The Spreadsheet That Broke on a Thursday Night
It’s 8:30 PM on a Thursday in March, the busiest service of the week, and Faisal is sitting in the back office of his JLT restaurant staring at a Google Sheet titled “Catering Orders - March.” There are 43 rows. Seven of them are highlighted in red, which means he isn’t sure whether the customer confirmed. Three are highlighted in yellow, meaning someone took the inquiry over WhatsApp but never entered the details. One has a note that just says “call back,” with no phone number attached.
His phone buzzes. A WhatsApp voice note from a corporate office manager in DIFC asking whether they can change their Friday lunch order from 40 people to 55 — for tomorrow. Faisal doesn’t know which order she’s referring to because the catering phone is shared across two locations and nobody logged the original conversation.
This is what running four restaurant locations in Dubai looks like when your ordering operation depends on a combination of delivery apps, WhatsApp threads nobody owns, and spreadsheets that are always two days behind.
Faisal manages operations across four locations: a casual dining spot in JLT, a shawarma and grill concept in Business Bay, a seafood restaurant in Dubai Marina, and a fast-casual branch in JVC that opened eight months ago. Combined, the four locations do roughly AED 1.6 million in monthly revenue. On paper, healthy. In practice, he’s watching a third of his margin walk out the door.
The Delivery App Tax
The math is brutal and every restaurant operator in the UAE knows it. The major delivery platforms charge 25-30% commission on every order. For a restaurant doing AED 400,000 per month through delivery apps — which Faisal’s four locations were — that’s AED 100,000 to 120,000 per month going to platforms. Not to rent, not to food costs, not to staff. To the apps.
“I’m basically running a kitchen for the delivery apps,” Faisal says. “They own the customer. They own the data. They set the terms. And if I raise my prices on their platform to cover the commission, they push me down the algorithm.”
The dependency was growing, not shrinking. Across his four locations, delivery app orders represented 58% of total revenue. Dine-in was 30%. Direct orders — phone calls, walk-in takeaway, and the handful of customers who ordered through WhatsApp — were just 12%.
The delivery apps also created an operational problem beyond commission. Each platform has its own tablet sitting on the counter. Orders come in through one delivery app’s interface, the other’s interface, and the POS system for dine-in. The kitchen gets tickets from three different sources with three different formats. During peak hours, the kitchen team juggles between screens, and orders get mixed up. Faisal estimates that delivery app order errors — wrong items, missing items, incorrect modifications — ran at about 9%, compared to 3% for dine-in orders taken face to face.
But the delivery apps also did something no one talks about openly: they absorbed the marketing cost. Faisal didn’t have to advertise to get orders. The apps brought the customers. The question was whether the 30% commission was worth what amounted to a customer acquisition and logistics service — or whether there was a way to keep the customers he’d already acquired without paying the toll every time.
The Catering Black Hole
Delivery apps were the margin problem. Catering was the chaos problem.
Corporate catering is a significant revenue stream for Dubai restaurants, particularly locations near business districts. DIFC alone has over 25,000 professionals, and the JLT and Business Bay clusters each have tens of thousands more. During Ramadan, iftar catering demand spikes dramatically — companies ordering for 50, 100, sometimes 200 people.
Faisal’s catering pipeline was entirely manual. A potential customer would reach out on WhatsApp, usually to the restaurant’s general number. Whoever happened to see the message would respond. Sometimes that was a floor manager, sometimes a cashier, sometimes Faisal himself. The conversation would happen over several messages: menu options, headcount, dietary restrictions, delivery timing, pricing.
The problem was follow-through. Nobody owned the catering pipeline. Conversations lived in WhatsApp threads on multiple phones. There was no central view of pending inquiries, no automated follow-up, and no system for converting an inquiry into a confirmed order with a deposit.
Faisal ran the numbers for Q4 of the previous year. His team had received 127 catering inquiries across the four locations. They’d converted 41 into confirmed orders. That’s a 32% conversion rate, which sounds acceptable until you consider that the average catering order was AED 3,800 and the 86 lost inquiries represented over AED 325,000 in potential revenue.
“I wasn’t losing these because people didn’t want to order,” Faisal explains. “I was losing them because we took six hours to respond, then forgot to follow up, then couldn’t find the conversation when the customer called back.”
The Reservation No-Show Drain
The dine-in side had its own problem: reservation no-shows. The Marina and JLT locations took reservations through WhatsApp, the restaurant’s Instagram DMs, and phone calls. No-show rates were running at 22%, which is above the Dubai F&B average of around 15-18%.
A no-show on a regular Tuesday isn’t catastrophic. A no-show on a Thursday or Friday night, when every table is pre-booked and there’s a waitlist, means lost revenue that’s gone forever. A party of six that doesn’t show up at 8:30 PM on a Friday can’t be replaced — by the time you realize they’re not coming, the peak window has passed.
