Enterprise Intelligence Platform
Transform your agency into a continuously learning intelligence system—capturing every signal, insight, decision, and outcome to drive smarter strategy and faster execution over time.
178 slides organized into 19 sections • Click any section to navigate
Enterprise AI transforms agencies from execution partners to intelligence partners
Projects that end
Insight in people's heads
Reactive execution
Headcount scaling
Intelligence that compounds
Insight in the system
Predictive action
Knowledge scaling
Leading analysts define the convergence of AI, orchestration, and automation
Business Orchestration & Automation Technologies
Adaptive Process Orchestration
We're implementing what leading analysts envision for enterprise automation
Six branded workflows that operationalize strategy into governed execution
Brief → Strategy → Content → Calendar
ImplementedPlanning → Optimization → Reporting
ImplementedDetection → Assessment → Response
In ProgressIngestion → Analysis → Visualization
ImplementedVoice → Planning → Creation → Approval
PlannedIntake → Response → Resolution
PlannedThe 8 core functions defined in the Paid Optimization Workflow Outline
Budget distribution across channels
ImplementedPerformance prediction using benchmarks
ImplementedUnified targeting recommendations
ImplementedStandardized naming conventions
ImplementedDetailed flighting calendar
ImplementedReal-time spend & KPI monitoring
ImplementedBudget rebalancing based on performance
ImplementedWoW analysis & strategic insights
ImplementedGoogle, Meta, LinkedIn, TikTok, GA4, CRM
Camunda v7 OSS engine with AI generation
MMM Allocator, Predictive Engine
Persona-based dashboards & DoA
CortexOne AI agents invoked via API • All published at cortexone.rival.io
ML forecasting
Bayesian optimization
Auto-execution
CAPI validation
Lift measurement
Unified read API
Resend integration
Twilio integration
Embedded LLM analysis
Content compliance
RAG preprocessing
Vector generation
Production-ready integrations with error handling, validation, and retry logic
OAuth 2.0 • Campaign CRUD
OAuth 2.0 • 9 Workers
OAuth 2.0 • 189 Tests
Token Auth • KB Ingester
n8n Signal Ingestion
Salesforce • HubSpot
Blue Silk AI • 150M+ Sources
Domain-Wide Delegation
Base Classes • Retry Logic
Rival Platform Foundation + Poetry Domain Layer
Claude-powered automation
Visual process orchestration
Unified signal taxonomy
Meta, Google, TikTok, LinkedIn, X
Generate complete BPMN processes from natural language using Claude
"Media planning workflow
with parallel approvals"
BPMN generation rules
+ validation
Complete process
+ diagram coordinates
Automated BPMN deployment integrated with GitHub
AI generates BPMN
BPMN validator checks
Git branch + PR
Auto-deploy to Camunda
processes/ (experiment) → packages/bpmn/ (promote) → CI/CD (deploy)
New processes start in processes/ for experimentation.
Once validated, they're promoted to the @rival/bpmn package for automated deployment.
Auto-commit, branch management, PR creation
13 rules ensure Camunda 7 compatibility
Seamless deployment to Camunda v7 engine
One-command QA environment setup
Real example: RIV-337 completed autonomously in a single Claude Code session
40-60% Effectiveness Improvement Through AI-Powered Routing
Automatic selection of optimal model based on task complexity
5-stage pipeline with feedback loop for continuous improvement
Production deployments with measured effectiveness
Runtime (LTS until Apr 2027)
Monorepo with 10+ packages
React web app with App Router
TypeScript backend with DI
BPMN engine with REST API
Enterprise authorization
Runtime API validation
Accessible UI components
End-to-end testing
Alpine-based database
Type-safe database access
Cache and sessions
Spring animations
Neo4j + pgvector
Containerized stack
Real-time voice AI
Conversational LLM
Speech-to-text
AI Avatar rendering
Automated vulnerability scanning
rival/ ├── apps/ │ └── web # Next.js dashboard ├── packages/ │ ├── api # NestJS REST API │ │ └── test/fixtures/ # BPMN test fixtures (CI/CD) │ ├── bpmn # Production BPMN (@rival/bpmn npm package) │ ├── workers # Camunda workers (5 planning) │ ├── authority-management # Enterprise Authorization │ └── integrations/ │ ├── core # Shared utilities │ ├── google-ads # Google Ads API │ ├── meta-ads # Meta/Facebook API │ ├── linkedin-ads # LinkedIn Marketing │ ├── tiktok-ads # TikTok Ads API │ ├── ga4 # Google Analytics 4 │ ├── crm # CRM integration │ └── google-workspace ├── functions/ # CortexOne AI agents └── processes/ # BPMN scratch/experimentation (local dev)
Secure, multi-tenant file storage with enterprise compliance controls
No service account keys in Cloud Run. Uses GCP Workload Identity for zero-credential access to GCS buckets.
Tamper-proof audit logs with cryptographic signatures. Every upload, download, and delete is immutably recorded.
Next.js 14 dashboard with enterprise-grade features
Full ARIA support, keyboard navigation
Content Security Policy, HSTS
Runtime validation with type-safe client
Smart forms, KB search, RBAC-protected routes
Interactive process diagram viewer
Mock data layer for development
7 role-specific views with proactive insights
Cmd+K with page detection + auth guardrails
Enterprise delegation of authority
AI-Powered Campaign Optimization Agents (GCP Python 3.13 Runtime) - All 6 Published ✓
Actions: ROAS forecasting, creative fatigue, audience saturation
✓ Published ($0.01)
Actions: CAPI signal quality, privacy-compliant validation
✓ Published ($0.01)
Actions: Auto-execute with budget guardrails & safety limits
✓ Published ($0.01)
Actions: Bayesian budget optimization, Media Mix Modeling
✓ Published ($0.01)
Actions: Geo-holdout, synthetic control, Bayesian lift
✓ Published ($0.01)
Actions: Unified read API for Google & Meta campaigns
✓ Published ($0.01)
11/24 functions published at cortexone.rival.io | 13 in development | API-invokable via BPMN workflows | $0.01/call pricing
How CortexOne Functions Think, Reason, and Act Autonomously
THINK - ACT - OBSERVE cycle enables multi-step reasoning with tool use
Risk-based thresholds ($5K budget, 50% change) trigger human approval
Fast rule-based (sync) or intelligent multi-step (agentic) modes
Complete audit trail with query hashing and trace correlation
Visual BPMN Modeler with CortexOne Marketplace Integration
bpmn.io + Rival Activity Palette
100s of AI Functions Available
Browse Marketplace
Select Functions
View Collection Cost
Save Collection
Use in Modeler
Tenant admins curate which marketplace functions are available to their process modelers
Poetry: World-Class Client Experience
Click persona to see task inbox:
Descope Identity Platform
Social OAuth + Magic Link + Enterprise SSO (7.5K MAU free)
ITSM-Style Ticket Queue for Support & Operations Teams
Filterable list with status badges, priority indicators, and SLA tracking
Full ticket view with description, assignee, timeline, and communication thread
Metrics cards showing queue health, response times, and resolution rates
Available at /operations/tickets • Support Agent persona • RIV-478
AI-Assisted Campaign Creation with Enterprise Controls
Campaign objectives, target audiences, industry verticals, platform channels - all editable by admins via /admin/settings
Context-Aware Intelligence with Auth Guardrails
Global shortcut
from any page
Route detection
intent classification
Persona-filtered
responses
Markdown + syntax
highlighting
Detects landing, login, dashboard, planning routes for tailored suggestions
Intent classification blocks auth-required questions from guests
720+ nodes, 665+ relationships, TOGAF ontology
Role-based knowledge filtering via graph
Dynamic questions based on page + auth state
Thumbs up/down with AI sentiment analysis
Context Examples: Landing page → "What is Poetry?" | Dashboard → "How is my ROAS trending?" | Planning → "What's the recommended budget for retail?"
