← all systems
live · autopilotRiZE MapsOutbound2026 — present

Hotel outbound agent

Race-timed B2B outbound to hotels and resorts in F1 host cities: researched targets, verified contact routes, AI-drafted first touches, compliance gates and a human approval queue. It drafts and never sends on its own.

Founder · designed + built
PythonSQLiteClaudeGoogle SheetsGmail
~/systems/rize-hospitality
running
  1. Target hotelscsv · 118 rows
  2. Race calendarLV · ATX · MIA
  3. Verify routesdept · confidence
  4. Draft agentclaude · concept
  5. Guardrailsdnc · prior · opt-out
  6. Approval queuehuman · approve
  7. Send-ready+ follow-ups
tool agent gate / human record in flight
118
target properties
3
US race markets
14
compliance rules
0
unapproved sends
Problem

Hotels near F1 races are a natural buyer for premium track art (suite gifting, VIP player development, guest experiences), but the buyer is hard to find, the window around each race is short, and a careless cold email to a resort group you already have a relationship with burns it.

Architecture
  1. 01

    Research

    118 properties across Las Vegas, Austin and Miami, each with parent group, buyer team to find, recommended product and why it fits.

  2. 02

    Timing

    A race calendar drives when each market enters the queue, so outreach lands in the window when hotels plan race-week gifting.

  3. 03

    Verify

    Contact routes are verified by department and confidence before a target becomes draftable.

  4. 04

    Draft

    An LLM drafts first touches in a single CONCEPT format, with property-specific detail and no fake relevance.

  5. 05

    Gate

    Hard rules block do-not-contact, groups with an existing relationship, international markets pending legal review, and anything missing sender identity or an opt-out.

  6. 06

    Approve

    Every draft waits in a human approval queue. The system is draft-only by design and covered by tests.

Design decisions
  • →Draft-only, human-approved. Small, high-value B2B lists punish mistakes far more than they reward speed.
  • →Warm relationships route around the cold engine. Prior-org matching stops the system from cold-pitching a resort that already inquired.
  • →Experiments sized honestly: 8–12 prospects per strategy, judged on real buyer dialogue, not statistical claims a sample this small can't support.
Results

118 researched targets, 26 verified routes and a 14-property first pilot ready for approval ahead of the Las Vegas Grand Prix, with zero sends that skipped a human.

What I'd build next
  • +Log replies back into the target sheet to auto-score which strategy (CONCEPT vs. PROPERTY) earns buyer conversations.
  • +Expand to international race markets once legal review clears.
Next system
RiZE Maps: Content + paid growth engine
→