The root cause was straightforward: no confirmation system. Someone would book via WhatsApp, and the only reminder was if the host remembered to message them on the day. Nobody had time to send 30 individual WhatsApp reminders on a busy Friday afternoon.
Deciding to Build a Direct Channel
By late 2025, Faisal had reached the conclusion that a lot of restaurant operators in the UAE eventually reach: the delivery apps are a necessary evil for discovery, but building a direct ordering relationship with repeat customers is the only path to sustainable margin.
The challenge was how. He’d looked at building a branded ordering app, but the economics don’t work for a four-location group. Development costs start at AED 80,000, and the likelihood of customers downloading yet another food ordering app is near zero. He’d tried promoting phone orders, but customers don’t want to call — they want to text.
WhatsApp was the obvious channel. Customer adoption was already there. People were already messaging his restaurants. The problem was that WhatsApp, as a human-operated channel, couldn’t scale. His staff were already overwhelmed handling the conversations they had.
In November 2025, Faisal engaged our team to build an AI agent that would turn WhatsApp from a chaotic inbox into a structured ordering, catering, and reservation system. The scope was specific:
- Direct ordering for delivery and pickup: Customer sends a WhatsApp message, browses the menu through the conversation, places an order, and pays — all without touching a delivery app
- Catering inquiry handling: Automated response to catering inquiries with menu options, pricing, availability, and structured follow-up until the order is confirmed or explicitly declined
- Table reservations with confirmation: Booking, automatic day-of confirmation, waitlist management, and no-show tracking
- Customer feedback collection: Post-meal feedback via WhatsApp to catch issues before they become Google reviews
Building the Menu Into the Conversation
The first technical challenge was making the menu conversational. A restaurant menu with 80+ items, each with modification options, dietary flags, and varying availability across four locations, is not something you dump into a WhatsApp message.
The agent was designed to work like a knowledgeable server. When a customer messages “I want to order dinner,” the agent asks which location, identifies the customer’s area for delivery feasibility, then navigates the menu through guided conversation. “Are you in the mood for grills, seafood, or something lighter?” narrows the menu before presenting specific items.
Each menu item includes its modifications (spice level, protein substitutions, side options) and its Dubai Municipality Foodwatch compliance flags. Dubai Municipality requires that all food establishments display allergen information, and Faisal’s operation was already flagging the 14 major allergens on his physical menus. The AI agent carries this data and proactively mentions it: “Just so you know, the mixed grill platter contains nuts in the dipping sauce. Want me to swap it for the garlic tahini?”
The menu syncs from a master Google Sheet that Faisal’s kitchen managers update. When an item runs out, the manager marks it in the sheet, and the agent stops offering it within minutes. This solved a persistent delivery app problem — customers ordering items that were out of stock, leading to substitution calls and cancellations.
The VAT and Pricing Layer
UAE VAT at 5% applies to all food and beverage sales. On delivery apps, VAT is baked into the displayed price and the platform handles the receipts. For direct orders, Faisal needed VAT-compliant invoicing.
The agent generates an itemized order summary with VAT calculated and displayed as a separate line item, matching Federal Tax Authority requirements. When the customer confirms the order, they receive a payment link. The receipt is automatically logged in a format that Faisal’s accountant can import directly during VAT return filing.
This detail matters more than it sounds. Several restaurant operators in Dubai have been caught out by direct ordering channels that don’t properly document VAT, leading to reconciliation headaches during quarterly FTA filings. Getting the invoicing right from day one saved Faisal from a problem he didn’t know he was about to have.
Handling 200 Nationalities in One Conversation
Dubai’s population is roughly 85% expatriate. Faisal’s customer base includes Emirati families, Indian professionals, Filipino workers, British expats, Pakistani students, and dozens of other nationalities. The WhatsApp conversations that come in reflect this: Arabic, English, Hindi, Urdu, Tagalog, and frequent code-switching between them.
The agent handles Arabic and English natively, with intent detection tuned for Gulf Arabic food terminology — “mashawi” (grills), “fatteh” (a Levantine dish), “machboos” (Emirati rice dish), and dozens of regional dish names that don’t translate cleanly. For Hindi and Urdu, the agent can understand basic ordering intent and food preferences, but routes complex conversations or complaints to a human operator.
One early issue: South Asian customers frequently order by description rather than menu item name. Instead of “chicken shawarma,” a message might read “that chicken wrap thing with the garlic sauce.” The agent was initially trying to exact-match against menu items and failing. The team built a fuzzy matching layer that maps common descriptions to actual menu items, trained on three months of historical WhatsApp conversations from Faisal’s restaurants.
The Catering Transformation
The catering module was where the agent delivered the most dramatic improvement.
