The estate
Three interface generations living in one hotel estate
Every hotel we have looked inside runs some mix of these. None of them is wrong; the failure begins when a bridge is built assuming the wrong one is on the other end.
Batch and file exchange
- Nightly exports
- Flat files
- Shared folders
- Manual re-keying
Persistent legacy interfaces
- Socket links
- Message protocols
- On-premise adapters
- Interface licences
Cloud-era APIs
- REST endpoints
- Webhooks
- Event feeds
- Sandbox tenants
The bridging layers
- Vendor connectors
- Integration middleware
- Custom direct interfaces
The failure modes
How a rate, a room and a reservation drift out of step
Three flows carry the commercial weight of any PMS integration, and each fails in its own way. Rates fail by mapping: the update flows, but a rate plan on one side lands on the wrong product on the other, and a package sells at the wrong price until a guest complains. Availability fails by timing: the sync runs, but in windows, and inside the window a room sells twice — or worse, the direct channel shows closed while the platforms are still selling. Reservations fail silently: a booking that could not be delivered sits in a queue nobody monitors, and the first symptom is a guest at the desk with a confirmation the PMS has never seen.
What the three have in common is that none announces itself. The integration “works” in every demo and every quiet week, and fails exactly when volume, rate changes and full occupancy arrive together. That is why the design question is never only how the systems connect but how the connection fails: what is detected, what is queued, what is retried, and who is told — decided before the season that will test it.
The paths
Connector, middleware or direct interface — judged by their bad days
These are categories, not products, and every estate ends up mixing them. The comparison that matters is not the feature list but what each path does on the day it breaks.
| Option | Time to a first working sync | Behaviour when it breaks | Who you depend on afterwards | Fit for the odd system in the corner | Best for |
|---|---|---|---|---|---|
| The vendor’s own connector | Fastest where both ends are on the vendor’s supported list — much of the work is configuration. | A support ticket into a queue you do not control, with the diagnosis on their schedule. | The vendor, for scope, fixes and the connector’s roadmap. | None — the unsupported system stays unsupported. | Standard pairs of mainstream systems with no unusual requirements. |
| Integration middleware | Moderate — one platform learns each system once, then new connections get cheaper. | Queues, retries and a monitoring surface in one place — if they were configured, which is the work that gets skipped. | The middleware operator and the discipline of whoever maintains the flows. | Good — an adapter can usually be written for the strange one. | Estates with many systems and more integrations coming. |
| A direct custom interface | Slowest to first sync, because everything is built — and exactly fitted, because everything is built. | Whatever failure design was engineered in — which can be the best of the three, or nothing at all. | Whoever built it — which is why the documentation and handover matter more than the code. | The reason this category exists. | The critical flow the other two paths cannot carry properly. |
Our role
Who holds the pen on the rate-code mappings
Every PMS integration stands on tables that say which rate code equals which product, which room type equals which listing. Those tables are the integration; the code merely enforces them.
- 01
The estate, mapped before anything is bought
We establish which generation each system speaks, what its vendor actually offers for integration and on what terms, and where the honest ceiling is — a real-time flow or a scheduled one.
- 02
The bridge, chosen by failure behaviour
We recommend connector, middleware or direct interface per flow, on how each fails and who can fix it, and we build and monitor the ones that fall to us with detection and retry designed in from the start.
- 03
The mappings, written down and handed over
The rate-code and room-type mapping tables are delivered as documents with an owner and a change process, so the next rate season is an edit by your team rather than an archaeology project through our code.
The commercial meanings — what each rate plan is for, which product it should appear as on which channel — are decisions only the property can make, so the mapping tables carry your team’s signature and our engineering merely refuses to let the systems disagree with them.
Common questions
What buyers ask
Our PMS vendor charges an interface fee for every connection. Is that normal?+
Common, yes — many PMS vendors license interfaces individually, price API access by tier, or certify third parties for a fee, and the terms vary widely by product and contract. What matters is establishing the licensing position in writing before any integration is designed, because a build that assumes an endpoint your tier does not include fails commercially after succeeding technically. It is also a real input to the path decision: where per-interface fees are high, routing several flows through one middleware connection can change the economics — a question to price per estate, not to assume.
Why do reservations arrive in the PMS while availability never flows back out?+
Because those are two different interfaces, and your estate may only have one of them working — or licensed. Delivering a reservation into the PMS is an inbound flow; publishing availability and rates outward is a separate outbound one, often a different interface generation, a different licence line, and a different configuration effort. Estates frequently end up with the inbound path connected because the booking engine vendor needed it, while the outbound path was never finished because it belonged to nobody. The one-way arrangement looks alive from the outside, and it is exactly how a property oversells a full weekend.
Is middleware just one more thing to break between our systems?+
It is one more component, and whether that is a net gain depends on what it replaces. An estate with a handful of point-to-point links, each built differently and monitored nowhere, already has more failure points than the same estate with one middleware layer — the difference is that the middleware’s failures happen in a place with queues, retries and an alerting surface, while a point-to-point link fails wherever it fails and tells no one. The honest caveat runs the other way: middleware only delivers that benefit if its monitoring and retry behaviour are actually configured, and an unwatched middleware platform is simply a more expensive place to lose reservations.
In this cluster
An estate map that names each system’s interface generation and the integration path with the least dangerous failure days.
Put the problem to us
