Check-In When the Internet Drops
A hotel does not stop at 23:00 because the connection did. Most front desk systems fall into four levels of offline behaviour — and the difference between level two and level four is the difference between a queue and a normal night.
What actually happens at the desk
It is 23:40. Three guests are waiting, one has a booking the system cannot see, the connection dropped twenty minutes ago and the receptionist is holding a phone with a screenshot of an email confirmation. This is not a rare technical event — it is a Tuesday in a hotel with a domestic connection.
Every hotel software vendor now says “works offline”. The phrase covers four completely different behaviours, and only the last one lets your staff keep working normally.
From “error” to “normal service”
Ask which one your system is at before you sign. Most marketing describes level four and delivers level two.
Level 1 — nothing
The screen shows an error or an endless spinner. Nothing can be read and nothing can be written. This is more common than vendors admit, and it is what a cloud-only PMS delivers the moment the line goes down.
Level 2 — read-only cache
The last-synced data is visible in the browser, so staff can at least answer “do we have this booking?” — but they cannot check anyone in. Bookings get written on paper and re-entered later, which is where errors enter the folio.
Level 3 — local writes, manual reconciliation
Check-ins work on the machine at reception and are stored locally. When the connection returns, somebody has to decide what to do with the data — often by comparing two lists by hand. Better than paper, still a bad night.
Level 4 — local-first, automatic reconciliation
Every write goes to a local database first and is synced in the background when the connection returns. The desk works normally, the room list is accurate, and the conflict rules are the software’s problem, not the receptionist’s.
Why level four is hard to build
“Offline support” sounds like a feature flag. It is closer to a rewrite, because a system that assumes a server is always reachable has to be redesigned around assumptions that no longer hold.
- The database moves. A local store — SQLite, for instance — becomes the source of truth on the machine, with the server as the thing it eventually agrees with.
- Writes have to survive a crash. If the machine loses power mid check-in, the record has to be either fully there or not there at all. That means journalling, not hope.
- Operations must be repeatable. When the connection returns, the same check-in may be sent twice. Every operation needs an identity so the second attempt is recognised as the same one, not a duplicate guest.
- Clocks disagree. Two machines, one booking, different system times. The reconciliation rules have to be written down and deterministic — “last write wins” is a decision with consequences.
- Staff need to see the state. A pending queue, a clear indicator, and a manual retry button. Software that silently holds unsynced data is worse than software that refuses.
- Some things cannot be offline at all. A channel manager is a live connection by definition, and anything that talks to a tax authority has to be designed for deferred transmission rather than pretending the authority is local.
Five minutes that tell you the truth
Ask for this during a demo, not after installation. Any vendor can refuse; only one of the answers is reassuring.
Unplug the network cable, then check a guest in
Not “look at the screen” — complete a check-in, assign a room, and open a folio. If the button greys out, you are at level 2.
While still offline, restart the machine
This is the test almost nobody runs. Data that lived only in memory is gone, and you find out whether the writes were real.
Plug the cable back in and watch what happens
Does it sync by itself? Is there a visible queue? Does it ask a human to resolve a conflict, or does it decide silently and get it wrong?
Ask what happens with two machines at the same time
Two receptionists, one connection drop, the same room. The answer is a rule, and the rule should exist before the situation does.
Ask which operations are explicitly not available offline
A vendor with a written list is telling you the truth. A vendor who says “everything works” has not tested it.
What your staff need to see
The engineering can be perfect and the night can still go badly, because the person at the desk has no idea whether their work has been saved. Three interface details decide that:
- An honest indicator. Not a subtle grey dot — a line that says “offline, 4 items waiting to sync”, visible from a metre away.
- A list they can open. Which bookings, which folios, at what time. If a guest disputes something the next morning, that list is the evidence.
- A retry they can press. When the connection returns but the sync has not started, staff should not have to call anybody.
And a procedure on paper for the one case software cannot solve: two devices, one room, both offline. It happens rarely, and it is the story that gets told about your hotel for years.
What people ask next
Isn’t a cloud system always online anyway?
No. “Cloud” describes where the data is stored, not how reliably the line behaves. A domestic fibre connection, a router reboot, a provider outage or a building works crew with a digger can each take you offline. The question is what happens next.
Can a browser-based system work offline at all?
Partly — service workers can cache a page and store some data, but a browser tab is a fragile place to hold a night’s front desk operations. Systems that take offline seriously usually run as a desktop application with a real local database.
What about card payments when the line is down?
That depends on your terminal and your acquirer, not on the PMS. Many terminals support offline authorisation up to a limit, with the risk of declined transactions later. Ask your bank what your floor limit is before you need it.
Do we still need a backup if it works offline?
More than ever. Local-first means the authoritative copy of last night’s work may be on one machine at reception. A nightly backup of that database is not optional.
How do we test our own setup?
Once a quarter, on a quiet afternoon: unplug the connection, check a guest in and out, restart the machine, plug it back in, and watch the sync. Ten minutes, and it turns an emergency into a routine.
Want to pull the cable on ours?
The front desk demo runs with generated data. Ask us for a live walkthrough with the network disconnected — it is the only demo worth having.
