RoleAxis

Insights · Food and beverage · September 10, 2026

The margin a restaurant owner cannot see until month end.

Food and beverage took 29% of all Saudi point-of-sale spend in the first quarter of 2025, about SAR 99 billion. Almost all of it runs on thin margins, and much of it is managed from a spreadsheet, weeks after the cost actually moved.

A very large market with very thin cushions

Knight Frank puts F&B at 29% of Saudi POS spend in Q1 2025, with retail and F&B together at roughly 45%. A 2023 mapping dataset counts on the order of 132,000 restaurants, cafes and bakeries in the Kingdom; treat that as an order of magnitude, not a census. What those businesses share is a cost structure where a few points of food cost decide whether a location makes money.

Why the margin goes invisible

The margin on a dish is a live number. It moves every time a supplier reprices an ingredient, every time a delivery arrives short, every time a recipe drifts in the kitchen. But in a typical operation the systems that hold those facts do not talk on any schedule faster than the monthly close: the POS knows what sold, the purchasing records know what was paid, and the recipe sheet that connects them lives in a spreadsheet nobody reconciles until month end. So the owner learns in the second week of the month what the kitchen did to margin in the first week of the last one.

None of this is a knowledge problem. Every operator knows the drill: explode the recipe into ingredients, price the ingredients at what you actually paid last, compare invoiced prices against the purchase order and the goods receipt, and chase the differences. It is simply too much arithmetic to do by hand every day across hundreds of items, so it gets done monthly, badly, or not at all.

What changes when the arithmetic runs daily

Once purchasing, stock and recipes live in an ERP, that arithmetic is automatable. This is the work our supply and inventory operations agent does today for businesses running Odoo: it reads the recipe structures, the purchase orders, the receipts and the stock levels, names the drift it finds in riyals, and drafts the fix, a transfer, a purchase hold, a supplier query, as a proposal. A person approves or rejects every one of those drafts before anything happens. The value is not that software is cleverer than the operator; it is that software does not get tired of the arithmetic on day two.

We have no customer case studies to show yet, and we will not invent any. When pilot results exist, they will be published with the method visible.

Sources

More from Insights: all posts. Related: Not every back-office task deserves an AI agent.