When a catering inquiry comes in, the agent immediately responds with an acknowledgment and three questions: event date, approximate headcount, and any dietary requirements. Within the first minute, the customer receives a structured response instead of the old pattern of “seen” and silence.
Based on the answers, the agent presents two to three menu package options at relevant price points, calculated automatically from the headcount. Corporate inquiries get a different set of options than personal celebrations. Ramadan iftar inquiries — which spike massively during the holy month — get a dedicated menu with traditional dishes and suhoor add-ons.
The agent follows up systematically. If the customer hasn’t responded in four hours, a gentle check-in. After 24 hours, a second follow-up with a slight incentive: “If you confirm by tomorrow, we can include complimentary Arabic coffee service for your event.” After 48 hours without response, the inquiry is marked as cold but logged for reactivation when the customer’s event date approaches.
The response time alone changed the conversion dynamics. In the first three months, catering inquiry-to-order conversion went from 32% to 54%. On average catering order values of AED 3,800, that’s substantial recovered revenue.
The Ramadan Stress Test
The real test came during Ramadan, which in 2026 fell in late February through March. Ramadan transforms the Dubai F&B landscape. Restaurants shift from lunch-heavy to iftar-heavy service. Operating hours change. Catering demand quadruples. And the entire ordering pattern shifts — very little ordering during the day, then an intense spike starting 90 minutes before iftar time.
During the first week of Ramadan, the agent handled over 600 ordering conversations per day across the four locations, with a peak of 180 concurrent conversations in the 90-minute window before iftar. The system maintained sub-2-minute response times even at peak load.
The Ramadan catering pipeline was the real story. The agent processed 94 iftar catering inquiries in the first 10 days of Ramadan, converting 51 into confirmed orders. The previous Ramadan, Faisal’s team had managed 38 total confirmed catering orders across the entire month.
Faisal had pre-loaded the agent with Ramadan-specific configurations: adjusted delivery time windows (no deliveries during fasting hours, peak availability at iftar), special Ramadan menus, and iftar package pricing. The agent also handled a common Ramadan request that had previously been a manual headache: recurring daily iftar orders for offices observing the holy month together.
The Reservation Fix
The reservation module was simpler but high-impact. When a customer books a table through WhatsApp, the agent confirms immediately with the date, time, party size, and location. It sends an automatic reminder 24 hours before and a final confirmation 3 hours before the reservation.
If the customer doesn’t confirm the 3-hour reminder, the table is released back to availability and the customer is notified: “We haven’t heard back, so we’ve released your table for tonight. Want to rebook for another time?”
No-show rates dropped from 22% to 9% within two months. More importantly, the released tables from non-confirmed bookings were now available for walk-ins or last-minute reservations, recovering revenue that had previously just evaporated.
The agent also manages a simple waitlist. When tables are full on a Thursday or Friday night, customers can join a WhatsApp waitlist and receive a notification when a table opens, typically from a cancelled or non-confirmed reservation. This created a second-chance revenue capture that didn’t exist before.
What Went Wrong
Three months in, not everything worked as planned.
Delivery logistics remained the hard problem. Faisal could take direct orders, but he still needed someone to deliver them. He contracted with a third-party fleet service, but their reliability was inconsistent. During peak hours, delivery times stretched to 55-60 minutes, compared to the 30-35 minutes customers were used to from the delivery apps, which have logistics down to a science. Several customers who tried direct ordering once switched back to the delivery app because the food arrived late.
Faisal ended up limiting the delivery radius for direct orders to 5 km from each location and being transparent about delivery times: “Direct orders typically arrive in 40-50 minutes. If you need it faster, we’re also on the delivery apps.” Honest messaging reduced complaints but also capped the percentage of customers willing to order direct.
Payment friction was real. On delivery apps, payment is one tap — your card is saved, Apple Pay works, and it’s done. The direct WhatsApp ordering flow required clicking a payment link, entering card details or paying via Apple Pay on a payment page, and confirming. It’s two extra steps, and for orders under AED 80, some customers didn’t bother. Cash on delivery helped — roughly 30% of direct orders opted for COD — but it introduced its own problems with change and no-shows.
The agent couldn’t handle complaints well. When a customer received cold food or a wrong order and messaged angrily, the agent’s attempt to gather information (“Can you tell me your order number and what was incorrect?”) came across as bureaucratic rather than empathetic. Complaints are escalated to humans now within the first response, but the initial handoff sometimes added frustration. A customer who’s already angry doesn’t want to explain the problem twice.
What It Handles Well vs. What It Doesn’t
Handles well:
- Repeat ordering for known customers (the agent remembers previous orders and suggests “your usual?”)
