Ask most F&B owners in the Gulf how marketing is going and they will tell you about orders. Orders are up, the feed looks good, the delivery apps are busy. Then ask what those orders kept after the platform took its cut, and the room goes quiet. That silence is the whole problem. In this business the busiest month and the most profitable month are often not the same month, and the dangerous number is the one nobody on the marketing side ever looks at.
So I will start where the money actually is. The biggest line bleeding most GCC food businesses is not a weak agency or a quiet feed. It is the delivery aggregator's commission, and a marketing function that keeps feeding the channel that charges it the most. Talabat, Jahez, HungerStation, Deliveroo, Careem — each takes roughly a quarter to a third of every order it carries. The real job of restaurant marketing is not to push more volume through that pipe. It is to decide which orders belong in it at all, and to move as many covers as possible onto channels you own. That is the work an AI marketing system does for a food business, and this playbook is how you build one.
The order that grows revenue and shrinks profit
Run the arithmetic the order count hides. A AED 90 delivery order on an aggregator at 30% commission hands the platform AED 27 before you have paid for a gram of food. Add the discount you ran to win it, the packaging, the rider promo, the occasional refund, and the contribution that order leaves on your table can be thinner than the same dish sold to a walk-in at half the menu price. These are illustrative figures, not a real P&L, but the shape holds across the region: scale that channel hard enough and you can grow top-line revenue every month while profit per cover slides backwards.
That is why "we did more orders" is the most misleading sentence in a food business. The goal is not orders. It is contribution margin per cover — the cash a cover leaves after commission, discount, packaging, and refunds. See the business through that lens and the marketing question flips, from "how do we get more orders" to "how do we keep more of each one, and move the next one to a channel that doesn't tax us a third."
A scenario you will recognize
Here is a clearly illustrative case, built from patterns rather than one real operator. Read it for the shape, not the numbers.
Picture a four-branch casual-dining group across Riyadh and Jeddah, listed on three aggregators, with its own app that almost nobody uses, running some discount nearly every week. The monthly review celebrates rising orders. What it never shows is that two-thirds of those orders arrive through the highest-commission aggregator, the weekly discount has trained regulars never to pay full price, and the group's own app — the one channel with no commission — sits ignored because nobody owns filling it. Reviews pile up unanswered. A guest who ordered direct last month has no reason to come back, because nothing brought them.
Nothing there is fixed by more ad spend. The group has demand. What it lacks is a marketing function that defends the margin on each order and steadily shifts covers toward the channels and the repeat behavior it owns.
Move margin to channels you own
Get the strategy straight before any tool, because the tools only matter once the goal is right. Every cover sits on a margin ladder. A dine-in cover keeps the most. An order through your app or a WhatsApp reorder keeps almost as much. An aggregator order keeps the least, sometimes alarmingly little once the discount is counted. Marketing's job is to push the mix up that ladder over time — not to abandon the aggregators, which buy discovery you cannot replace, but to stop treating them as the destination. They are the top of the funnel. Your owned channels are where the margin and the relationship live.
In practice that means three habits most teams never build: capturing the guest's contact at the moment of an aggregator order so you can invite them back direct; making direct and WhatsApp ordering genuinely easier than reopening the app; and engineering repeat frequency on purpose instead of hoping for it. AI does not change that strategy. It lets a small team run it across four branches, two languages, and thousands of guests without drowning.
The agents that defend the margin
I do not stand up one big "AI." I wire a few narrow agents, each owning a single job, each handing a person the calls that carry reputation or money. For a food business the roster is set by the margin problem, not a generic content line.
An owned-channel router watches where demand lands and works the levers that pull repeat orders off the aggregators: a post-order message inviting a direct reorder, a reason to choose the app, a nudge timed to when that guest usually orders again. A menu-and-offer margin agent models what each item and promotion actually contributes after commission and discount, then flags the loss-leaders you are scaling by accident and the high-margin items you under-sell. A local-search agent keeps your Google Business Profile and map-pack presence sharp at every branch — hours, photos, menus, the "near me" surface where a hungry guest decides in ten seconds — because a strong map listing sends covers to your own door, not through a commission. A review-response agent drafts fast, on-brand replies to every rating so nothing festers, leaving the serious complaints for a human. And a WhatsApp repeat-order agent turns your best guests into a channel of their own: a saved order, a quick reorder, a reminder of the dish they always get.