Multi-Modal AI Interaction with Local LLM
Beyond Presence AI avatar with real-time lip sync
Audio-only with real-time bar visualizer
Traditional chat interface with markdown support
| LLM | Ollama (llama3.2) - Local |
| STT | Deepgram |
| TTS | Deepgram (aura-asteria-en) |
| VAD | Silero - Local |
| WebRTC | LiveKit Cloud |
Powered by: LiveKit Cloud + Deepgram + Ollama + Beyond Presence
Manager
Director
CMO
👍/👎
AI Sentiment
Notify User
40+ nodes including Signals, Causal AI, Patterns & Anomalies - Click "Data Scientist" to explore
Immersive visualization of Signals, Causal AI, Cross-Client Learning & 80+ relationships
Transform content creation from art to measurable science
Powered by ContentWRX methodology | Partnership with Colleen Jones / Content Science View Breakout →
Track which themes and archetypes drive business outcomes
Score content against brand pillars for voice compliance
Identify underserved themes, audiences, or formats
Self-improving recommendations that compound over time
25,000+ lines of AI-powered content intelligence | 3 Application Modes: Agency, Platform, Competitive
Deep User Modeling for Truly Personalized Experiences
Primary & secondary jobs, goals, success metrics, pain points for each persona
Individual users synced from PostgreSQL with expertise tracking and learning history
16 skills with prerequisites, learning paths, and certification tracking
Identify what users should know vs. what they currently know
Users can hold multiple roles - unified task inbox across all assigned personas
New persona for regulatory compliance review and content approval workflows
"What should I know as a Campaign Manager?"
"What skills do I need for Meta API integration?"
"What's my learning path to become GDPR expert?"
"What topics has Sarah recently viewed?"
"What's Alex's expertise level in OAuth?"
"What knowledge gaps do I have in compliance?"
AI Under Control + Industry Regulations Built-In
Poetry builds compliance into the platform from day one—
a strategic advantage for us and
a competitive edge for our clients.
18 Controls Across 5 Categories - Complete Autonomous Agent Oversight
Complete Framework: Every AI decision is logged, every model call is traced, every output is evaluated— automated governance that scales with your agents
BPMN-Driven Automated Evidence Collection & Gap Detection
Automated Compliance: BPMN-orchestrated evidence collection with real-time gap detection— compliance becomes a continuous process, not an annual scramble
Role-Based Views for Clients and Auditors
Dual Persona Design: Clients manage their compliance posture while auditors verify evidence— 108 unit tests ensure reliability across both portals
10 Camunda External Task Workers for Real-Time Compliance
Missing & stale evidence
Framework & overall
Email & Slack
Real-time metrics
Complete trail
Content scanning
Policy checks
External services
Service checks
Automated gathering
Always-On Compliance: Timer-triggered BPMN workflows continuously detect gaps, calculate scores, and alert— compliance issues found in hours, not audit cycles
SOC 2 Type 2 + Rival AI Governance - 120-150 Controls Target
Right-Sized Framework: 120-150 controls targeting SOC 2 Type 2 certification— inspired by RI AI CoE 214-control framework, adapted for commercial SaaS
7-Layer Zero-Trust Defense + 5-Gate CI/CD Pipeline | SOC 2 CC Aligned
SOC 2 Alignment: CC6.1 (Access) • CC6.6 (Encryption) • CC6.7 (Transmission) • CC7.2 (Monitoring) • CC8.1 (Change Management)
Default DENY • Explicit Allowlist • Complete Audit Trail
Audit Compliance: HMAC-SHA256 tamper-proof logs • SOC 2 CC6.1 • HIPAA Security Rule • GDPR Article 30
GCP Firewall + Cloudflare Zero-Trust: No Direct IP Access
IaC: infrastructure/kubernetes/cloudflared-config.yaml | GCP: rival-qa-allow-cloudflare-only, rival-qa-deny-direct-ingress
9 Critical Security Issues Identified & Fixed (January 2026)
PR: #553 | Methodology: Multi-agent security review + LLM-as-Judge validation | Evidence: .claude/memory-bank/evidence/cdd/RIV-550/
Unified GCP Infrastructure + External Services Cost Intelligence
PR: #633 | Knative service: cost-reporter | Schedule: 6AM/12PM/6PM/12AM ET
SEO → AEO (Agent Experience Optimization)
Optimize for
human clicks
Optimize for
AI retrieval + actions
Schema.org Product, Organization & LocalBusiness vocabularies in JSON-LD so AI reads meaning, not just words
SSR for crawlable content, clean canonical URLs, XML sitemaps, /llms.txt declarations
RFC 9309 robots.txt compliance, WAF-level enforcement, rate limiting & selective access
AI Preferences working group (aipref) for site-level machine-readable policy declarations
Real-time product APIs, structured commerce feeds enabling AI agents to browse & transact
AI referral traffic, LLM citation tracking, bot crawl analytics & AI-driven attribution
As AI shopping agents become the new top-of-funnel, Poetry positions brands to be discoverable, trustworthy, and actionable for both human users AND AI retrieval bots.
Rival Platform - Agentic SDLC Metrics ↑ 69% Complete
14 SubAgents automating the full software development lifecycle
Generated: 2025-12-28 | Data refreshed on demand via /dev-summary
Real-Time Performance Intelligence Powered by AI
Dashboard available at /campaigns/dashboard • Data refreshes every 15 minutes
What-If Analysis with Real-Time AI Predictions
Normal campaign performance baseline
Declining returns scenario simulation
Overspend/underspend pacing analysis
Single platform outage resilience
Holiday/event traffic surge modeling
Creative exhaustion prediction
Available at /campaigns/simulator • CortexOne functions power AI predictions • RIV-491
6,500+ Records for Demo, Testing & QA
State rules • Responsible gaming • GeoComply • HITL reviews
50+ institutions • Gainful employment • Lead gen consent
Seed command: pnpm --filter @rival/db seed • Located in packages/db/prisma/seeds/
Complete workflow from campaign inception to creative brief with AI + DMN governance
AI Proposes → DMN Decides → Human Approves → BPMN Enforces
Orchestrated workflow with DMN governance gates at each decision point
Process: plan-campaign.bpmn • Work Item: RIV-224
State machine controlling all campaign phases with message-based coordination
Process: campaign-lifecycle.bpmn • Master state machine
Simultaneous deployment to 4 ad platforms with verification and rollback
All platforms verified -> Status: ACTIVE -> Notify stakeholders
Verification fails -> Pause all platforms -> Log details -> Notify team
Process: launch-campaign.bpmn • Parallel gateway pattern with rollback
Tactical daily adjustments and strategic weekly reallocations
Adaptive Approval: Small changes auto-approve | Large changes require human review
Processes: daily-optimization.bpmn, weekly-reallocation.bpmn
Campaign state management with DMN-driven authorization
Processes: pause-campaign.bpmn, resume-campaign.bpmn, close-campaign.bpmn, emergency-stop.bpmn
Advanced AI-powered campaign optimization with predictive analytics and autonomous execution
Process: cortexone-optimization.bpmn • 4 CortexOne serverless functions
AI-powered sentiment analysis with priority-based routing and closed-loop notification
Process: feedback-workflow.bpmn • Closed-loop feedback with AI sentiment analysis
8-agent pipeline from brief intake to content calendar generation with parallel research phases
workflow-incident-create worker with priority routing to senior-engineers or incident-managers.
