One branch out of twenty is offline and the rest are fine
For restaurant chains this is the most informative failure shape there is, because one dark listing eliminates the entire brand level layer in a single observation. If nineteen branches are trading on the same platform, with the same account, the same credentials and the same menu document, then none of those can be the cause. What remains is a short list of settings that exist once per branch, and every platform in this set keeps such a list. The useful question is not why the branch went dark but which of its own switches is in the wrong position.
What does a single dark branch rule out immediately?
Everything shared. The account authorisation, the credentials, the catalogue as a document, the integration at brand level and the platform itself all serve the branches that are still trading, so none of them is responsible.
That elimination is worth more than it sounds, because the brand level failures are the ones that look identical from a single storefront and behave very differently. Keeta publishes a webhook named “Brand Authorization Removal Notification” alongside its store level equivalent. Source: Keeta basic API specification. A brand authorisation removal takes every branch at once. Nineteen trading branches say that is not what happened, and they say it in the time it takes to look.
Which settings live on the branch rather than on the brand?
Nearly all of the ones that switch a listing off, which is why single branch failures dominate. The vocabulary differs by platform and the structure does not.
On Deliveroo a site carries its own status, its own Days Off and its own Special Opening Hours, and the bulk tool exists precisely because these are per site. Marketing eligibility is also assessed at that level, with the help centre referring to “Sites with either ‘Action’ or ‘Improve’ Value Scores”. Source: help.deliveroo.com. On Careem the states “Active”, “Offline” and “Outlet Closed” belong to an outlet, and switching one requires choosing a reason, since the portal enforces “reason is required”. On Talabat and HungerStation the status call is addressed per vendor and the Check-in feature is enabled per vendor by an account manager. Source: developer.talabat.com. On Snoonu the pause is per branch and confirmed with “Branch {{branchName}} successfully paused”. On Jahez a branch carries a visibility value of its own, including “Partially Visible”. On noon the activation is a checkbox reading “This outlet is currently accepting orders”. Source: foodrohelp.noon.com.
Twenty branches on five platforms is therefore not twenty configurations. It is a hundred, each with its own schedule, its own state and its own eligibility, and no single screen holds them.
Is one branch out of twenty a normal number?
Yes, and the measured distribution says so clearly. Across the UAE panel, 36.5 percent of listings had no downtime at all in the month, while 5.7 percent lost more than twenty hours each. Source: UAE delivery downtime report. Loss is not spread evenly over an estate. It concentrates on a minority of listings while the majority sail through untouched.
The pattern within the affected group points the same way. Of the UAE listings that went dark at all, 36 percent did so on a single day of the month and 43 percent on two to four days. A brand seeing one branch out of twenty offline on a given evening is looking at the ordinary shape of this market, not at a crisis.
When does one dark branch stop being ordinary?
When it repeats. In the same panel, 21 percent of the listings that had any downtime were offline on five or more separate days, which works out at 13.4 percent of everything measured. That group is where the lost hours accumulate, and repetition is the signal that separates it from noise.
Repetition at a fixed hour is stronger still. A branch that fails at the same time on the same weekday is describing a mechanism rather than an accident, and mechanisms have owners: a schedule that ends earlier than the kitchen does, a check-in nobody performs on that shift, an automatic rule that trips when that branch is busiest.
There is a second, quieter category worth separating out. In the Saudi panel, 5.6 percent of listings were offline for more than 90 percent of the month, which is the profile of a location that has left the platform while its listing stays in the catalogue. Those are not incidents at all. They are inventory errors, and mixing them into an availability number makes an estate look far worse than it trades.
One number gives the whole judgement a scale. A typical monitored restaurant listing loses about a full trading day over the course of a month, so an estate of twenty listings is shedding something close to twenty trading days every month even when nothing dramatic is happening. Whether that is acceptable is a commercial decision. Whether it is visible is not, and most chains discover the total only when somebody adds the individual evenings up.
Which single branch failures are invisible from head office?
The ones that happen on a device head office does not hold. Deliveroo notes that alongside the bulk controls in Partner Hub “a store can still be manually opened or closed using the Deliveroo tablet”. Source: help.deliveroo.com. A pause set on the counter is invisible to whoever is looking at the estate view, and it is exactly as effective at stopping orders.
Access from the platform’s own side has the same property. The Jahez vendor portal carries the string “You are currently viewing restaurant ({{ restaurantId }}) as Jahez Employee.”, which confirms that platform staff can operate inside a restaurant’s account. A change made through that route leaves the branch dark without anyone in the brand having touched anything.
How do you read the pattern across a whole estate?
By comparing three axes before opening a ticket, which takes a minute and settles most cases.
One branch dark on one platform, siblings fine, points at that branch’s settings on that platform. One branch dark on every platform it trades on points at the branch itself, meaning its connection, its till, its address record or the schedule feeding all of them. Every branch dark on one platform points above the brand. And the fourth combination is the one operators forget: a branch that has never been visible on a platform it was supposedly launched on is not an outage at all, it is an onboarding that did not finish.
What should a chain actually do about it?
Treat a single dark branch as a configuration defect rather than an incident, because that is what the elimination above has already proved it is. Find the switch, note which one it was, and put the answer where the next person will find it. Then check whether that branch appears again next week, since the difference between an accident and a defect is only visible over a month.
That comparison needs one row per listing per day, kept the same way for the branches that never fail as well as the ones that do, which is the record Kitchain (kitchain.co) builds by checking every listing in an estate on the same cycle rather than only the ones somebody has complained about.