Summary
Business Challenge
The business ran on someone else’s platform
Customer data, payments, and two years of bookings lived in a third-party product. Anything it could not handle, staff did manually.
About 730 manual tasks a year
Staff logged 545 tasks in nine months, from assigning rooms to tracking hotel release deadlines and adjusting commissions. Some carried direct financial risk: missing a release deadline could mean paying for empty rooms.
Failed payments and routine questions
Staff chased down missed payments and answered routine questions about balances, forms, and flights.
What We Did
1
Validated scope before committing
We mapped the client’s manual tasks and costly hotel release deadlines. A two-week checkout and payments pilot validated the estimate before the Phase 1 price was fixed. The website vendor developed in parallel against a mock API.
2
Built a shared platform
Checkout, portals, and staff AI tools use one API with per-user permissions, keeping availability and pricing consistent. Logged, replayable jobs prevent duplicate charges after a restart. Sensitive passport, medical, and bank fields are encrypted individually.
3
Automated retreat operations
The platform pairs travelers who name each other, manages waitlist offers, and sends form and flight reminders. Staff assign remaining roommates and can override pairings. Hotel release deadlines are flagged in advance. When a package changes, the platform recalculates prices, balances, refunds, and commissions, then routes adjustments for approval.
4
Connected payments and finance
Stripe checkout uses a 15-minute room hold to prevent overselling. Failed installments are retried, then reviewed by staff. No booking is automatically canceled. HubSpot logs bookings and messages, QuickBooks Online tracks finances by trip, and Wise handles payouts with finance approval.
5
Created role-based portals
Travelers manage payments and trip requirements. Leaders see their own rosters, rooming, and projected commissions. Staff manages operations and finance across the portfolio through five permission-based roles.
6
Added AI with human oversight
The assistant answers from live booking data within user permissions and hands unresolved questions to HubSpot. It scored 95%+ on 100+ pre-launch test questions. AI-drafted follow-ups require human approval. Staff query data through an MCP server with read-only, permission-scoped access. Sensitive data stays outside the AI layer.
7
Migrated and verified
Two years of booking history moved over with daily syncing until cutover and reconciliation on every import. A limited soft launch preceded wider release. Final acceptance requires the first production month-end close to reconcile to the cent across Stripe, Wise, the platform, and QuickBooks Online.