What is missing from that list is a generic "make more content" engine. Velocity is not the constraint in F&B. Kept margin is. Every agent above exists to protect or repatriate a riyal. The next five sections are where those agents actually earn it.
Local search: win the map before the feed
"Near me" is the highest-intent query in food, and it gets decided on a map in about ten seconds. A guest who finds your branch on Google, taps directions, calls, or books a table costs you nothing per cover — no commission, no discount, usually no ad. That makes branch-level local search the cheapest acquisition channel a GCC restaurant has, and most groups treat it as an afterthought while they argue about the feed.
The work is unglamorous and it compounds. Every branch gets its own Google Business Profile, in Arabic and English, with photos taken this quarter, hours that are actually correct — including the Ramadan switch, which a startling share of Gulf listings gets wrong every year — and a menu link that lands on your own ordering page, not an aggregator app. If you take reservations, the booking link lives on the profile too, because a booked table is the most predictable high-margin cover on the ladder. The local-search agent from the roster keeps all of this current across branches and flags the drift a human only notices at review time: the branch whose photos are two years old, the "temporarily closed" flag nobody lifted after the refit.
The menu is a search asset, not a PDF
A menu locked inside a PDF or an Instagram image is invisible — to search engines, and to the AI answer engines a growing share of guests now ask first. Published as a crawlable bilingual page instead, with every dish, description, and price in Arabic and English, it starts working for you: "best mandi in Jeddah", "gluten-free breakfast in Dubai Marina", "iftar set menu Doha" are all queries a structured menu page can win and a PDF never will.
Dish-level content earns its keep twice more. The same descriptions feed your ads and your social without anyone starting from a blank page, and structured menu data gives Google something concrete to surface in the listing itself. One rule from the human-judgment section applies here with full force: no allergen, nutrition, or ingredient-origin claim ships unless the kitchen has confirmed it. A wrong price annoys a guest. A wrong allergen line is a different category of mistake.
Paid offers that don't train guests to wait
The standing weekly discount is the most expensive habit in GCC food marketing, because it stacks on top of the commission. Run 20% off through a channel that takes 30%, and an AED 100 order collects AED 80, hands over AED 24, and leaves AED 56 before you have plated the food — illustrative arithmetic, but the shape is real. Paid media in this vertical should do one narrower job: buy the first *direct* order or the first booked table, so the discount is a one-time acquisition cost for a guest you now own rather than a permanent subsidy flowing through a platform.
In practice that means geo-targeted search and Maps campaigns around each branch, branded defense so an aggregator cannot buy your own name out from under you, and offers built as single-use direct incentives instead of standing codes that teach regulars never to pay full price. Judge every dirham of it on two numbers: the platform's cost per order, and the contribution that order kept after commission and discount. It is the same discipline I run on Google Ads engagements in Dubai — and for teams that work in Arabic, the same service page exists in Arabic.
Loyalty and CRM: engineer the second visit
In food, the economics live in the second visit and the fifth; the first order mostly exists to earn them. Yet plenty of GCC groups own no guest record at all — the aggregator holds the customer relationship and rents it back as sponsored placement. Loyalty therefore starts with capture: a phone number taken at the order, the reservation, or the wifi login, flowing into a CRM the group controls. From there the plays are mostly automated and boringly effective: the saved WhatsApp reorder, a nudge timed to each guest's usual gap between visits, a follow-up that recovers a no-show booking instead of shrugging at it. The messaging mechanics are the ones I walked through in WhatsApp commerce for GCC stores.
Judge the program on visits, not points. "Members enrolled" is the gross number; the net one is the incremental covers those members added per quarter, and at what discount cost. A scheme that mostly rewards visits that would have happened anyway is a discount program wearing a loyalty badge.