Process: social-media-content-workflow.bpmn • Styx Workflow Suite
8-agent hub-and-spoke pattern for multi-channel content production with Opus quality gate and configurable workspace output
Process: content-creation-workflow.bpmn • Styx Workflow Suite • ClaudeAgentWorker + WorkspaceProvider
6-agent linear workflow for automated ticket handling with intelligent escalation and SLA enforcement
Process: customer-service-workflow.bpmn • Styx Workflow Suite
9-agent data pipeline with collapsed sub-processes for collection, analysis, and visualization
Process: reporting-analytics-workflow.bpmn • Styx Workflow Suite • Wayfinder Pattern
6-agent event-driven pipeline for early risk detection, assessment, and coordinated response
Process: crisis-reputation-workflow.bpmn • Styx Workflow Suite • Spectre Pattern
109 specialized workers execute BPMN service tasks across 6 categories
All workers registered with Camunda 7 external task API
Policy-as-code governance with auditable decision logic
| Input | Rule |
|---|---|
| Privacy Framework | GDPR/CCPA required |
| Tracking Enabled | +30 pts if true |
| Client Tier | Min budget by tier |
Hit Policy: COLLECT (SUM)
| Tier | Range | Max Δ |
|---|---|---|
| Enterprise | $50K-$10M | 50% |
| Growth | $10K-$500K | 30% |
| Starter | $1K-$50K | 20% |
Hit Policy: FIRST
| Metric | Threshold |
|---|---|
| Confidence Score | ≥ 0.70 |
| Variance | ≤ 25% |
| Coverage | ≥ 80% |
Hit Policy: FIRST
| Check | Rule |
|---|---|
| Data Sources | 1st/2nd party only |
| PII Handling | Hashed required |
| Consent | Explicit opt-in |
Hit Policy: COLLECT
| Level | Pattern |
|---|---|
| Campaign | [Client]_[Obj]_[Date] |
| Ad Set | [Audience]_[Geo] |
| Ad | [Format]_[CTA]_v# |
Hit Policy: FIRST
| Budget | Approver |
|---|---|
| < $25K | Manager |
| $25K-$100K | Director |
| > $100K | VP/C-Level |
Hit Policy: FIRST
Location: packages/bpmn/poetry/decisions/ • 34 DMN files total
Industry-specific decision tables for regulatory compliance validation
pharma-fda-opdp-compliance.dmn
finra-2210.dmn
cms-marketing-guidelines.dmn
43 total DMN decision tables • Industry-specific compliance rules
Defense, Alcohol/Cannabis, and AI Governance decision tables
dod-contract-compliance.dmn
state-abc-rules.dmn
ai-act-transparency.dmn
All DMN tables evaluated at process runtime via Camunda 7 decision service
6-stage HITL compliance pipeline for defense & aerospace advertising
ITAR (International Traffic in Arms Regulations) - Controls export of defense articles on the USML. EAR (Export Administration Regulations) - Controls dual-use items on the CCL. Putting controlled data in a public ad = illegal export to the world.
Reviewers can choose REDACT instead of REJECT. Process ends, user removes controlled content, then starts a new process instance. Each submission is a separate audit record for DCSA compliance.
Process: defense-compliance-gate.bpmn • 5 DMN tables • 5 HITL gates • Work Item: RIV-268
Each stage is a DMN decision table with human escalation path
Target: Lockheed Martin, RTX, Northrop Grumman, Boeing Defense, Anduril, Shield AI
Complete workflow executed via Rival Functions API + Camunda 7 orchestration
poetry-media-mix → mmm-allocator (real API)
poetry-forecasting → predictive-engine (real API)
poetry-audience, poetry-framework, poetry-creative-brief
Generate mock DecisionProposal when API unavailable
Process Instance: 364acb4f-e08c-11f0-9d52-f63bc50e7e8b • Business Key: demo-e2e-final
Two complementary decision engines for different purposes
Purpose: Business rules & regulatory compliance
Audience: Business analysts, compliance officers
Editing: Visual table editor (Camunda Modeler)
Integration: Native BPMN (businessRuleTask)
Purpose: Authorization & access control
Audience: Developers, security engineers
Editing: Code editor (Rego language)
Integration: REST API or WASM (in-process)
DMN: packages/bpmn/poetry/decisions/ | OPA: packages/authority-management/policies/
Delegation of Authority (DoA) with OPA WASM for sub-millisecond authorization
Auto-route to correct approver based on spend amount
Temporary authority transfer with expiration
Complete decision log for compliance
Customer-specific policies via Rego
Most marketing platforms lack integrated Delegation of Authority. Rival provides enterprise-grade DoA with sub-millisecond authorization via WASM, enabling real-time policy evaluation without infrastructure overhead.
ADR: ADR-0003 (DoA) | ADR-0012 (WASM) | Policies: packages/authority-management/policies/
Visual AI Agent workflows integrated with Camunda BPMN
POST /webhook/research-analyze
POST /webhook/optimize-budget
Workflows: packages/n8n/workflows/ | Strategy: packages/workers/src/unified/strategies/n8n-strategy.ts
All personas use password: Demo2025!Secure
sarah.johnson@nissan-demo.com
mike.chen@nissan-demo.com
lisa.finance@nissan-demo.com
alex.thompson@poetry.ai
jamie.support@poetry.ai
dev.engineer@poetry.ai
morgan.kim@poetry.ai
process.modeler@poetry.ai
compliance@poetry.ai
tenant.admin@poetry.ai
platform.admin@poetry.ai
auditor@external-audit-firm.com
demo@rival.io
Source: apps/web/e2e/fixtures/personas.ts | Seed: packages/db/prisma/seeds/users.ts
Unified multi-channel notification service for all platform workflows
Source: packages/api/src/shared/notification.service.ts | Worker: packages/workers/src/cortexone/email-service.worker.ts
AI-forward agency intelligence with causal validation and cross-client learning
Proving causation, not just correlation
Zep/Graphiti bitemporal pattern
Network effect moat
Poetry's moat is NOT any single AI capability but the integration of three layers that competitors struggle to replicate: Causal AI (the "why"), Temporal Knowledge Graph (the "what"), and Workflow Orchestration (the "how").
