ometere123/the-skills-bible/advertising-systems-engineer

Advertising Systems Engineer

Purpose

Deliver and measure advertising under explicit commercial, privacy, safety, and fairness constraints.

Use this skill when

• A platform sells or serves advertising.
• Campaign budgets and pacing need control.
• Measurement and billing must reconcile.

Do not use this skill as the primary workflow when

• The task is general recommendation ranking.
• The user seeks advertising creative strategy.
• Consent and privacy authority are unresolved.

Required inputs

• Advertisers, publishers, and users.
• Campaign and inventory model.
• Targeting and consent policy.
• Auction, pacing, measurement, attribution, and billing.

Classify missing information as blocking, important, or optional. Inspect the actual policies, contracts, domain models, workflows, records, integrations, data, permissions, incidents, reports, operational procedures, and regulatory constraints before asking for information that can be verified directly.

Core operating principles

• Model the domain's authoritative entities, states, invariants, actors, and evidence before selecting technology.
• Separate legal, policy, commercial, operational, and technical rules so each can change under accountable ownership.
• Treat money, identity, safety, entitlement, inventory, consent, and irreversible actions as high-integrity state.
• Use explicit state machines, idempotency, audit trails, reconciliation, and exception handling for consequential workflows.
• Preserve tenant, customer, patient, citizen, partner, and operator boundaries throughout data and action paths.
• Design for real-world delay, duplication, dispute, correction, cancellation, partial completion, and external-system failure.
• Keep generated recommendations and probabilistic scores separate from authoritative decisions and records.
• Minimise sensitive data and apply purpose, retention, access, and deletion controls across primary and derived copies.
• Make operational outcomes observable through stable identifiers, timelines, ownership, and evidence.
• Validate domain behaviour with qualified stakeholders, representative scenarios, historical failures, and safe recovery exercises.

Workflow

1. Model campaign

• Define objective, creative, audience, placement, schedule, bid, budget, frequency, and status.

2. Model inventory

• Describe eligible surfaces, formats, capacity, quality, context, and publisher restrictions.

3. Apply privacy and eligibility

• Use consent, age, geography, sensitive-category, publisher, and advertiser policies before targeting.

4. Design candidate selection

• Filter by compatibility, policy, budget, pacing, frequency, and relevance.

5. Design auction or priority

• Use transparent bid, quality, reservation, guaranteed, or house rules with deterministic records.

6. Control budget and pacing

• Reserve spend, pace across time, cap frequency, and handle delayed events and overspend.

7. Serve creative safely

• Validate assets, destinations, tracking, malware risk, disclosure, accessibility, and rendering.

8. Measure delivery

• Define request, opportunity, impression, view, click, conversion, invalid traffic, and deduplication.

9. Attribute and bill

• Use versioned windows and models, preserve uncertainty, reconcile delivery, and generate chargeable records.

10. Validate

• Test budget race, delayed conversion, consent withdrawal, invalid traffic, auction tie, and reporting correction.

Decision rules

Required outputs

• Campaign and inventory models.
• Privacy eligibility.
• Candidate and auction rules.
• Budget and creative controls.
• Measurement, attribution, and billing.
• Advertising tests.

Domain-system quality gates

• Domain gate: Core entities, terminology, state transitions, and authoritative sources are agreed.
• Invariant gate: Money, safety, identity, entitlement, inventory, and policy invariants are enforceable.
• Authority gate: Every decision and state change has a trusted actor, permission, and accountable owner.
• Evidence gate: Consequential outcomes retain provenance, timestamps, reasons, and supporting records.
• Exception gate: Disputes, reversals, corrections, manual review, and partial failure have explicit paths.
• Privacy gate: Sensitive data is purpose-bound, minimised, access-controlled, retained, and deleted appropriately.
• Interoperability gate: External identifiers, schemas, versions, and reconciliation rules are documented and tested.
• Operational gate: Monitoring, support, incident response, recovery, and business continuity are ready.
• Fairness gate: Policies and automation are reviewed for unjustified exclusion, unequal burden, and appeal access.
• Lifecycle gate: Onboarding, change, migration, deprecation, offboarding, archival, and retirement are governed.

Evidence standard

For consequential domain conclusions, identify the exact policy, contract, jurisdiction or operating context, actor, system version, record, event time, effective time, external reference, and source evidence. Distinguish authoritative fact, derived calculation, policy interpretation, probabilistic assessment, and unresolved dispute. Do not claim legal, clinical, financial, or safety correctness from software behaviour alone.

Failure handling and escalation

Pause or escalate when:

• The authoritative rule, source of truth, accountable owner, or affected party is unclear.
• Money, inventory, identity, entitlement, clinical, public-safety, or legal state may be corrupted.
• An automated decision could materially disadvantage a person without review, explanation, or appeal.
• External systems disagree and reconciliation cannot establish authoritative state.
• Required consent, licence, contract, credential, or jurisdictional authority cannot be verified.
• Production testing could create real charges, orders, messages, appointments, claims, or public records.
• Residual financial, legal, safety, privacy, or operational risk exceeds delegated authority.

Report the exact blocker, verified state, affected parties, evidence, safe options, and smallest required decision.

Common anti-patterns

• Targeting before consent check.
• Budget checked after impression.
• Click equals conversion automatically.
• Advertiser report differs from billing source.
• Sponsored content undisclosed.

Completion checklist

• [ ] Campaigns and inventory have authority.
• [ ] Privacy precedes targeting.
• [ ] Budgets cannot overspend materially.
• [ ] Delivery events are defined and deduplicated.
• [ ] Billing reconciles.

Domain implementation record

# Domain Implementation Record
 
## Outcome
[Business, public-service, clinical, operational, or user outcome]
 
## Domain scope
[Actors, entities, jurisdictions, policies, environments, and exclusions]
 
## Authoritative state
[Systems of record, ownership, identifiers, and effective-time rules]
 
## Workflow
[States, transitions, approvals, exceptions, reversals, and deadlines]
 
## Invariants
[Money, safety, identity, entitlement, inventory, privacy, and fairness]
 
## Integrations
[Partners, protocols, schemas, credentials, delivery, and reconciliation]
 
## Evidence and audit
[Events, records, reasons, provenance, access, and retention]
 
## Validation
[Scenarios, qualified review, historical cases, recovery, and residual gaps]
 
## Operations
[Monitoring, support, incident response, continuity, and manual fallback]
 
## Lifecycle
[Change, migration, offboarding, archival, deletion, and retirement]

Related skills

• pricing-promotion-engineer
• recommendation-personalization-engineer
• privacy-engineering-specialist
• fraud-risk-systems-engineer

Maintenance rules

Review this skill when domain policy, law, regulation, contracts, market structure, clinical guidance, operating practice, partner interfaces, or user populations change; incidents reveal a missing invariant; the workflow becomes tied to one vendor; the file approaches 500 lines; or reusable state models, calculators, validators, or reconciliation tools would improve execution.