party Risk
Availability dependencyCircuit breaker on payment API; graceful degradation (queue transactions for retry); multi-processor fallbackREQ-006: Availability

5.3 Boundary 6: Microservices → Third-Party Email Provider

ThreatControlRequirement Traceability
Account takeover via email compromiseContractual MFA requirement on email provider admin accounts; monitor provider breach disclosures; disable password-reset-via-email for high-value actions; fallback to app-based recovery with velocity limitsREQ-001: Identity & Access Management
PII leakage in email contentNo PII in email templates beyond first name; all sensitive notifications require login to portal; suppress full statementsREQ-004: Confidentiality
Sub-processor complianceDPA with email provider; data residency commitment; no onward transfer; breach-notification SLAREQ-008: Third-Party Risk
Phishing via spoofed provider domainSPF/DKIM/DMARC enforcement on provider-sending domain; brand monitoringREQ-009: Brand Protection
Email enumerationGeneric “check your email” responses; no confirmation of email existence in API responsesREQ-004: Confidentiality

6. Compliance Mapping

Regulation / StandardTriggerControl Implication
PCI DSSPayment processor integrationScope minimization via hosted payment page (SAQ A); tokenization; no CHD enters platform boundary
Privacy Regulations (GDPR/CCPA)Personal data in profile storePrivacy-by-design; data minimization; pseudonymization; consent logging; retention enforcement
Fraud / Financial CrimeLoyalty points as stored valueTransaction monitoring; anomaly detection; velocity limits; insider controls; reconciliation engine

7. Consequences

Positive

  • Inherits validated enterprise patterns (TOGAF, SABSA, CSA), which reduces design risk and audit friction.
  • Zero Trust architecture eliminates implicit trust and limits the blast radius of a compromise.
  • Dual threat-modeling (STRIDE + LINDDUN) covers security and privacy. §3.4 resolves the collision between STRIDE and LINDDUN non-repudiation.
  • Loyalty-specific fraud controls (§3.2) address the primary business risk that generic web-app security misses.
  • PCI scope minimization via SAQ A avoids ~$150K–$300K/year in additional audit overhead.
  • Traceability to documented requirements (REQ-001 … REQ-011) enables verification and validation in later phases.

Negative / Risks

  • Complexity: Mutual TLS and per-service authorization increase operational overhead.
  • Latency: Additional authentication/authorization hops may increase API response times. Requires a performance baseline.
  • Third-party dependency: Payment processor and email provider security postures sit outside direct control. Requires contractual SLAs and continuous monitoring.
  • Fraud false positives: Anomaly detection on points transfers may flag legitimate behavior. Requires tuning and customer communication.
  • PCI residual risk: Any misconfiguration in tokenization or segmentation could expand audit scope. Quarterly self-assessment required.
  • ADR-001: Platform-Wide Identity Federation Strategy (OAuth 2.0 / OIDC)
  • ADR-002: API Gateway Selection and Rate-Limiting Policy
  • ADR-003: Microservices Service Mesh and mTLS Configuration
  • ADR-004: Customer Profile Data Store Encryption and Retention

9. References

  • TOGAF Standard, Version 9.2 — ADM Security Architecture
  • SABSA — Sherwood Applied Business Security Architecture
  • CSA Enterprise Architecture v4
  • NIST SP 800-207 — Zero Trust Architecture
  • OWASP Threat Modeling Cheat Sheet (STRIDE)
  • LINDDUN Privacy Threat Modeling Framework
  • MITRE ATT&CK for Enterprise
  • PCI DSS v4.0 Requirements and Testing Procedures
  • SAQ A and SAQ A-EP Eligibility Criteria (PCI SSC)

Fix Log: Validation & Critique Response

Defects found during adversarial review of this ADR, and the fixes applied.

ADR-006 (Retail Loyalty) — Fixes Applied

#IssueFix
1Nothing loyalty-specific. STRIDE table would fit any web app. Missing points theft, laundering, promo abuse, insider issuance, balance manipulation.Added §3.2 Loyalty-Specific Threats (Fraud & Abuse) with five threat rows: Points Theft via ATO, Points Laundering, Referral/Promo Abuse, Insider Point Issuance, Earn/Burn Logic Manipulation.
2PCI decision not made. “Tokenization” and “scope minimization” are outcomes, not decisions. Missing SAQ A vs. A-EP vs. D choice.Added §4 PCI-DSS Scope Decision: explicitly selected Hosted Payment Page (SAQ A). Rejected Direct Post (SAQ A-EP/D) with costed rationale.
3Two trust boundaries (5: payment processor, 6: email provider) had no controls mapped. Email provider is live ATO path and sub-processor.Added §5 Trust Boundary Controls with dedicated subsections for Boundary 5 (payment processor) and Boundary 6 (email provider).
4LINDDUN covered only 2 of 7 categories. Unawareness and Non-compliance are live for loyalty. STRIDE non-repudiation collided with LINDDUN non-repudiation; claimed “without duplication” but they collide.Expanded LINDDUN to all 7 categories. Explicitly resolved collision: STRIDE non-repudiation = system protection (immutable audit logs); LINDDUN non-repudiation = data-subject protection (privacy dashboard). Architecture implements both.
5MITRE ATT&CK and CVSS named but not used. No techniques cited. CVSS described as “post-implementation” which is out of scope for pre-design ADR.Added §3.3 MITRE ATT&CK Mapping with three actual techniques (T1589, T1110.004, T1078). Removed CVSS from threat-modeling stack; added explicit exclusion note.
6“Domain 1 Traceability” leaked study origin. No CISO knows what it means.Replaced with real requirement IDs (REQ-001 … REQ-011) with descriptive names.
7§6 referenced ADR-001 (AI vendor / ECOA / SR 11-7) and ADR-003 (PAM) — bank decisions in a retailer’s ADR.Replaced cross-references with retail-loyalty-appropriate ADRs (Identity Federation, API Gateway, Service Mesh, Data Store Encryption).
8ADR-006 had no rejected alternatives for its main security architecture decision.Added §2.4 Rejected Alternatives for the overall approach: Greenfield Custom Framework, Perimeter-Based Architecture, and Frameworks Without Zero Trust.
9Boundary 6, first row: “Control” cell restated the threat instead of stating a control.Replaced with actual control: contractual MFA on provider admin accounts, monitoring breach disclosures, disabling email-based password reset for high-value actions, app-based recovery with velocity limits.
10T1496 (Resource Hijacking) incorrectly mapped to loyalty points. T1496 refers to compute resource hijacking (cryptomining).Replaced with T1078 (Valid Accounts) — abuse of compromised legitimate accounts for points transfer/redemption. Kept T1110.004 (Credential Stuffing) and T1589 (Gather Victim Identity Information).
11Unsourced dollar figures for PCI cost savings.Added rough order-of-magnitude label with basis: QSA hours, ASV scanning, penetration-test scope.

Tags: architecture, security, cloud, compliance, ADR, PCI-DSS, loyalty-platform, zero-trust