Schemas: packages/knowledge-base/poetry/schema/ | Functions: cortexone-functions/
8 signal categories with real-time to quarterly capture frequencies
n8n: packages/n8n/workflows/ | Workers: packages/workers/src/ | Schema: signals-intelligence.cypher
DoWhy/EconML causal inference + 67% faster anomaly detection
Daily 6AM trigger → Parallel n8n collection (GA4, News, Media, Competitive) → Anomaly detection → Slack alerts
signal-aggregation.bpmn
Severity routing (Critical: 1h, High: 4h SLA) → Causal analysis → Human-in-loop investigation → Resolution
anomaly-response.bpmn
McKinsey Finding: 40% of ad spend is wasted without causal validation. Poetry's integration of DoWhy/EconML with the MMM Allocator provides "Validated ROI" badges on recommendations.
Functions: cortexone-functions/causal-validator/, cortexone-functions/anomaly-detector/
Network effect moat with privacy-safe aggregate patterns
Schema: packages/knowledge-base/poetry/schema/cross-client-patterns.cypher
5-phase framework from strategy to continuous improvement
Industry, Competition, Economics
COMING SOONBrand Audit, Customer Insight
COMING SOONCreative Brief, Program Design
COMING SOONMedia Mix, Forecasting, Launch
✓ BUILTDaily Optimization, Weekly Review
✓ BUILT
PRD: .claude/memory-bank/prds/mop-phases-1-3.md • Jira: RIV-541, RIV-542, RIV-543
AI-assisted strategic context gathering • Coming Q1 2026
DMN Governance Gate: Business Context Approval • Strategic Planner review required before Phase 2 • Causal validation of market assumptions
Jira: RIV-541 • BPMN: business-situation.bpmn (planned)
Deep customer understanding with causal validation • Coming Q1 2026
DMN Governance Gate: Insights & Objectives Approval • Account Director review required • Causal validation of behavioral insights • k-anonymity verification for cross-client patterns
Jira: RIV-542 • BPMN: segmentation-insights.bpmn (planned)
Creative strategy with cross-client pattern learning • Coming Q2 2026
→ Handoff to Phase 4: Approved creative brief triggers plan-campaign.bpmn
• Media Mix Allocation • Performance Forecasting • Campaign Launch
Jira: RIV-543 • BPMN: big-idea.bpmn (planned)
Three-layer continuously learning architecture • The "Strategic Brain" differentiator
The "Why" - Validates causation, not just correlation
The "What" - Bitemporal memory for complete history
The "How" - Privacy-safe aggregate patterns
Competitive Moat: Every new client makes the intelligence better for all clients. Only Poetry knows that "Pharma CTV campaigns with frequency >3.5 see 30% CTR decline after 21 days"
Schema: packages/knowledge-base/poetry/schema/ • 31 Cypher files
Comprehensive signal taxonomy feeding the Knowledge Graph
Schema: packages/knowledge-base/poetry/schema/signals-intelligence.cypher
From strategic insight to continuous optimization in one intelligent system
Phase 1 Beta
Business Situation
Phases 2-3 Beta
Insights + Big Idea
Full MOP GA
All 5 Phases Live
PRD: .claude/memory-bank/prds/mop-phases-1-3.md • Jira: RIV-541, RIV-542, RIV-543
Embedding Agency Expertise into Domain-Specific Language Models
Agency writing style
Brand voice generation
Expert explanations
Senior analyst reasoning
ADR: .claude/memory-bank/decisions/ADR-007-poetry-llm-fine-tuning-architecture.md
BPMN Orchestration + CortexOne Functions + GPU Training Compute
BPMN: packages/bpmn/poetry/processes/poetry-llm-fine-tuning.bpmn
Strategic benefits of combining BPMN governance with CortexOne agility
BPMN for governance. CortexOne for intelligent compute. Human-in-loop where it matters.
Functions run on optimal hardware - CPU, GPU, FPGA, or ASIC. Not locked to any cloud.
Workloads matched to best execution venue. GPU completes in 1/10th CPU time.
Invoke CortexOne functions directly. No infrastructure awareness required.
Processor-level encryption. Deploy on-premise, cloud, or managed.
90%+ cost/time reduction proven. 65x efficiency gains (STORM).
| Approach | Governance | Compute Efficiency | Hardware Flex | Hybrid ✓ |
|---|---|---|---|---|
| BPMN Only | ✓ Excellent | Limited (CPU) | None | ✓ |
| CortexOne Only | Limited | 65x gains | CPU/GPU/FPGA | API-based |
| Hybrid | ✓ Excellent | 65x gains | CPU/GPU/FPGA | ✓ Automatic |
STORM: 3,120 → 48 compute hours via intelligent hardware routing
25 files, 5,774 lines delivering the complete fine-tuning pipeline
poetry-llm-fine-tuning.bpmn
cortexone-functions/poetry-*/
cortexone-functions/training/
PRD: .claude/memory-bank/prd/PRD-poetry-llm-fine-tuning.md • Branch: feature/RIV-500-poetry-llm-fine-tuning
Fine-tuned models for agency-specific tasks powered by CortexOne intelligent routing
Marketing Voice
Brand Reputation
Investment Analysis
CortexOne intelligent routing: CPU/GPU/FPGA/ASIC • Not locked to any cloud
Training: QLoRA fine-tuning • Serving: Knative auto-scaling • Governance: BPMN orchestration
Continuous model improvement through user feedback on RAG responses
User flags inaccurate RAG response with AlertTriangle button
AI pre-labels sentences, user confirms and categorizes errors
JSONL export feeds DeBERTa-v3 fine-tuning pipeline
Training Architecture: User Feedback → JSONL Export (70/15/15 split) → DeBERTa-v3 Cross-Encoder → Improved Hallucination Detector
Components: HallucinationModal.tsx • API: /api/hallucination/detect • Jira: RIV-666
Every traditional agency relationship has the same problem.
When talent leaves, so do your insights. Every transition resets your learning curve.
Every campaign reinvents the wheel. Last quarter's learnings stay in someone's inbox.
You're paying premium rates for insights your competitors got last quarter.
What if your agency got smarter with every campaign?
Imagine an agency relationship where intelligence compounds over time.
Knowledge stays in people's heads
Knowledge lives in the system
Year 1: We learn your business. Year 2: The system knows patterns you haven't discovered. Year 3: Insights compound faster than any competitor.
The Learning Agency
Every campaign makes us smarter for your business. Every insight stays in the system. Every quarter builds on the last.
Powered by proprietary technology that captures, connects, and compounds every insight.
Real outcomes from compounding intelligence
"In Q3, Poetry identified a pattern across our campaigns that we'd never seen. It turned out our top-performing creative had a specific element that worked 3x better on mobile. That insight alone paid for the entire engagement."
Every interaction makes the system smarter
Data flows from every touchpoint. Signals from 8 categories feed the system continuously.
