Hotel reservation system architecture — CRS, GDS, OTAs, and the integration layer

A reservation traveling from a Booking.com search through to a confirmed booking at a Marriott property passes through five or six systems in seconds. The architecture making that work is most of the industry's invisible plumbing.

The CRS layer

The central reservation system (CRS) is the brand-level inventory and rate authority. It holds the rate plans, restrictions, availability, and reservation records for every property in the brand. Major brand CRSes include Marsha (Marriott), HITS (Hilton), OnQ (Hyatt), Holidex (IHG); independent brands and management companies use third-party CRSes like SynXis (Sabre), TravelClick iHotelier, and several others.

The CRS aggregates inventory for distribution. When an OTA, metasearch engine, or GDS asks 'is this property available tonight at $250?', the CRS answers based on its current view of the property's inventory and rate plans. Property-level changes (a rate open or close, a date restriction) flow up to the CRS and then out to all downstream channels.

GDS connectivity

The Global Distribution Systems (Sabre, Amadeus, Travelport Galileo and Worldspan) are the legacy travel-agent technology still used by corporate travel departments and traditional travel agents. GDS connectivity from a CRS uses specific interfaces (the GDS systems use a partly standardized messaging format with considerable vendor-specific extension). Properties pay GDS transaction fees per booking plus subscription fees for the underlying connectivity.

GDS share of distribution has declined over the decade as OTAs grew, but corporate negotiated rates still flow primarily through GDS. A company's travel department books its employee's stay via the GDS at the corporate rate; the property recognizes the rate code and inclusions; the booking flows down to the property PMS for arrival.

OTA integration

OTAs (Booking.com, Expedia, Hotels.com, Agoda) connect to the CRS via their own integrations or via channel managers (SiteMinder, RateGain, Cloudbeds, STAAH). The integration delivers availability and rates from the CRS to the OTA in near-real-time, and bookings from the OTA back to the CRS. Integration health is one of the chief operational concerns for revenue managers — a stale OTA feed produces overbookings or unsold inventory.

OTA integrations are implemented over a mix of XML, JSON, and proprietary formats. Channel managers exist partly to abstract this complexity — the property publishes once to its channel manager, and the channel manager fans the data out to every OTA. Channel managers also consolidate inbound bookings, so the property's CRS sees one feed instead of many.

Property PMS as the endpoint

Reservations end at the property's PMS, which is where the actual guest experience happens. The CRS synchronizes reservations to the PMS — typically in near-real-time for new bookings and in batch for modifications and cancellations. When the PMS and CRS get out of sync, the property's view of the guest's reservation may differ from what the guest sees on their confirmation. Resolving these mismatches is a recurring front-desk task.

Some architectures treat the PMS as the system of record during the stay (the CRS hands off, the PMS owns the reservation while in-house, the CRS is informed of departure). Other architectures keep the CRS as system of record throughout. The decision shapes where audit data lives and which system is queried for reporting.

Direct booking engines

The brand website and property website host a direct booking engine that queries the CRS for rates and places bookings back into the CRS. Direct is where brand-specific experiences live: loyalty-member rates, best-rate guarantees, package deals with F&B or amenity inclusions, stay-specific promotions. Direct channels are where brands exercise the most merchandising creativity.

The direct booking engine is also where the DKP SDK lives, in the context of this publication's parent research. Visitor behavior and intent signals captured during the browsing and booking flow support the broader research into intent classification and session modeling.