Skip to content
Perspective · Delivery

When AI orders dinner, who earns the margin?

As management consultants to restaurants in Bahrain, the UAE, Qatar, Saudi Arabia, Oman and Kuwait, we start here: an agent asked to order dinner ranks restaurants on price, delivery time and rating. It scans no list and answers to no paid position, so the lines priced to a browsing customer are questioned first.

A chef tossing a pan over an open flame while another plates a dish at the pass during dinner service, the night skyline through the windows
Exhibit 1 · Restaurants

Where QAR 100 of menu value went on one Doha restaurant's talabat statements

QAR 31

Across 476 orders over five months, the restaurant kept QAR 31 of every QAR 100 in menu value, before food, staff and rent.

Source: talabat statements of one Doha restaurant, 476 orders over five months, as reported in Moneyball Finance, 16 October 2026.

What moves first

About QAR 24 of menu value went to the GEM deal fee and paid visibility. That is roughly 66% of talabat's charges and 78% of what the restaurant kept. Visibility alone, through Premium Position and banners, was about 12% of menu value.

On today's list, restaurants pay to be seen and to run deals. The customer gets a lower price, and the restaurant bears the discount. An agent ranking on price, delivery time and rating has no fixed list and no position to buy, which puts the 12% at risk first.

Discounts stay part of the price an agent compares, and we expect the QAR 12 fee for running them to move next. Card fees and refunds remain.

What is harder to replace

We expect agents to unbundle aggregator economics one line at a time. Visibility goes first and deals next. Delivery density and customer history will be harder to replace.

The platform's own figures point the same way. talabat's marketplace lines added up to 24.3 points of GMV in Q2 2026, ads were 3.4, and At Par attributes 8 or 9 of the points to winning the customer, which is the slice an agent puts in play. eCommerceNews AU put the fee at 4% for one agent checkout model in early 2026, before it was scaled back. At a 20% share of GMV it would equal 0.8 points.

The board's questions

  • A Tenant should ask what share of its platform spend buys a place on a list, and what it would pay for a ranking that an agent actually reads.
  • A platform, or Sovereign, should ask whether to cut the price of visibility before agents do, sell agents a ranking product, or defend the current figure.
  • A Carrier should ask whether delivery density, which agents cannot yet replace, is priced into its contract.
  • A Merchant should ask whether its menu, prices and delivery promises can be read and compared by an agent.
  • Every board should ask for margin per order by line, and not only for orders won.
How we help

From one statement to the group's margin

We rebuild the order-level margin for restaurant and food delivery groups, line by line, and test it against agent scenarios. Restaurant operators start with the aggregator margin tool, and we set out our approach for restaurant and food delivery groups under that sector. We set the group's position on agents through Digital, AI & Agentic Commerce.

Sources

  1. talabat statements of one Doha restaurant, 476 orders over five months, as reported in Moneyball Finance, 16 October 2026.
  2. talabat, Q2 2026 results, released 12 August 2026: marketplace lines totalling 24.3 points of GMV; advertising and listing fees at 3.4%.
  3. The 8 or 9 points paying for winning the customer: At Par analysis, published in Moneyball Finance, 9 October 2026.
  4. ChatGPT checkout fee of 4% for Shopify merchants: eCommerceNews AU, 3 February 2026; scaled back per an OpenAI post of 24 March 2026, as quoted in the press. 0.8 points at a 20% share of GMV: At Par arithmetic.

Related

Bring last month. Leave with a score.

Sit down with Qasim Abbas. One table, your numbers, and a clear read on where you stand against par.