Decision Orchestration Workflow
7 phases · 32 steps · 6 loop-backs · 8 AI agents · 3 decision personas
P0: Daily brief
P1: Signals
P2: Command card
P3: Decision
P4: Execution
P5: Outcome
P6: Memory
AI agents
Decisions
Loop-backs
Phase 0: AI daily brief NEW
0.1 Generate daily brief
AI compiles overnight signals, pending decisions, urgent tasks, closing campaign dates, KPI status, and new opportunities into a single morning report
Command tab
0.2 Customise brief focus
User selects focus areas: Market Signals, Campaigns, Analytics, Demand, Schedule. Brief adapts to show relevant content only.
0.3 Surface pending actions
Show decisions pending, total potential upside, urgent actions. Acts as an assistant reminding the user of approvals and closing dates.
0.4 KPI health check
Flag if main KPIs (Cover Value, Function Yield, occupancy, ROAS, direct %) are tracking above or below targets. Highlight deviations early.
0.5 Share & export brief NEW
Send brief by email to GM, Revenue Manager, department heads. Schedule daily/weekly/monthly delivery. Download as PDF. Recipients receive full intelligence digest without platform login.
View Intelligence?
YES: Navigate to Command cards
NO: Brief logged, check later
Phase 1: Signal ingestion
1.1 Collect raw signals via APIs
PMS, third-party channels, Google Ads, Meta Ads, interstate travel, FX rates, Google Trends, events, competitor rates, politics, world economy indicators
Signals tabIntegrations tab
1.2 Manual signal input NEW
User can manually add signals: local intel, event tips, staff feedback, owner observations. Merged into signal feed alongside API data.
1.3 Filter, classify, score
Type (airline, trend, event, FX, competitor, political, economic), impact (low/med/high), confidence %. Connect as many data sources as possible.
1.4 Contextualise against memory
Cross-ref with past decisions, brand guidelines, seasonal patterns, historical outcomes from Memory
Memory tab
Signal meets threshold?
NO: Log to monitoring feed → Loop back 1.1
YES: Proceed to Phase 2
Phase 2: Command card generation
2.1 Select strategic persona NEW
Strategy mode: Conservative, Balanced, or Aggressive. Shapes the card's priorities, risk tolerance, and recommended actions.
2.2 Assign specialist agent
Route to correct intelligence agent based on signal nature and persona mode
Preemptive Intelligence
Opportunity detection
Tradeoff Negotiator
Conflict arbitration
Performance Alert
Urgent fixes
2.3 Build command card
Observation, Interpretation, Opportunity, Proposed Strategy, Strategic Tradeoffs, Expected Impact (revenue, bookings, Cover Value, timing)
2.4 Attach RAG data sources
Cite verified data: event calendars, Google Trends, market data reports, historical patterns. Zero hallucination policy.
2.5 Risk assessment + institutional memory
Risk level + mitigation. Reference past similar decisions: "Last time you approved this, occupancy held at 91%"
Guardrails pass?
Brand risk, budget limits, confidence, irreversible check
NO: Modify or escalate → Loop back 2.3
YES: Queue card
2.6 "Create your own" card NEW
User types a question or decision and AI generates a full command card in the same structured format. Manual card creation path.
Phase 3: Decision card
3.1 Present command card to user
Full card: observation, interpretation, opportunity, strategy, tradeoffs, expected impact, data sources, institutional memory, risk, persona context
Command tab
3.2 Chatbot: Ask about strategy NEW
User can ask questions about the card before deciding. "What if we wait 2 weeks?" "Show me the downside scenario." AI responds in context of the brief.
User decides (4 paths)
APPROVE and EXECUTE
ADJUST & APPROVE → edit terms, then execute
DECLINE & LEARN → preference logged
DECLINE → Loop back Phase 1
3.3 Log decision to memory
Record timestamp, signal, brief, user choice, feedback. All decisions (including declines) train the system.
Memory tab
Autonomy mode? (set in Settings)
SIGNAL: AI monitors and alerts, cycle ends
OBSERVE: AI drafts the execution plan
CO-PILOT: Auto-execute within guardrails
Phase 4: Execution orchestration
4.1 Generate execution plan
Break strategy into steps: what to do, who does it (agent), timeline, expected outcome date, success metrics.
Execution tab
4.2 Connect via APIs to execute NEW
Push campaigns to Meta Ads, Google Ads, email platforms. Rate changes to channel managers. API connections for direct execution (Co-Pilot mode).
4.3 Assign execution agents
Campaign Optimizer
Budget, bidding
Ads Performance
Quality, creative
Campaign Architect
New campaigns
CRM Agent
Email, loyalty
Rate Agent
Pricing
4.4 Track execution status NEW
Action Ledger lists every action with status: Queued, In Progress (% complete + agent name), Completed. Live controls: Pause, Cancel, or Rollback any in-flight action.
Execution tab
Checkpoint passed?
NO: Pause, revise → Loop back Phase 2
YES: Continue
High-risk threshold?
YES: Escalate to user → Loop back Phase 3
NO: Complete execution
Phase 5: Outcome and reporting
5.1 Track execution metrics
Monitor actual vs projected: revenue, ROAS, occupancy, Cover Value, direct share, commission reduction. API pull from ad platforms, PMS, analytics.
Analytics tab
5.2 Generate outcome report NEW
Summarise the full card: original signal, strategy chosen, execution plan, actual results, key learnings. Pilot Performance Report aggregates predicted vs actual across decisions.
Results tab
5.3 Decision Lineage and attribution
Over/under performance analysis. Every decision traced as Decision Lineage: Signal → Card → Decision → Execution → Result. Root cause if missed.
Flywheel tab
5.4 Save to venue folder NEW
Archive complete card + execution + outcome into venue's knowledge folder. Builds permanent institutional record. Mark as Realised, Pending, or Missed.
Memory tab
5.5 Push results via APIs NEW
Feed performance data back to ad platforms (Meta, Google). Update campaign optimisation based on actual results. Close the execution loop.
Phase 6: Memory and learning
6.1 Feed outcomes to memory
Push all outcome data into venue memory. Results, learnings, what worked, what didn't. This compounds intelligence over time.
Memory tab
6.2 Update decision preferences
Adjust user model: approval rates, risk preference, market priorities, average decision time, preferred agent types.
6.3 Update revenue patterns
Avg approved deal, highest ROI market, peak opportunity window, most approved type. Guest preferences: LOS, sweet spot Cover Value, top conversion drivers.
6.4 Update company brain NEW
Essential and repeatable learnings: brand rules, seasonal patterns, market-specific playbooks. Knowledge base grows with every cycle.
6.5 Save to decision history
Full timeline: Approved/Declined/Suppressed with outcomes, projected vs realised revenue, agent used, feedback logged. Exportable.
6.6 Suppress / amplify patterns
Low-value types appear less. High-value patterns get proactive monitoring. Anti-noise mechanism keeps signal quality high.
Cycle complete
Return to Phase 0 with enhanced intelligence
Continuous cycle with compounding intelligence
7 phases · 32 steps · 6 loop-back paths · 8 AI agents · 3 decision personas · API-driven execution · Every cycle makes the system smarter