Reviews: the rating is a ranking input and a P&L input
A half-star on Google or a delivery app moves real money. It shifts your map-pack ranking, your placement inside the aggregators, and the conversion of every guest who checks before ordering — which in this region is nearly all of them. So treat reviews as a weekly operating channel rather than a reputation chore for the end of the month: respond fast, in the language the review was written in, to all of them, with the review-response agent drafting and a person approving anything sensitive.
The second use is the one most operators skip. Reviews are free operational telemetry, and an agent that clusters complaints by branch and dish will surface "the Jeddah branch's packaging fails on rice dishes" weeks before it shows up in any number. And the line from the human-judgment section stands: anything touching food safety or an allergic reaction goes to a person immediately, with no templated reply in front of it.
Two numbers: orders on the dashboard, riyals in the till
The discipline I hold to in food is simple: one number is a lie, so always show two. The first is the flattering figure the aggregators and ad platforms hand you free — orders, gross GMV, reach, cost per order. The second is the one that survives contact with reality: contribution actually kept after commission, discount, packaging, and refunds, reconciled against the POS, split by channel and branch.
"A thousand orders last month" tells you nothing if a third went to commission, a standing discount shaved the rest, and the till barely moved. Show the two figures together and the weekly meeting flips, from "let's run another discount" to "which channel and which item actually left cash on the table, and how do we move more covers there." That question is the whole game, and a team fixated on the gross count never reaches it. I made the full argument for this in the two-number report and why dashboards lie.
The wiring behind the two numbers
A two-number report only exists if the plumbing behind it does, and in food that plumbing has a specific shape. The POS is the source of truth; ad platforms and aggregator dashboards are witnesses with an interest. Aggregator statements get reconciled against the POS monthly, because the commission, refund, and promo-fee lines buried in those statements are where the second number actually lives. Direct ordering carries promo codes and UTMs so a repeat order can be traced back to the nudge that produced it. And everything reports per branch and per channel, because a group-level average hides exactly the branch that is quietly losing money.
None of this needs an enterprise stack — a four-branch group can run it off POS exports and a reconciliation an agent refreshes weekly. What it does need is sequence. Wiring comes before scaling: put spend behind an unmeasured channel and all you buy is a bigger number you cannot trust.
Where the human has to stay
In food the line you protect from automation is not subtle. AI does not cook, does not taste, and cannot tell you a dish is going out wrong; no routing agent saves a brand whose food slipped while everyone watched the dashboard. The plate is the marketing — if it is off, every kept riyal upstream evaporates downstream. So a human keeps the final word on any food-safety or allergic-reaction complaint, where a polished automated reply is the dearest shortcut in this industry, and no claim about a menu fact, a nutrition number, or an ingredient origin ships unless it traces back to real data. The host's read of a room and the chef's read of a plate are exactly what you keep off the machine.
A 60-day build
Start by seeing the truth, not by changing anything. In the first two to three weeks, connect the aggregators, the ad platforms, and the POS so the second number exists at all — contribution per channel and per branch, next to the orders you were already celebrating. You will probably dislike what the high-commission channel keeps once you can see it. That discomfort is the point.
In the next weeks, build the margin agent and the owned-channel router. Find the promotions quietly losing money and stop scaling them. Stand up the post-order invitation that pulls aggregator guests toward a direct reorder, and make direct ordering genuinely easier than it is today. Only then turn on the WhatsApp repeat-order and review-response agents, because driving repeat frequency is only worth it once you can see which repeats actually pay.
By the end of the cycle you have a system that knows what every order keeps, defends the margin item by item, and moves covers toward the channels you own. For where this sits in a wider build, see AI marketing for restaurants & F&B in the GCC.
The opinion, plainly
What passes for "AI for restaurant marketing" here is mostly faster caption-writing with a bill attached, engineered to push more volume down the one channel that is quietly eating your profit. The unglamorous edge almost nobody takes: know your contribution per cover after the aggregator's cut, stop scaling the orders that lose money, and move every cover you can onto channels that don't charge you a third. Do that and you out-earn a competitor running twice your orders, because you keep what they hand to a platform.
Next step
If you want to find out what your orders are really keeping after commission, request a restaurant & F&B marketing audit. Prefer a direct conversation? Message Ahmed on WhatsApp.