The checkout bill for agent-led commerce
Our management consultants tell boards in Bahrain, the UAE, Qatar, Saudi Arabia, Oman and Kuwait that the cost of taking a payment barely changes when an agent triggers it. What changes is who charges for access to the customer, and talabat's 24.3 points of GMV show how large that charge already is.
The card fee is the wrong place to look
Across the Gulf gateways that publish prices, taking a card payment costs the seller between about 1 and 4 on every 100. A mada payment costs the same whether a person or an agent triggers it. An agent working for the buyer has no reason to choose a cheaper rail for the seller, so the processing line stays where it is.
Access to the customer carries the price
Today talabat charges restaurants for that access. Commission, deals, ads and fees added up to 24.3 points of GMV in Q2 2026, and ads alone were 3.4. That is the baseline.
When ChatGPT ran checkout for Shopify merchants in early 2026, eCommerceNews AU put the fee at 4% of the sale, on top of normal card processing. The arrangement was scaled back in March 2026. The 4% remains larger than talabat's whole ads line.
A 4% agent fee on one fifth of GMV equals 0.8 points
0.8 points
If 20% of GMV came through an agent charging 4%, the fee would equal 0.8 points of total GMV. At the same 4%, it would match talabat's whole 3.4-point ads line only when about 85% of GMV came through agents. Both figures are At Par arithmetic on an illustrative share.
Source: talabat Q2 2026 results, 12 August 2026; eCommerceNews AU, 3 February 2026, for the 4% fee; At Par arithmetic.
A cost or an investment
Whether the fee is a cost or an investment depends on one thing. If the agent brings orders talabat would never have won, 4% is a customer acquisition cost like any other. If it brings customers who would have opened the app anyway, it is a toll on contribution talabat already had.
The designs published by Google and OpenAI keep the merchant as merchant of record. talabat would still process, settle and handle refunds. What it would lose is the start of the purchase, and with it the easy reorder that a saved card and a saved history make possible.
We expect the aggregators that come out ahead to be those that can measure incremental contribution from agent-sourced orders, after the cost of acquiring them. Referral attribution says where an order came from. It cannot say whether the group would have had the order anyway.
The board's questions
- A Tenant should ask whether an agent sending it orders directly would charge less for access than the platform it uses today, and who would then hold the customer record.
- A Merchant should ask what it would pay an agent for distribution, and whether it can tell new customers from the ones it would have won anyway.
- A Sovereign that runs a marketplace should ask whether it is still the marketplace when an agent brings the customer, or has become a merchant paying another platform for distribution.
- A Carrier should ask whether a fee at the front door will be recovered from the delivery rate it is paid.
- Every board should ask for contribution per agent-sourced order after acquisition cost, with new customers counted separately.
Pricing the front door
We build the incremental-contribution view for agent-sourced orders, test an access fee at different shares of GMV, and set the terms the group will accept from an agent channel. We deliver the work through Digital, AI & Agentic Commerce and the financial model through the Office of the CFO.
Sources
- talabat, Q2 2026 results release, 12 August 2026: marketplace lines adding up to 24.3 points of GMV; advertising and listing fees at 3.4%.
- Card fees: published pricing of Telr (Saudi Arabia), SADAD, Stripe (UAE) and Amazon Payment Services, compiled for Moneyball Finance, 19 October 2026.
- The 4% fee on ChatGPT checkout for Shopify merchants: eCommerceNews AU, 3 February 2026. Scaled back: OpenAI post of 24 March 2026, as quoted in the press.
- Google Universal Commerce Protocol and OpenAI Agentic Commerce Protocol documentation, read 18 October 2026.
- Scenario arithmetic: At Par analysis. Published in Moneyball Finance, 19 October 2026.
Related
- Margin Intelligence: where 24.3 points of GMV goLayer 6 of Moneyball Finance. Of talabat's 24.3 points of GMV, only the slice that pays for winning the customer is in play when agents place orders.
- When AI chooses for the customer, every position on the matrix movesAgents put paid visibility, the customer relationship and platform terms in question. Tenants, Merchants, Carriers and Sovereigns each face a different answer.
- Restaurant aggregator marginThe contribution a restaurant keeps from one aggregator order, and what it adds up to in a month.
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.