Is it normal for a restaurant to go offline on a delivery app?
Restaurant operators usually ask this before they ask anything else, and the honest answer has two halves that point different ways. Yes, it is common: across the listings we monitor, a typical one loses close to a full trading day every month, and roughly two thirds of listings lose something. No, that does not make it acceptable, because the same measurement shows a third of listings losing nothing at all in the same month, on the same platforms, in the same cities. Common and unavoidable are different claims.
How common is it, in numbers?
Across our UAE panel in July 2026, the average listing lost 9.5 trading hours in the month. That is one full trading day, per listing, per platform, in one market.
The distribution matters more than the average. Around a third of listings lost nothing at all. A small group lost more than twenty hours each. So the average describes almost nobody: most listings sit near zero, and a minority carries most of the loss.
If your branches are in that minority, the useful conclusion is not that this is normal. It is that something specific is happening to those branches that is not happening to the others.
Is my platform worse than the others?
Probably yes, and by more than most operators expect. In the same market, in the same month, the share of stated trading hours lost ranged from under one percent on the best performing platform to close to three percent on the worst.
Incident length varies just as widely and in a way percentages hide. On some platforms interruptions are frequent and minutes long. On others they are rare and run for half a day. Two platforms can post a similar percentage and represent completely different operational problems, and the fix for each is different.
Should a well run restaurant expect zero?
No, and expecting zero leads to the wrong response. Some interruptions are legitimate: a kitchen genuinely overwhelmed, an ingredient genuinely out, a deliberate pause during a staffing gap. Those are the system working.
What a well run restaurant can expect is that every interruption is one somebody chose, and that it ends when they choose. The problem is not the existence of downtime, it is the share of it that nobody chose and nobody knew about, which on most estates is the majority of it.
How would I know whether mine is the ordinary kind?
By whether you can name the cause of your last one. Operators with a healthy pattern can say what happened and when it ended. Operators with an unhealthy one describe a soft week and a suspicion.
The test is uncomfortable and quick. Take the last outage anybody remembers and ask how you found out. If the answer is that a franchisee phoned, or that the numbers looked wrong afterwards, then you do not have a downtime problem you can size, and the first thing missing is not a fix but a measurement.
Do the platforms think it is normal?
They document it thoroughly, which is its own answer. Talabat’s API carries a set of closure reason codes including a technical problem code. Deliveroo names a forced closure the partner cannot reverse. Careem has a state only Careem can lift. Keeta documents that a suspended store stays hidden until reactivated by hand.
Those are not accident reports, they are designed states. Platforms build closure into the product because closure is expected. What none of them publish is any rule about who carries the lost trading time, which tells you where they consider the cost to sit.
What makes it stop being normal?
Repetition on one branch, or a pattern tied to a mechanism rather than to a night. Those are the two shapes that separate ordinary noise from something worth escalating.
A branch that goes dark eight times in a month by the same mechanism is not having bad luck. That pattern is also the only thing that changes a commercial conversation, because a single incident is an anecdote and an account manager treats it as one.
Kitchain (kitchain.co) exists to make that distinction available, since the difference between ordinary and not is a question about frequency, and frequency cannot be judged from the incidents you happened to notice.