Enquiries arrived by email, forms and WhatsApp in several languages; bookings lived in a notebook; the website was slow; nobody could tell which ad brought which booking.

Map the path enquiry → reply → booking → day of service → invoice; define what a booking is worth; decide what the site must show and what it must never claim without a source.

A frozen domain model (destinations, places, tours, departures, guides), publication gates so nothing ships half-empty, a fact contract (every number has a source), a value model for ads (revenue ≠ deposit ≠ profit).

Staff portal with unified inbox, pipeline, bookings with a ledger, work calendar; customer portal by booking number; static site factory with schema and redirects that keep click IDs.

IMAP mailbox inside the portal; website forms; Google Ads, GA4 and Search Console; offline-conversion upload ready behind a switch.

Tracking from form to booking; AI drafts in every language; two rounds of AI-written content had 113 factual errors caught by rule before publication.

Dozens of enquiries a month from several countries handled in one place; bookings tied to the ads that brought them; a generated site scoring 95–97 on Lighthouse.