AI agents find patterns, validate causation (not just correlation), detect anomalies.
Recommendations flow to the right person at the right time. Human-in-the-loop approval.
Every outcome feeds the knowledge graph. The system remembers what worked—forever.
🔑 The secret: Every insight is connected to every other insight. Patterns emerge that no human could spot alone.
Three outcomes that matter most to marketing leaders
Optimize spend across channels with AI that learns your audience's patterns.
Know what content works before you publish. Learn from every piece of creative.
Dashboards that answer questions before you ask. Executive-ready insights.
+ Social Media Management • Crisis & Reputation • Customer Service → See all 6 solutions
How we measure success—and give your CFO ammunition
We don't just measure correlation—we prove causation with holdout tests and causal AI.
Marketing Mix Modeling that actually updates in real-time, not quarterly.
Signal loss recovery that works in a post-cookie world. iOS 14.5+ ready.
✓ Quarterly business reviews with executive dashboards
✓ Monthly ROI reports your CFO will actually read
The talent you'd hire if you could—augmented by intelligence that never forgets
Your strategic partner with full context on your business
Analytics expert backed by AI pattern recognition
Channel expertise enhanced by predictive modeling
5 specialized agents working 24/7 on your behalf
The difference: When your Account Director learns something, the entire system learns it. When they're out, the AI agents maintain continuity. Knowledge compounds—it never walks out the door.
The compounding advantage over time
| Traditional Agency | Poetry | |
|---|---|---|
| Knowledge Retention | Walks out with people | Lives in the system forever |
| Learning Curve | Resets with each campaign | Compounds exponentially |
| Pattern Recognition | Human memory limits | AI finds hidden connections |
| Reporting Speed | Weekly/monthly cycles | Real-time dashboards |
| ROI Proof | Correlation-based claims | Causal AI validation |
Year 3 with Poetry = More intelligence than Year 10 with a traditional agency
Your journey to compounding intelligence
30-minute conversation to understand your challenges and goals
90-day engagement on a focused initiative to prove value
Expand across channels and watch intelligence compound
No long-term commitment required. Prove value in 90 days, then decide if the relationship is right for you.
Concrete deliverables from your pilot program
Real-time visibility into performance across all channels
CFO-ready metrics proving value vs. your previous approach
At least 3 insights you didn't know about your audience
Your business intelligence captured and connected—forever
12-month plan for scaling the intelligence advantage
Account Director + Data Strategist + AI Agent access
The real deliverable: A foundation of intelligence that will grow more valuable every quarter.
Ready to build an agency relationship where every quarter makes you smarter?
Intelligence that compounds, not resets
ROI your CFO can measure
Prove value in 90 days
Powered by the Rival intelligence platform
Secure, Flexible Identity Management Powered by Descope
Passkeys, Magic Links, and biometric authentication eliminate password vulnerabilities
SAML 2.0 and OIDC support for Okta, Azure AD, OneLogin, and custom IdPs
Per-tenant identity configuration with isolated user pools and custom branding
Multiple secure options for every user preference
Seamless integration with your corporate identity provider
Industry-standard federation protocol for enterprise identity
Modern OAuth 2.0-based identity layer for web and mobile
Simple onboarding for enterprise customers
Customer submits SSO request with IdP details
We create SSO connection in Descope
SP metadata ↔ IdP metadata exchange
Validate SSO flow and go live
Isolated, secure identity management for each customer
End-to-End Workflow Demonstration
See how 7 AI agents collaborate through a BPMN-orchestrated workflow to generate a complete 8-week content calendar with human-in-the-loop quality gates.
Nissan North America - EV Awareness Campaign
| Client | Nissan North America |
| Goal | Brand Awareness |
| Duration | 8 weeks (Jan 29 - Mar 26) |
| Budget | $150,000 |
Tesla, Ford Mach-E, Chevy Bolt, Hyundai Ioniq, Kia EV6
BPMN-Orchestrated 7-Agent Pipeline
Industry Analysis and Audience Research run simultaneously, reducing total execution time.
Confidence-based gates (≥0.85 auto-approve) ensure human oversight on low-confidence outputs.
Full campaign strategy and 120 posts generated in under 20 minutes.
Validates and structures the campaign brief
{
"clientName": "Nissan North America",
"industry": "Manufacturing",
"targetPlatforms": ["instagram", "tiktok", "linkedin"],
"campaignGoals": "awareness",
"audienceDescription": "Auto enthusiasts aged 25-45,
eco-conscious consumers, first-time EV buyers...",
"contentPeriod": 8,
"competitors": ["Tesla", "Ford Mach-E", ...],
"budget": 150000
}
{
"structuredBrief": {
"clientName": "Nissan North America",
"startDate": "2026-01-29",
"endDate": "2026-03-26",
"totalWeeks": 8,
"platformCount": 3,
"isComplete": true,
"missingFields": []
},
"validationPassed": true,
"confidence": 0.6
}
38 seconds
Yes - Confidence 0.6 < 0.85 threshold
All required fields present, dates calculated
Industry Analysis + Audience Research run simultaneously
Duration: 2m 11s | Confidence: 0.6
Duration: 1m 50s | Confidence: 0.6
| Age Range | 25-44 |
| Interests | Technology, Trends |
| Values | Quality, Innovation |
| Pain Points | Finding reliable solutions |
Both agents completed in ~2 minutes total instead of ~4 minutes sequential. This pattern is enforced by the BPMN parallel gateway.
Creates core brand messaging themes
"Nissan North America delivers exceptional value"
"Building relationships that matter"
"Delivering real results"
| Formal ↔ Casual | 5/10 - Balanced |
| Serious ↔ Playful | 4/10 - Professional |
| Reserved ↔ Bold | 6/10 - Confident |
| Traditional ↔ Innovative | 7/10 - Forward-looking |
Confidence: 0.6 triggered manual review
Review Note: "Approved for Nissan Ariya EV campaign demo - messaging pillars look solid"
Platform-specific strategies and posting schedule
Focus: Visual storytelling
| Frequency | 5/week |
| Best Times | 9AM, 6PM |
| Content Types | Reels, Carousels |
Focus: Trend participation
| Frequency | 4/week |
| Best Times | 12PM, 7PM |
| Content Types | Short videos, Duets |
Focus: Thought leadership
| Frequency | 3/week |
| Best Times | 8AM, 12PM |
| Content Types | Articles, Polls |
Weeks 1-2: Brand introduction, awareness building
Weeks 3-5: Community engagement, user content
Weeks 6-8: Call-to-action, lead generation
Weekly content mix and theme distribution
| Day | Posts | Focus |
|---|---|---|
| Monday | 3 | Week kickoff |
| Tuesday | 2 | Educational |
| Wednesday | 3 | Mid-week engagement |
| Thursday | 2 | Behind-the-scenes |
| Friday | 3 | Weekend prep |
| Saturday | 1 | Lifestyle |
| Sunday | 1 | Community |
120 posts with captions, hashtags, and asset requirements
{
"id": "f43e6314-2bb2-4463-b345-fb279095f596",
"platform": "instagram",
"postType": "reel",
"scheduledDate": "2026-01-30",
"scheduledTime": "09:00",
"dayOfWeek": "Friday",
"weekNumber": 1,
"caption": "[REEL] Brand Excellence - Short-form video showcasing Brand Introduction",
"hashtags": ["#brandintroduction", "#NissanAriya", "#EVlife"],
"assetRequirements": { "type": "video", "duration": "15-60s", "aspectRatio": "9:16" },
"theme": "engagement",
"pillarReference": "Brand Excellence",
"callToAction": "Share your thoughts in the comments!",
"status": "DRAFT"
}
Confidence-based review triggers ensure quality
≥ 0.85
Auto-Approve
< 0.85
Human Review Required
0.6
REVIEWED0.6
REVIEWED0.6
REVIEWED0.6
REVIEWED0.6
REVIEWEDHITL gates ensure AI outputs meet quality standards before proceeding. In production, high-confidence outputs auto-approve for efficiency, while uncertain outputs get human oversight.
