When AI chooses for the customer, every position on the matrix moves
Platforms in Bahrain, the UAE, Qatar, Saudi Arabia, Oman and Kuwait earn part of their margin because a person scans a list; as management consultants, we ask what follows when an agent ranking on price, delivery time and rating has none to scan. Paid visibility goes into question first, then the customer relationship itself.
The board-level question
The board-level question is whether agents create contribution a group did not have, or transfer contribution it was already earning to someone else. Tenants, Merchants, Carriers and Sovereigns each face a different answer, and each has different moves available before the terms are reset.
Where the evidence already shows it
Visibility is a priced line
3.4%
talabat earned 3.4% of GMV from advertising and listing fees in Q2 2026, and its marketplace lines added up to 24.3 points. Placement is sold to people who scan, and an agent ranking on price, delivery time and rating does not.
Source: talabat Q2 2026 results, released 12 August 2026.
Access to the customer has a price
4%
eCommerceNews AU put the fee for ChatGPT checkout for Shopify merchants at 4% in early 2026, and OpenAI scaled the arrangement back in March. Scenario: at a 20% share of GMV it equals 0.8 points.
Source: eCommerceNews AU, 3 February 2026; OpenAI post of 24 March 2026, as quoted in the press; At Par arithmetic.
Notice is worth something to a fleet
SAR 14.2m
Jahez Group's logistics arm, Logi, took quarterly adjusted EBITDA from SAR 2.0m to SAR 14.2m in Q2 2026. An agent that books ahead gives a fleet more notice, and the contract decides who keeps the gain.
Source: Jahez Group Q2 2026 results; At Par's reading.
Four positions, four answers
Tenant
Pays for reach, and an agent changes what reach costs. On one Doha restaurant's statements, about QAR 24 of every QAR 100 of menu value went to deal fees and paid visibility. The move is to know how much of that an agent would make unnecessary.
Merchant
Owns the brand and the customer relationship, and will meet an agent at checkout. The published Google and OpenAI designs keep the merchant as merchant of record, so the payment stays. The start of the purchase will go to the agent.
Carrier
Is paid for utilization, and an agent that books ahead creates notice a fleet can sell. The exposure is a contract that hands the notice to the platform that received it.
Sovereign
Holds the customer and the network, and carries the advertising and deal lines itself. The choice is to sell agents a ranking product, reprice visibility, or open agent interfaces where a human approves every commitment.
Created or transferred
The test for each position is incremental contribution. An agent that brings an order the group would never have won has created contribution, and a fee for it is an acquisition cost. An agent that brings a customer who would have opened the app anyway has moved contribution, and a fee for it is a toll.
Referral attribution says where an order came from. It cannot say which of the two it was. A group needs to measure incremental contribution after acquisition cost before it signs terms with an agent, and before a platform signs terms with it.
Memory raises the stakes. If a customer's agent holds the record of what arrived late and what was refunded, the platform no longer holds the part of the relationship that makes the next order easy.
The board's questions
- A Tenant should ask which platform charges would survive a ranking by price, delivery time and rating, and what it would pay an agent instead.
- A Merchant should ask whether it can tell an agent-sourced customer who is new from one it already had.
- A Carrier should ask whether its terms pay for flexibility, reserved capacity and committed service levels, and who holds the data that proves each.
- A Sovereign should ask which of its own revenue lines an agent would price down first, and whether it will move before the terms are set.
Scenario work, a position and agent-ready interfaces
Scenario work on what is exposed in your margin, a position for the group on agents and platforms, and agent-ready catalogues and interfaces.
We set the position through Digital, AI & Agentic Commerce. The margin scenarios draw on the Office of the CFO, and every group starts with the Control Diagnostic.
Sources
- talabat, Q2 2026 results, released 12 August 2026: advertising and listing fees at 3.4% of GMV; marketplace lines totalling 24.3 points of GMV.
- ChatGPT checkout fee for Shopify merchants: eCommerceNews AU, 3 February 2026; scaled back per an OpenAI post of 24 March 2026, as quoted in the press.
- Jahez Group, Q2 2026 results: Logi adjusted EBITDA from SAR 2.0m to SAR 14.2m.
- One Doha restaurant's talabat statements, 476 orders over five months, as reported in Moneyball Finance, 16 October 2026.
- Google Universal Commerce Protocol and OpenAI Agentic Commerce Protocol documentation, read 18 October 2026.
- Scenario arithmetic worked by At Par.
Related
- When AI orders dinner, who earns the margin?One restaurant's statements show where the money in a dinner order goes. Visibility is the first line an agent puts at risk, and each position on the matrix holds a different piece.
- Who keeps the value of a 6 pm delivery window?Agents will book delivery ahead of time. Whether the notice reaches the fleet, and who keeps the saving, is a question of position on the Control Matrix.
- Digital, AI & Agentic CommerceYour technology plan, set up and connected, including AI in operations and agent-ready commerce.
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.