Airlines need to add another layer to the shopping framework
to determine the shopper's identity and intent prior to computation, according
to a white paper published by by Travlr ID and Futures Lab.
The white paper notes a "turning point" in the
airline retail system, with airlines globally processing more than 100 billion
shopping queries each data, according to data from the International Air
Transport Association. That comes in part from metasearch sending replicated
queries across hundreds of suppliers, corporate reshopping bots repricing
itineraries and generative AI agents exploring fares, according to the white
paper. IATA projects automated requests will make up more than 70 percent of
offer generation by 2027.
"Each of these automated interactions triggers the same
expensive sequence of fare construction, rule evaluation and tax calculation as
a genuine booking attempt," according to the white paper. "The result
is a continuous drain of CPU cycles and network capacity. Airlines are, in
effect, paying to serve machines that will never travel."
Airlines' shift away from the traditional EDIFACT
distribution model has compounded this problem. In a white paper published last
month, IATA said the global distribution systems have been a buffer for
airlines in handling high volumes of availability requests, "partially
shielding the airline's systems from direct load." That is not the case
with New Distribution Capability, however.
"This API-based, real-time request/response model means
that each search, such as a round-trip query, can generate dozens of calls to
various airline systems," according to the white paper. "For
instance, a calendar shopping request for one month might result in around 450
requests…to explore all combinations of outbound and inbound dates. The cost
and feasibility of handling these requests falls directly on the airline."
IATA has shifted the way it measures airline shopping away
from "look-to-book" amid this increasing automation, instead looking
at "offer-to-order," which measures how many offers displayed end up
with orders, as well as "CPU-to-order," which measures how much each
booking costs in computation. The white paper, however, suggest airlines should
adopt a "profile-to-order" approach, in which requests are sorted by
the likelihood of an actual purchase before an offer is generated.
With that, airlines could determine whether a request merits
an offer or something that requires less computation, such as a cache or light
offer, or should be throttled. Airlines could set governance tiers within
that—for example, giving corporate and duty-of-care agents full computing
access while limiting responses to public and anonymous channels—or set up
other models, such as contracted computing quotas with distributors, according
to the white paper.
"In this model, identity clarifies who may request,
intent clarifies why the request exists and consent clarifies how much
computation is authorized," according to the white paper. "The result
is a retail ecosystem that is both efficient and transparent."
The white paper includes a call to action for the industry,
including the creation of a "neutral, non-commercial consortium"—with
participants including airlines, GDSs, travel management companies, AI
developers and regulators—to help evolve profile-to-order into an
"industry-recommended practice," the establishment of a
"multi-actor pilot" to measure computation reduction and formalized
interoperability guidelines between profile governance and NDC and One
Order.
"Through these
steps, profile-to-order becomes not only a technological improvement but a
foundation for ethical, efficient and agent-ready travel retail,"
according to the white paper. "The industry’s next leap is not a new
standard but a new mindset. Computation itself becomes a governed asset—one
that must be managed with the same discipline as safety, emissions and
financial reporting."