Complete workflow in ~17 minutes
15:51:57 - 15:52:35 (38s)
15:52:35 - 15:54:46 (2m 11s)
15:54:46 - 15:56:39 (1m 53s)
16:03:24 - 16:06:17 (2m 52s)
16:06:17 - 16:08:41 (2m 24s)
| Total Duration | 16 min 44 sec |
| AI Processing Time | ~10 minutes |
| HITL Review Time | ~7 minutes |
| Posts Generated | 120 posts |
| Posts/Minute | ~7.2 posts |
A human content strategist would typically need 40-60 hours to create a comparable 8-week content calendar with this level of detail.
~200x Faster
CIB seven (Camunda 7 fork) manages workflow execution
Total BPMN processes deployed across all solutions
AI agents implemented as external task workers
HITL review gates for human approval
Activity History (Path Taken): StartEvent_SocialMedia → Start (4ms) Task_BriefIntake → Brief Intake Agent (37.6s) Gateway_ResearchSplit → Parallel Split Task_IndustryAnalysis → Industry Analysis Agent (130.9s) ┐ Task_AudienceResearch → Audience Research Agent (110.5s) ├ Parallel Gateway_ResearchJoin → Parallel Join ┘ Task_MessagingPillars → Messaging Pillars Agent (112.8s) Task_ReviewPillars → HITL Review (7s) Task_Strategy → Strategy Agent (172s) Task_ReviewStrategy → HITL Review (15s) Task_ContentPlanning → Content Planning Agent (45s) Task_ReviewContent → HITL Review (7s) Task_ContentCalendar → Calendar Agent (0s) Task_FinalApproval → Final Approval (15s) Task_GenerateOutput → Google Sheets Export (60s) EndEvent_Complete → Complete
Enterprise-grade AI workflow orchestration
BPMN workflow orchestration with external task pattern for AI agent integration
AI agents with tool use, web research, and structured output generation
Content calendar exported to Google Sheets for team collaboration
Authorization policies for multi-tenant content access control
Modern React frontend with TypeScript API backend
Kubernetes orchestration with serverless scaling for workers
Transform content creation efficiency
Multi-channel content production with brand consistency
Brand Voice → Editorial → Newsroom → Executive → Press-to-Social → Editorial Review
Newsroom, Executive, and Press-to-Social run simultaneously
Brand Voice Review + Final Content Approval
Press Release, LinkedIn, Instagram, Twitter, Executive Blog
Hub-and-spoke pattern with parallel content generation
Press releases, news articles, announcements with journalistic standards
Thought leadership posts, executive memos, LinkedIn content
Transforms PR content into platform-native social posts
Nissan Ariya EV Launch - Multi-channel assets
Two complementary workflows for end-to-end content automation
From strategic planning to content production in under 25 minutes
See the full platform in action
See the complete platform with your own campaign data
Run a 30-day pilot with one brand or campaign
Enterprise rollout with custom integrations
Demo executed: January 29, 2026
Social Media: 71299725-fd2a-11f0-8853-4e8806c1e7d6
Content Creation: ca03c49c-fd2f-11f0-8853-4e8806c1e7d6
{poetry}
The visual identity system for Poetry and all branded materials across the Arc Machina ecosystem.
Version 2.0 — January 2026The Poetry logo is the word "poetry" wrapped in curly braces, rendered in a light serif-italic typeface.
Purple gradient system on a near-black foundation.
Central to Poetry's visual identity.
linear-gradient(135deg, #7B2FBE 0%, #9B4DCA 40%, #C084FC 70%, #E9D5FF 100%)
linear-gradient(180deg, #0D0D0D 0%, #1A1A2E 50%, #2D1B4E 100%)
linear-gradient(135deg, #F3E8FF 0%, #E9D5FF 30%, #DDD6FE 60%, #C4B5FD 100%)
linear-gradient(90deg, #7B2FBE, #9B4DCA, #C084FC, #E9D5FF, #7B2FBE) height: 4px
Host Grotesk for headlines. Inter for body. Serif italic for logo.
Poetry is part of the Arc Machina startup studio ecosystem.
Built inside a startup studio ecosystem, our ventures move faster, integrate deeper, and scale intelligence, not overhead.
iOS (Swift/SwiftUI) + Android (Kotlin/Jetpack Compose)
Offline-First • Real-Time WebSocket • Biometric Auth • Push Notifications • Home Screen Widgets
Native Apps with Shared Backend Infrastructure
Swift / SwiftUI
Kotlin / Compose
Socket.io at /api/mobile/stream
Live Performance Data with WebSocket Updates
Swipe Gestures + Offline Queue for Uninterrupted Decision-Making
AI-Powered Monitoring with Severity-Based Grouping
Delta Sync Protocol with Optimistic Locking
Local SQLite cache
Compare versions
Source of truth
Enterprise-Grade Mobile Security Architecture
Glanceable Intelligence + Proactive Alerts
ONE app binary that dynamically brands itself per tenant
Rival is B2B2B. Organizations like Poetry serve multiple tenants like Nissan and Toyota. Every user now sees their tenant's brand — colors, logo, fonts, tagline — instead of hardcoded Rival indigo.
