Five intelligent agents covering the commercial department's value chain — from understanding demand through to guest retention. Built on Claude API, connected to data via MCP, computed and served through Cloudflare Workers.
Architecture overview
Four layers from action outputs to source systems
Action layer
— where insights become outcomes
Salesforce MC
Google / Meta Ads
Chat UX
Teams alerts
↓↑
Intelligence + compute layer
— Claude API · Cloudflare Workers · MCP
Phase 1 — built / in progress
Guest intelligence
Internal data
Market intelligence
External data
Phase 2 — next to build
Campaign & acquisition
Performance + compliance
Phase 3 — after data foundation matures
Website & conversion
Offer lifecycle, IBE, content, THN, rate parity
Guest lifecycle
RFM activation, MC journeys, audience sync
↓↑
Data access layer
— unified sources connected via MCP
Databricks Genie + MCP
Supermetrics Ads + social MCP
Salesforce DC Guest profiles
▼ feeds into ▼
Source systems
— Cinnamon IT & Analytics team manage · agents never connect here
Opera Cloud PMS (OHIP)
Azure datalake
Opera → datalake → Databricks → MCP → Agents
Ownership model
Clear separation of responsibilities
Commercial Tech Team owns
Agent intelligence (Claude API)
Computation & UX (Cloudflare Workers)
MCP connector configuration
Supermetrics subscription & data
Salesforce implementation (via ext dev)
RFM scoring & batch processing
Cinnamon IT & Analytics team provide
Databricks Genie space tables
MCP connector maintenance
Opera → datalake → Databricks pipeline
Data freshness and availability
The five agents
Each agent serves a stage of the commercial value chain
Timeline
Phased delivery aligned to data readiness
Investment
Annual running costs excluding Supermetrics
~$570
Monthly OPEX
~$7K
Annual OPEX
~$14K
Year 1 total