Phase 1: New Tours
We replaced our simplistic, error-prone location extraction with a sophisticated AI-driven model. The AI now reads itinerary descriptions like a human would, identifying and tagging locations with much higher accuracy. This happens automatically for all new imported tours, which end up as usual in the respective TimTam queue. Manual operators will find the same functionality behind the Refresh Map from Itinerary button in the Operator Dashboard. We've also added the ability to flag "no-location" days like transit days at sea, and introduced Location Logging so that if the AI finds a place that doesn't exist in our DB yet, it gets logged to bridge the gap and improve coverage over time.
Accurate locations mean better maps for customers, more relevant search results, less manual cleanup for our teams and most importantly, meeting the expectations of our travelers.
Phase 2: Existing Tours
The previous phase conquered new tours; Phase 2 brought that same AI precision to our existing catalog. We built and deployed a sophisticated outlier detection algorithm to identify tours with broken maps or mismatched itinerary locations, isolating 7.6k problematic tours for immediate AI intervention. We also refined the AI to ignore "adventure summaries" like last-day recaps to prevent map clutter. By running the extraction across the board, we achieved a 90% success rate in fixing identified outliers.
Our tour inventory has never been in better shape regarding location completeness and accuracy.
Phase 3: Streamlined Operator Dashboard
Phase 1 & 2 cleaned our history; Phase 3 simplifies the future. We overhauled the operator dashboard to make location management a faster, more intuitive experience, moving from a clunky two-step process to a single, streamlined UI with fewer buttons. The map is now permanently visible, providing real-time visual feedback during edits. We've replaced general "Tour Locations" with required Day Locations to ensure maximum itinerary accuracy, and added new transit mode options for River Cruise and Sailing days so operators can mark days as Transit without requiring a specific map pin.
The result is lower friction for operators and higher data quality for our travelers, with inventory that stays accurate from the moment it's updated.
Phase 4: Auto-Syncing Updates
The final piece of the puzzle. Phase 4 ensures that as itineraries evolve, our maps evolve with them automatically. If an imported tour's itinerary text changes by more than 5% at a day level, the AI automatically triggers a fresh location extraction. We've also implemented a sophisticated distance-and-angle check to catch errors before they reach travelers, flagging locations with a 200km minimum distance or 1.2x the average route distance, as well as sharp turns where the route angle is narrower than 60°. Any detected outliers or tours with ambiguous titles are automatically routed to the TimTam Content Queue for manual review.
This concludes the AI Location Extraction project. Our maps are now living data points that stay accurate even as operators update their tours, maintaining the highest quality standards on the marketplace.