How many delivery apps should one restaurant be on?
Restaurant chains usually answer this question by comparing commission rates, which is the smallest of the numbers involved. The real unit is a listing: one branch on one platform, with its own schedule, its own status vocabulary, its own promotion tool and its own account manager. A twenty branch chain on four apps is running eighty listings, and availability on them differs enough that the same brand can be reliable on three and quietly broken on the fourth. The right number is the number you can actually service.
What is the unit of cost when you add a platform?
The listing, not the platform, and this is why the arithmetic surprises people. Availability, prices, promotions and delivery area are all set per listing and behave differently per listing, so a branch trading on four apps is four independent operational objects rather than one restaurant with four sales channels. We measure availability this way for exactly that reason, and it is the reason our platform figures for the same city sit so far apart.
Aggregating across the four hides the thing you added the platform to find out. A brand can be entirely healthy on three platforms and dark on the fourth, and an average across all four describes a restaurant that does not exist. Anybody presented with a single group wide availability figure is looking at a number that cannot be acted on by anyone.
How different are the platforms, in numbers rather than in impression?
Far enough apart to change the decision. In our July 2026 UAE panel, 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. That is a factor of four across five platforms in one market and one month. Mean interruption length varied in the same panel from 1 hour 50 minutes to 3 hours 34 minutes, and the platform with the shortest incidents is not the platform with the fewest.
Markets move the answer as well. The panel wide figure was 1.68 percent in the UAE, 2.88 percent in Saudi Arabia and 0.54 percent in Kuwait, and the Saudi number falls to 0.77 percent once listings that were dark for almost the whole month are excluded. A chain deciding how many apps to run in three Gulf markets is not making one decision, it is making three, and the platform that costs least in trading hours differs between them. What the two bases mean and when each applies is in our methodology.
What recurring work does each additional platform actually add?
A queue you do not control, on top of the screens you do. Careem publishes its own change turnarounds as “Operational hours = 24 hours” and “Menu changes = 7 working days”, and routes part of its promotional tooling through a person, since “You can add offers/discounts/cpc activation by emailing partnerssupport.uae@careem.com or by emailing your Account Manager.” Jahez synchronises menus once a day, and according to Foodics “Jahez has an enabled auto sync everyday at 3 AM, If a manual sync is required please contact your Jahez Account Manager.”
The pattern repeats. Talabat’s Check-in feature is off by default and “To activate the check-in flow, please contact your account manager.” HungerStation grants portal plugin access the same way, saying “you will be given access by your account manager or during registration with us”, and directs partners to “reach out to your Account Manager to get the Vendor ID.” Deliveroo’s automatic opening requires “At least 95% availability to request Auto-open” and then a request through the Hub. Each new platform therefore adds a relationship to maintain and a set of turnaround times to plan around, and neither of those scales down for a small estate.
Which costs are fixed per platform and which scale with branches?
The expensive one is fixed, which is the opposite of what most cost models assume. Learning a platform’s vocabulary is paid once and does not get cheaper: that closure is called “Outlet Closed” here and FORCED_CLOSURE there and “Hidden” somewhere else, that this one auto reopens on a timer and that one does not, that a promotion publishes in five minutes on one and up to fifty on another. That knowledge sits with two or three people and is what actually limits how many platforms a chain can run well.
The variable cost per branch is comparatively light. Adding a twenty first branch to a platform you already run is a configuration task. Adding a fifth platform to twenty existing branches is twenty new listings plus a whole new set of rules, and the second is the one that quietly degrades the first four, because attention is the constrained resource rather than headcount.
What does a listing nobody is watching cost?
More than nothing, which is the answer most operators get wrong. A dormant listing stays in the catalogue and stays visible to customers, so it continues to represent the brand while being unable to sell. In Saudi Arabia in July 2026, 5.6 percent of the listings in our panel were offline for more than 90 percent of the month, and removing that group cut the reported downtime rate from 2.88 percent to 0.77 percent. Those are cards without a business behind them, and somebody signed each of them.
There is a contractual edge to it as well. Foody’s business terms list among the grounds for suspension, cessation or deletion the “Failure to receive orders via the platform for at least two months, in a way that the Partner appears inactive – unavailable on the platform.” So a listing left running on a platform nobody services is not a free option. It is a slowly maturing liability that ends either in a customer finding a dead card or in the platform ending the relationship on its own terms.
How should a chain actually decide the number?
By asking whether you can answer three questions about each platform before you add the next one. Who at head office knows this platform’s closure vocabulary and who can lift each state. What the turnaround is for a menu change, a schedule change and a promotion on it. And what this platform’s availability looked like at your branches last month, as a number rather than as an impression.
A platform you cannot answer those for is a platform you are not running, you are merely listed on. That is a defensible position for a launch quarter and a poor one for a second year. The useful discipline is to treat adding a platform as a staffing decision rather than a commercial one, since the commission is negotiated once and the servicing is forever.
What should be measured before adding one and after?
Before: the incremental demand you expect, against the servicing load described above. After: the same availability measurement you already run on the others, from the first week, because a new platform has no internal history and the first month is when configuration errors are cheapest to correct.
The measurement has to be independent of each platform’s own reporting to be comparable across them, since every vendor defines availability in its own terms and none of them will describe a competitor’s. Reading each public storefront from the customer side gives one definition applied identically to every platform, which is what Kitchain (kitchain.co) does, and it is the only way a chain can put four platforms in the same column of the same table and mean the same thing by each row.