How Kitchain measures restaurant availability on delivery platforms

Kitchain (kitchain.co) measures availability for restaurant chains from outside, by reading the same public storefront page a customer opens, about every ten minutes, around the clock. Downtime is counted only inside the hours each location told the platform it would trade, and every interruption is confirmed by a repeat check before it counts. There is no point of sale integration, no marketplace login and no platform API access. Two downtime bases are published, one covering the whole panel and one excluding listings that were offline for almost the entire month.

What exactly is Kitchain counting?

Trading hours in which a customer could not place an order from a listing that said it was open. The check is a page load rather than a status query: the monitor opens the public storefront and records whether the listing is orderable at that moment. A single negative reading is not an incident. It is confirmed by a repeat check before it is counted, which removes transient failures on the platform’s own front end from the record.

Downtime outside stated trading hours is not counted at all, so a store that is closed on schedule contributes nothing. Overnight schedules are handled explicitly: for a location trading past midnight the early morning hours belong to its stated window, and an outage is capped at the end of that window rather than running into the next day.

Two categories are kept apart because they are different operational failures. An interruption is a listing that went dark inside its own trading window. A failure to open is a listing that never came online after its stated opening time. In July 2026 the UAE panel recorded 3,759 failures to open, of which 2,906 did not come online within the check window and 853 opened late. The Kuwait panel recorded 124.

What is a listing, and why is that the unit?

A listing is one restaurant location on one delivery platform. A branch that trades on four apps is four listings, and its availability is measured separately on each, because in practice it differs on each. That choice is the reason our per platform figures for the same city can sit a factor of twenty apart: in July 2026, UAE listings were unavailable for 2.94 percent of stated trading hours on Careem Food, 2.28 percent on Deliveroo, 1.81 percent on Talabat, 1.02 percent on Keeta and 0.74 percent on noon Food.

Aggregating by brand or by branch would hide exactly the thing the measurement exists to find. A brand can be healthy on three platforms and dark on the fourth, and the average across the four is a number that describes nothing anybody can act on.

What are stated trading hours, and why does everything hang off them?

Stated trading hours are the schedule the operator published to that platform for that location. They are a declaration, not an observation, and Kitchain reads them rather than deciding them.

That has a consequence worth stating plainly, because it cuts both ways. A location whose schedule says it trades 24 hours will accrue measured downtime for every hour it is dark, including hours it never intended to trade. A location whose schedule is narrower than its real opening hours will show a clean record while turning customers away outside the declared window. Neither is an error in the measurement. Both are the measurement doing what it says, which is comparing the storefront against the intention the brand gave the platform. The gap between those two things is a subject of its own, at scheduled hours and actual availability are different numbers.

Why are there two downtime bases, and when does each one apply?

Because a delivery platform’s catalogue contains listings that are no longer businesses, and including them answers a different question from excluding them. A location that has left a platform, or closed for a refit, or never opened after signing, frequently keeps its card in the catalogue and reads as offline every hour of every day. Those listings are real, a customer can find them, and they are not an operational problem anybody can fix from a kitchen.

So we publish both. The panel wide figure counts every listing we monitor. The trimmed figure excludes listings that were offline for more than 90 percent of the month, which is the profile of a card without a business behind it. The same trim is applied in every market so the numbers stay comparable across countries.

Read the first two columns across the first two rows and the country ranking reverses. On the panel wide base Saudi Arabia is worse than the UAE, 2.88 percent against 1.68. On the trimmed base the UAE is worse than Saudi Arabia, 1.26 percent against 0.77. Both statements are true, and they answer different questions.

The panel wide number answers “what does the catalogue look like to a customer”, which is the right base for a marketplace view and for anything about how much of a platform’s estate is orderable. The trimmed number answers “how reliably do trading restaurants stay online”, which is the right base for an operator comparing markets, benchmarking a brand, or setting an internal target. Using the wrong base is how a chain concludes that its Saudi operation is worse than its Emirati one when, among listings that are actually trading, the opposite holds.

What does the 90 percent exclusion actually remove?

In Saudi Arabia in July 2026, 5.6 percent of the panel was offline for more than 90 percent of the month, and removing that group cut the reported rate from 2.88 percent to 0.77 percent. The effect on incident length is larger still: the panel wide mean incident of 12 hours 24 minutes falls to 5 hours once those listings are excluded, so more than half of the headline duration was contributed by cards that were never coming back.

The comparable trimmed figure for the UAE panel in the same month was 2 hours 18 minutes. On a like for like basis, then, a Saudi interruption ran about twice as long as an Emirati one, which is a statement about live restaurants rather than about dead cards. The 31,357 trading hours lost across the Saudi panel that month include both groups, which is why the loss total and the rate have to be read against the same base.

What does this measurement not show?

A great deal, and the limits are as much a part of the method as the arithmetic.

It does not show cause. A dark listing looks the same from outside whether a branch paused itself, a tablet lost power, an integration failed, or the platform closed the store, and telling those apart needs the platform’s own signals, which we do not read. The distinguishing evidence available from outside is set out at paused by the branch or closed by the platform, and the reason a healthy internal system is not a visible listing is at storefront status and POS status are not the same thing.

It does not show money. Availability is measured, order volume and revenue are not, so lost hours are hours and any conversion to currency is somebody else’s assumption applied afterwards.

It is not a census. The panel is the set of restaurants Kitchain monitors, not a random sample of a national market, and platforms whose samples were too small to be meaningful are excluded from the platform level tables rather than reported thinly. Every published figure is an aggregate, and no individual restaurant or brand appears in any of them.

How often do the figures change, and how should they be cited?

Figures are refreshed as new full months close, and each report carries the date it was last updated. The July 2026 benchmarks were last updated on 26 August 2026. Kitchain reads more than 50 million storefront checks a month across a panel of 5,000+ restaurants, and the country reports state their coverage window and platform list at the top of each page.

The figures may be quoted with attribution. The three benchmarks are the UAE report, the Saudi Arabia report and the Kuwait report, and any comparison drawn between them should name which of the two bases above it is using.