Same APK, Different Brand
Login State Machine
White-Label Architecture Overview
Core Pattern: TenantThemeManager
// TenantThemeManager.kt object TenantThemeManager { val theme = mutableStateOf( TenantTheme.Default ) fun loadTheme(config, ctx) { theme.value = TenantTheme .fromBranding(config.branding) // Cache to SharedPreferences cacheTheme(ctx, theme.value) } }
Two Endpoints, Two Auth Levels
Built with Defense in Depth
// RivalApiClient.kt private val authInterceptor = Interceptor { chain -> val builder = chain.request() .newBuilder() getAuthToken()?.let { builder.addHeader( "Authorization", "Bearer $it" ) } getTenantId()?.let { builder.addHeader( "X-Tenant-ID", it ) } chain.proceed(builder.build()) }
Premium First-Time Experience
The “Wow Moment” — Generic to Branded in 1.5 Seconds
Pre-Configured Onboarding via Shareable Links
app.rival.io/invite/Xk9mP2invitations table
POST /api/mobile/invites
GET /invites/:code (public)
/settings/invites
/settings/invites with email input, role selector, pending invite table, copy-link, and status badges (Pending / Used / Expired).Dev → QA → Production pipeline with infrastructure-as-code
Docker Compose with 20 services • Single docker compose up
Live data from poetry-standard cluster • Captured Feb 7, 2026
| POD | IMAGE | CPU | MEMORY |
|---|---|---|---|
| api | rival-qa-docker/api:latest | 2m | 295Mi |
| web | rival-qa-docker/web:latest | 1m | 166Mi |
| cib7 | cibseven:run-2.1.0 | 3m | 657Mi |
| postgres | pgvector:pg17 | 20m | 70Mi |
| neo4j | neo4j:5.15 | 9m | 502Mi |
| redis | redis:7-alpine | 10m | 3Mi |
| cloudflared (x2) | cloudflared:2024.12.2 | 8m | 35Mi |
| TOTAL (8 pods) | 53m | 1,728Mi | |
Managed services for data tier • Knative autoscaling • Multi-node HA
| Component | QA | Production |
|---|---|---|
| PostgreSQL | In-cluster (pgvector) | Cloud SQL |
| Redis | In-cluster pod | Memorystore |
| Neo4j | In-cluster pod | Neo4j Aura |
| App Services | Static Deployments | Knative (autoscale) |
| Image Tags | :latest / :sha | :stable |
| VMs | Spot (70% savings) | On-Demand (SLA) |
| Registry | rival-qa-docker | rival-prod-docker |
Single PostgreSQL instance with pgvector • 4 databases • 3 users • 3 schemas in rival DB
public (app), eams_audit (audit), knowledge_base (KB)
Multi-tenant foundation • Organization → Tenant → User hierarchy • 115 models across 21 domains
Enterprise-grade 12-step wizard for agentic workflow creation, testing & deployment
Each step calls real backend APIs — zero simulations, zero hardcoded data
Four pillars powering the Solution Builder experience
Full lifecycle: dashboard, detail view, edit, and delete with confirmation
13 violations eliminated across Tier 3 Facades and Forbidden Patterns
End-to-end flow from wizard UI to deployed solution with auth at every layer
Delivered across 3 PRs with full SDLC orchestration and CDD compliance
Platform Changelog & Version Tracking
Current Version: 2026.02.9 | Last Release: February 8, 2026
8 releases in one day: 2026.02.2 – 2026.02.9
February 7, 2026 • Security Pipeline Release
| Area | Change | Impact |
|---|---|---|
| Pipeline | 7-gate security scanning based on policy2control | Enterprise-grade security |
| Security | GCM authTagLength: 16 for createDecipheriv | CWE-347 vulnerability fixed |
| Dependencies | @isaacs/brace-expansion → 5.0.1 | SCA vulnerability resolved |
| Scripts | license-scan.sh, dast-scan.sh added | Gates 5 & 7 implementation |
| Version | Date | Type | Key Changes |
|---|---|---|---|
| 2026.02.9 | Feb 8, 2026 | Security | Encrypted push tokens, optimistic locking, mobile onboarding |
| 2026.02.7 | Feb 8, 2026 | Feature | Report scheduling backend + UI, cron poller, connector sync |
| 2026.02.5 | Feb 8, 2026 | Feature | Premium white-label onboarding, brand reveal, invite system |
| 2026.02.2 | Feb 8, 2026 | Feature | Rival Mobile (iOS + Android): 10 features, 4 backend modules |
| 2026.02.1 | Feb 7, 2026 | Security | 7-gate security pipeline, GCM auth fix, license scanning |
| 2026.01.7 | Jan 25, 2026 | Security | Docker Trixie images, Snyk CVE fixes, KG Explorer |
| 2026.01.6 | Jan 20, 2026 | Feature | Social platform APIs, Google Workspace clients |
| 2026.01.5 | Jan 18, 2026 | Feature | Google Workspace integration, OAuth/SA auth |
Format: YYYY.MM.release
CHANGELOG.md
Enterprise-grade access to organization-wide data via service account impersonation
⚠️ Requires each user to authenticate
✓ No user interaction required
Index all docs, emails, and files for AI-powered search
eDiscovery, retention policies, regulatory compliance
Automated data ingestion without user interaction
Aggregate insights across all users and teams
Interactive visualization of Neo4j knowledge entities with gap detection and AI ideation
Campaign, Client, Signal, SemanticEntity types
Find disconnected clusters and isolated nodes
Suggest bridges between disconnected entities
Inspect node properties and relationships
Fail-Closed Services, Least Privilege IAM, Key Rotation & Operational Safeguards
PR: #947 | 12 files modified across API, Terraform, Cloudflare | GCP Security Advisory Response
Three Pillars: Metrics, Logs, Traces — Unified with AI Governance
View Full PRD | ADR-0007 ACCEPTED | GOV-011 through GOV-018 | SOC2 CC7.2
From QA to rival.io — Phased Rollout for Enterprise Readiness
A comprehensive 7-phase, 5-week production deployment plan bringing the full Rival platform to a dedicated GCP project with HA infrastructure, observability, and enterprise SLOs.
Dedicated GCP Project • Regional HA • Cloudflare Edge
7 Phases over 5 Weeks — Each Phase Has Exit Criteria
Day 1: Launch-Critical • Day 2: Feature-Flagged Expansion
@rival/shared feature flagsOptimized for Early Stage • Scales with Demand
| Component | Cost |
|---|---|
| GKE Autopilot (Regional) | $450-650 |
| Cloud SQL HA (4vCPU/15GB) | $320-380 |
| Redis HA (5GB) | $140-175 |
| GCS + Registry + Build + Egress | $40-70 |
| Cloudflare Pro + Descope | $20 |
| Observability (in GKE) | $50-100 |
| Total Baseline | $1,050-1,500 |
31 Tickets across 5 Sprints
View Full PRD | 10 Terraform Modules | 7 Day 1 Integrations | $1,050-1,500/mo
This section documents the original 8 core functions defined in the client's "Paid Optimization Workflow Outline" spreadsheet. These are the baseline requirements against which all development is measured.