- Catering inquiries from first contact through to confirmation and deposit collection
- Menu browsing with dietary restriction filtering (halal certification details, allergen flagging, vegetarian/vegan options)
- Reservation booking, reminders, confirmation, and waitlist management
- Post-meal feedback collection (a simple 1-5 rating request sent 2 hours after delivery)
- Ramadan-specific scheduling and menu adjustments
- Bilingual Arabic/English ordering with food-specific vocabulary
Doesn’t handle well:
- Complex complaints requiring empathy and service recovery
- Custom menu requests outside the standard modification options (“can you make the biryani but with less oil and extra crispy onions on the side in a separate container”)
- Group ordering where multiple people are messaging from the same WhatsApp number
- Negotiation on catering pricing for large events (customers expect to haggle above 100 people, and the agent doesn’t have authority to discount)
- Delivery tracking — the agent can confirm an order is out for delivery but can’t provide real-time driver location the way delivery apps do
The Numbers After Three Months
By mid-February 2026, the results across the four locations:
Direct order percentage: 12% to 38%. More than a third of orders now come through WhatsApp rather than delivery apps. This is concentrated among repeat customers — first-time customers still discover the restaurants through the delivery apps, but after two or three orders, the agent sends them a message: “Order direct next time and save. Same food, no app markup.”
Monthly commission savings: AED 47,000. The 26 percentage points shifted from delivery apps to direct orders, at an average commission rate of 28%, translates to roughly AED 47,000 per month in commissions that Faisal no longer pays. The AI agent’s operating cost is approximately AED 5,200 per month across all four locations.
Catering conversion: 32% to 54%. Response speed and systematic follow-up turned more inquiries into confirmed orders.
Reservation no-shows: 22% to 9%. Automated confirmations and the release-and-notify system cut no-shows by more than half.
Customer data ownership: This is the metric that doesn’t show up on a P&L but matters most to Faisal. He now has direct WhatsApp contact with over 2,800 customers, along with their order history, preferences, and feedback. On delivery apps, that data belongs to the platform. “When I launch a new dish, I can message 2,800 people who’ve actually ordered from me. Try doing that through a delivery app.”
The Customer Data Play That Changes Everything
Faisal is blunt about why customer data matters: “Delivery apps are renting me customers. Every month I pay commission, I’m paying rent. Direct WhatsApp customers are owned.”
The agent tracks ordering patterns: what each customer orders, how often, at what times, from which location. This data feeds two things. First, personalized reorder prompts. A customer who orders chicken shawarma from the Business Bay location every Wednesday at 1 PM gets a message at 12:30: “Ready for your usual Wednesday shawarma? I can have it ready by 1.” Personalized prompts have a 34% conversion rate, compared to 7% for generic promotional messages.
Second, it feeds menu development. Faisal noticed through the ordering data that customers frequently asked for items that weren’t on the menu — specific combination plates, family-size portions of popular dishes, and healthy bowl options. He’s used this data to plan his Q2 menu update, something he would have done by gut feel before.
What Faisal Would Do Differently
Four things, in his words:
Solve delivery first. “I spent all this effort building the ordering channel and then stumbled on the delivery. If I could do it again, I’d lock in a reliable fleet partner before launching direct orders. The channel is only as good as the fulfillment.”
Don’t fight the apps; layer on top. “I tried to pull customers off the delivery apps aggressively at first. It annoyed some of them. Now I let the apps do the acquisition and use WhatsApp for retention. New customer? The delivery app is fine. Third order? Here’s why you should go direct.”
Start with catering, not ordering. “Catering had the highest ROI and the lowest operational complexity. No delivery logistics, higher margins, and the AI follow-up alone was worth the investment. If I’d started with just catering, I would have seen returns in week one.”
Get staff buy-in early. “My floor managers saw the WhatsApp agent as a threat at first. Two of them were the ones handling catering inquiries, and they thought they were being replaced. I should have involved them in the setup so they could see it was handling the tedious parts, not replacing judgment.”
Where It Goes Next
Faisal is planning three expansions. First, integrating the ordering agent with a dedicated delivery fleet management tool so that direct orders get the same tracking experience customers expect from the apps. Second, building a loyalty program into the WhatsApp channel — every fifth direct order earns a discount, tracked automatically through the agent. Third, expanding the feedback collection into a structured review pipeline where satisfied customers are nudged to leave a Google review and dissatisfied customers are routed to a service recovery flow before they ever reach a public review platform.
The delivery apps aren’t going away. Faisal still does 62% of his revenue through them. But the trajectory has shifted. Three months ago, his direct channel was an afterthought. Now it’s his fastest-growing revenue source, his highest-margin channel, and the only one where he actually knows his customers by name.
“Every restaurant in Dubai is complaining about delivery app commissions,” Faisal says. “Most of them are complaining while still paying. I decided to build the alternative instead.”