Someone sees a beach on Instagram. Days later they search destinations, watch videos, compare hotels, change dates, talk to their family, look at flights, come back to hotels, and eventually book.
The average travel purchase journey can stretch for months. The travel industry averages a 71-day journey from initiation to final booking.
The traveler experiences all of this as one journey.
Inside a company the size of Expedia, things look different.
may have something to offer.
may have something to offer.
may have something to say.
may have an email scheduled.
may see another opportunity.
Each team can optimize its own interaction and still create a fragmented experience.
Because the traveler doesn't experience Expedia's organization.
To them, it's all Expedia.
So I started somewhere different: with the traveler.
One journey.
Different needs.
We modeled the journey through five broad stages:
Inspiration
Destination themes and travel motivations may matter most.
Shop
Current searches, dates, destination, party, and products under consideration.
Attach
Part of the trip has already been purchased.
In-Trip
The active itinerary and immediate traveler needs take precedence.
Post-Trip
The journey continues after the stay ends.
A traveler dreaming about Hawaii doesn't need the same thing as someone comparing three hotels tonight.
Someone who just booked a hotel doesn't need the same thing as someone landing tomorrow morning.
So the question wasn't simply:
What does Expedia know about this traveler?
What is this traveler trying to accomplish right now?
We already knew a lot
about the traveler
Expedia wasn't suffering from a lack of customer data.
We were building increasingly rich customer profiles by organizing signals that had historically lived across the business.
A lodging profile could describe property and amenity preferences.
A flight profile could capture patterns in how someone traveled by air.
Engagement with destinations could reveal affinities for themes such as beaches, skiing, nightlife, family activities or outdoor recreation.
Other signals could help us understand social patterns — whether someone tended to travel alone, as a couple, or with children.
Add loyalty, purchase history, searches, customer value and other behavior, and the result was an extraordinarily rich picture of the traveler.
Generative AI could synthesize those attributes into something much easier to understand:
This traveler frequently takes short domestic trips, typically travels as a couple, prefers upscale lodging, and shows strong affinity for beach destinations, food and nightlife. They tend to book lodging before airfare and have historically favored flexible flight options.
Useful intelligence.
But not necessarily useful right now.
A traveler with a strong historical affinity for beach vacations could currently be planning a ski trip with their children.
Their profile isn't wrong.
The relevance of the profile has changed.
The journey became
the contextual filter
Journey state helped determine which parts of that massive customer profile mattered at a particular moment.
Profiletells us who the traveler tends to be.
Journeytells us which parts of that understanding matter now.
And current behavior can make that context even sharper.
The clickstream knew what happened.
AI could tell us what it meant.
Consider a traveler planning a trip over several days.
Searches Miami for two adults. Browses beachfront hotels.
Returns to Miami. Changes the dates. Filters for four-star hotels and pools. Repeatedly views two properties.
Searches flights to Miami for the same dates.
Returns to one of the hotels and adds it to the cart.
To Expedia's underlying systems, that's a stream of searches, clicks, filters, product views and cart events.
An LLM can turn those events into something much more useful:
The traveler appears to be actively planning a Miami trip for two adults. They are comparing lodging and flight options, with the strongest interest in higher-end beachfront hotels with pools. One hotel has progressed from repeated consideration to the cart, suggesting increasing purchase intent.
The same reasoning can produce structured context another system can use:
{
"journey_phase": "SHOP",
"destination": "Miami",
"party": "2 adults",
"product_focus": ["lodging", "flight"],
"lodging_preferences": ["beachfront", "4-star", "pool"],
"purchase_intent": "high",
"leading_signal": "hotel added to cart"
}
Now every downstream experience doesn't have to interpret days of behavioral data.
More importantly, Expedia doesn't merely know a lot about this traveler.
It can understand which parts of what it knows matter to what the traveler is doing now.
Context changes
the decision
Now imagine several Expedia experiences competing for the traveler's attention.
A destination recommendation might be eligible.
Loyalty may have a promotion.
Flights may see an opportunity.
Lodging may want the traveler to complete the hotel booking.
All can be individually reasonable.
But only some are relevant now.
So the question becomes:
Given who this traveler is, where they are in their journey, and what they're trying to accomplish right now, what is the most useful thing Expedia can do next?
That's where customer intelligence becomes decisioning.
Profile
Tells us what we know.
Journey
Tells us what matters now.
Decisioning
Determines what to do about it.
The AI wasn't
the starting point
It's tempting to look at this system and see an AI problem.
It wasn't.
- We first needed to organize fragmented customer data into meaningful traveler understanding.
- We needed to model the journey.
- We needed to understand current behavior.
- And we needed to determine which context mattered to the decision being made.
Then AI became extraordinarily useful.
It could synthesize complex behavior into natural language a human could understand and structured context another system could act upon.
That's the idea behind Big Splash.
We don't start by asking:
What can AI do?
What does the system need to understand?
Because more data isn't necessarily more context.
And the intelligence of the model isn't the same as the intelligence of the system.