| # | Function | Original Description | Status | Implementation |
|---|---|---|---|---|
| 1 | Media Mix Allocation | Budget distribution across channels (Meta, YouTube, Search, Programmatic, TikTok, LinkedIn, X) based on objectives, audiences, seasonality, and media approach | IMPLEMENTED | packages/workers/src/planning/media-mix.worker.ts |
| 2 | Media Forecasting | Predicts expected campaign performance using industry benchmarks, platform data, and historical results (CPM, CPC, CPE, CPV, CPL, ROI) | IMPLEMENTED | packages/workers/src/planning/forecasting.worker.ts |
| 3 | Audience Identification | Analyzes channel targeting capabilities and produces unified targeting recommendation (interest groups, demographics, job titles, retargeting pools, lookalikes) | IMPLEMENTED | packages/workers/src/planning/audience.worker.ts |
| 4 | Campaign Naming Generator | Creates consistent, searchable campaign names following unified naming structure | IMPLEMENTED | packages/workers/src/planning/framework.worker.ts |
| 5 | Campaign Flighting | Transforms forecast and mix allocation into detailed flighting calendar | IMPLEMENTED | packages/workers/src/planning/framework.worker.ts |
| # | Function | Original Description | Status | Implementation |
|---|---|---|---|---|
| 6 | Daily Optimization | Real-time spend/KPI monitoring, overspend/underspend detection, automated budget recommendations | IMPLEMENTED | packages/workers/src/poetry/daily-optimization.worker.ts |
| 7 | Weekly Reallocation | 7-day performance review, budget rebalancing, learning phase awareness | IMPLEMENTED | packages/workers/src/poetry/weekly-reallocation.worker.ts |
| 8 | Weekly Performance Snapshot | WoW analysis, creative performance signals, strategic insights email | IMPLEMENTED | packages/workers/src/poetry/snapshot.worker.ts |
All 8 functions require integration with:
| Category | Implemented | Remaining | Coverage |
|---|---|---|---|
| Planning Flow | 5/5 | 0 | 100% |
| Monitoring Flow | 3/3 | 0 | 100% |
| Total | 8/8 | 0 | 100% |
The goal is to build an automated agentic workflow for optimizing paid media campaigns. This system will ingest user constraints, historical data, and live performance metrics to recommend budget allocations, forecast performance, assist with audience targeting, flighting, and ongoing optimization (daily and weekly).
Version 1.0 (MVP) focuses on 8 core agents for market launch with recommendation-based workflows. Enhanced capabilities including predictive intelligence, incrementality testing, and autonomous execution are planned for subsequent releases (see Section 2.1 Release Roadmap).
The system is composed of 8 Core Functions or "Agents" that handle specific parts of the paid media lifecycle:
Goal: Operational paid media workflow with core planning and optimization
| # | Agent | Function | Priority |
|---|---|---|---|
| 1 | Media Mix Allocation | Budget distribution across channels | P0 |
| 2 | Media Forecasting | Performance prediction using benchmarks | P0 |
| 3 | Audience Identification | Targeting recommendations | P0 |
| 4 | Campaign Framework | Flighting & naming conventions | P0 |
| 5 | Creative Brief | Asset management & assignment | P0 |
| 6 | Daily Optimization | Real-time monitoring & recommendations | P0 |
| 7 | Weekly Reallocation | Budget rebalancing | P0 |
| 8 | Weekly Performance Snapshot | Strategic reporting | P0 |
Execution Mode: Recommendation-only (human approval required for all changes)
Goal: Add predictive capabilities and advanced measurement
| # | Agent | Function | Business Value |
|---|---|---|---|
| 9 | Predictive Performance Engine | ROAS forecasting, fatigue prediction | Prevent 30-50% performance drops |
| 10 | Incrementality Testing Orchestrator | Causal measurement, geo tests | Identify 20-40% wasted spend |
| 11 | MMM-Driven Allocator | Marketing Mix Model integration | 5-15% efficiency gains |
Goal: Enable autonomous optimization with human oversight
| # | Agent | Function | Business Value |
|---|---|---|---|
| 12 | Autonomous Execution Engine | Direct platform API execution | 11% uplift, 30+ hrs/wk savings |
| 13 | Privacy-First Measurement Manager | Server-side tracking, Conversion APIs | Future-proof for cookie deprecation |
Safety Controls: Max change limits, anomaly detection, 15-min human veto window
| # | Agent | Function |
|---|---|---|
| 14 | Cross-Channel Attribution Manager | Multi-touch journey mapping, Shapley value attribution |
| 15 | Audience Learning System | Auto-expand/contract audiences based on performance |
| 16 | Dynamic Creative Optimization (DCO) Engine | AI-generated creative variants |
| 17 | Competitive Intelligence Monitor | Competitor spend and creative tracking |
Scope: Version 1.0 (MVP) - All agents below are targeted for initial market launch
Goal: Determine how budget should be allocated across channels based on constraints and historical data.
Goal: Forecast performance outcomes based on the allocated mix using industry benchmarks.
Goal: Identify and recommend target audiences using platform targeting capabilities.
Goal: Structure the campaign flighting and naming conventions.
Goal: Manage creative assets and their assignment.
Goal: Monitor daily spend and KPIs, making real-time adjustments.
Goal: Rebalance budgets weekly to hit monthly targets efficiently.
Goal: Summarize weekly performance and provide strategic insights.
Enterprise client portal with:
V2: Predictive Performance Engine, Incrementality Testing, MMM-Driven Allocator
V3: Autonomous Execution Engine, Privacy-First Measurement Manager
V4: Cross-Channel Attribution, Audience Learning System, DCO Engine, Competitive Intelligence
Research-backed performance targets based on 2024-2025 industry data.
| Metric | Industry Baseline | Poetry Target | Improvement |
|---|---|---|---|
| Time to build media plan | 8-12 hours | <30 min | 96%+ reduction |
| Planning completion rate | 40-60% | >80% | 50%+ improvement |
| Recommendation acceptance | N/A (manual) | >70% | AI-enabled |
| Client NPS | 25-35 (agency avg) | >50 | 50%+ improvement |
| Metric | Industry Average | Poetry Target | Improvement Goal |
|---|---|---|---|
| Search CTR | 6.11% | 7.5%+ | +23% |
| Search CPC | $4.22 | <$3.80 | -10% |
| Search Conversion Rate | 7.04% | 8.5%+ | +20% |
| Display CTR | 0.46% | 0.55%+ | +20% |
| Metric | Industry Average | Poetry Target | Improvement Goal |
|---|---|---|---|
| CTR | 1.49% | 1.85%+ | +24% |
| CPC | $0.40-$0.65 | <$0.55 | -15% |
| Conversion Rate | 8.25% | 10%+ | +21% |
| CPM | $5-$15 | <$10 | Optimized reach |
| Metric | Industry Average | Poetry Target | Improvement Goal |
|---|---|---|---|
| CTR | 0.84% | 1.0%+ | +19% |
| CPM | $9.16 | <$8.00 | -13% |
| Engagement Rate | 5.96% | 7.0%+ | +17% |
| Video Completion | 60-70% | >75% | +10% |
| Metric | Industry Average | Poetry Target | Improvement Goal |
|---|---|---|---|
| CTR | 0.35-0.65% | 0.8%+ | +40% |
| CPC | $5.39-$8.00 | <$5.00 | -25% |
| Conversion Rate | 6.1% | 7.5%+ | +23% |
| InMail Open Rate | 52% | >60% | +15% |
| Metric | Industry Research | Source | Poetry Target |
|---|---|---|---|
| Marketing Automation ROI | 544% average | Nucleus Research | 600%+ |
| Time Savings | 300+ hours/year | Salesforce | 400+ hours |
| Campaign Performance | 25-40% improvement | McKinsey Digital | 35%+ |
| Manual Task Reduction | 60-80% | Gartner | 75%+ |
| Decision Speed | 3x faster | Forrester | 4x |
Document Version: 2.1 | Last Updated: 2025-12-22 | Status: Active - Research-Backed KPIs Added