Restaurant marketing is not a content problem. It is a systems problem wearing a content costume — and the brands that win in the Gulf are the ones that stop buying tools and start building a machine that publishes, protects the brand, and tells the truth about money.
My name is Ahmed Ayoutty. I spent 13 years building marketing for the Saudi market and ran three agency groups before going all-in on AI-native marketing systems. I work fully remotely across the GCC and the United States, which suits F&B well: your problems do not respect office hours, and neither does the system I build for you. This page is about one thing — how a restaurant, cafe, cloud kitchen, or multi-branch F&B group builds an AI marketing system that actually moves contribution, not just order count.
From the outside, F&B marketing looks like pretty plates and a busy Instagram. Underneath, it is one of the most punishing publishing operations in any industry. You are shipping content constantly — new items, dayparts, weekend offers, Ramadan menus, branch-specific promos — in Arabic and English, across your own channels and three or four delivery aggregators that each demand a different format and tone.
Reviews land every single day on Google, on Instagram, and inside every delivery app, and a slow or clumsy reply to an angry one will cost you more than a week of ad spend. Meanwhile the number everyone stares at — orders — quietly lies. It says nothing about what survived a large aggregator commission, a near-permanent discount, and the refunds. You can grow orders every month and shrink the business at the same time, and most teams never see it because no one is reconciling the dashboard against the POS.
When I build one of these, I do not buy "an AI." I build a handful of narrow agents, each owning one job, each leaving a human in control of judgment and the publish button. Five roles carry the weight.
A research agent assembles the raw material before anyone writes a word: competitor menus and price points in your area, what diners are actually searching and asking for, the recurring themes buried in your reviews, and the dayparts and occasions you are under-serving. A draft agent turns one item spec or one offer into a bilingual set — menu descriptions, aggregator listing copy, and a few social variants — in minutes, with Arabic written as Arabic, not translated as an afterthought. A QA agent then checks every draft against your brand voice, against allergen, halal, and health-claim rules, against the real price and availability, and against your list of claims that are never allowed. Only then does a publish-and-route agent push approved content to your channels and, just as importantly, triage the inbound — routing an angry review, a catering enquiry, or a press question to the right human fast, instead of letting it rot in a shared inbox.
The fifth agent is the one most vendors quietly skip. A measure agent reconciles the orders your aggregators and ad platforms report against what your POS says actually became net contribution after commission, discount, and refunds — per channel, per branch, per offer. It is the least glamorous piece and the one that changes how the company makes decisions, because it is where the truth lives. If you want this wired into a broader build, that is exactly what I do in AI marketing systems.
Let me walk a clearly illustrative scenario — this is a composite, not a client result, and the point is the shape of the work, not the numbers.
Picture a four-branch casual-dining group across Riyadh and Jeddah, listed on three delivery aggregators plus its own app, heavy on Instagram and TikTok, running some kind of discount almost every week. The dashboard shows orders climbing month over month, so on paper marketing is winning. On the floor, the founder cannot tell you which of those orders actually made money. Listings go out in English first and Arabic "when there is time," so the Arabic menu is thinner and converts worse. And the genuinely valuable enquiry — a 200-cover corporate catering request that landed in the Instagram DMs on a Thursday night — got answered Monday, by which point it was booked elsewhere.
Notice that nothing here is solved by more ads. The group already has plenty of orders; what it lacks is publishing speed, brand-safe consistency, and an honest scoreboard. After instrumenting net contribution, a composite group like this typically discovers something uncomfortable — say, that one aggregator tier and the always-on 25%-off bundle are quietly unprofitable, while its own-app orders and one full-price signature combo carry the whole business. Shift budget and attention there, publish bilingual listings the same day items launch, and route the high-value enquiries to a human within minutes. Order count might even dip slightly. Contribution goes up. That trade is the entire game.
Here is the rule I will not bend on: every report shows two numbers, never one. The first is the flattering top-of-funnel figure — orders, gross GMV, reach, cost per order. The second is the number that survived contact with reality — net contribution after aggregator commission, discounts, and refunds, reconciled against the POS.
One number alone is how F&B marketing lies to itself. "We did a thousand orders last month" means nothing if a heavy commission, a 25%-off, and the refunds left contribution flat or negative. Put both numbers side by side and the conversation in the weekly meeting changes overnight. It stops being "let us run another discount" and becomes "which channel and which offer actually made money, and how do we sell more of that?" That second question is where margin is made. I made the full case for it in the two-number report and why dashboards lie.
I will not invent a restaurant case study to sell you one. The most relevant proof I can point to comes from a different vertical — education — and the honest framing is that the method transfers, not the numbers.
FIT Institute competes in a sector dominated by globally recognised names. After deploying a systematic Generative Engine Optimization strategy, its content began appearing in Google's AI Overviews across its catalog and being cited alongside — in some queries ahead of — PwC content on overlapping topics. On the paid side, the same engagement turned 121,330 AED of ad spend into ~912,550 AED of collected revenue — roughly 7.5× clean ROAS. Education has no product to return, so gross and collected converge here; I still report both numbers, by rule.
The mechanism transfers directly to F&B: a diner who used to Google "best brunch near me" now asks an AI assistant first, and the brands that build citable, brand-safe, bilingual content now — wired into a system that publishes fast and measures honestly — will own that surface before their competitors notice it exists.
Especially to you. Aggregator dependence is exactly where order count and real profit drift apart, because commissions and discounts eat the margin invisibly. The measure agent reconciles aggregator orders against POS contribution per channel, so you can finally see which platforms and offers are worth the volume — and which are buying you busy, unprofitable kitchens.
Yes, and it is built that way on purpose. Arabic is drafted as Arabic, not machine-translated after the fact, because Gulf diners feel the difference and it shows in conversion. Bilingual parity — same speed, same quality in both languages — is a default of the build, not an upsell.
It starts with a discovery call and a scoped proposal. I usually instrument the truth first — connecting aggregators, ad platforms, and POS so the measure agent can show both numbers honestly before we change anything — then build the draft and QA agents around your highest-volume item type, and only then turn on routing. System builds run on defined milestones, typically four to twelve weeks depending on complexity. No open-ended retainers without clear deliverables.
Bring a real problem — a discount you cannot tell is profitable, an Arabic menu that lags the English one, or a review queue nobody owns. We will figure out which bottleneck is costing you the most and whether I am the right person to build the fix.
Request a diagnostic →