Three ways to pay for an AI agent, and what each one does to a franchise system

Three contracts appeared in one week

Between 6 and 8 October 2026, three ways of paying for AI in a multi-unit business showed up in public.

On 7 October 2026, Restaurant Technology News reported that Kutlerri.ai raised USD 4 million from Narahari Investments for managed restaurant AI agents, with fees tied to the orders its catering agent generates. Its reference customer, the fast-casual group RASA in the Washington D.C. area, is reported to have produced USD 450,000 in new catering revenue over ten months without an additional hire, which the vendor says exceeded RASA spending on the service by more than ten times. These are company-reported figures.

On 8 October 2026, a release described Deliverect and Yext connecting live store data, including menus, hours, availability and allergens, into the Yext Knowledge Graph so it flows to search engines, maps and AI assistants. Deliverect is described as covering more than 80,000 restaurant locations, with the About section saying more than 85,000, across more than 70 countries, with 30 million API calls a day. This is platform infrastructure, normally sold as a subscription.

On 6 October 2026, an article on Franchising.com by Joe Smithberg of Right at Home described an internal build: a dashboard called RightFocus that merges operational data from several systems and lets owners benchmark against comparable offices in the same system.

Three vendors, three commercial shapes. The shape matters more than the price.

Shape one: the subscription

A fixed fee per unit per month is the default in franchising because it is easy to mandate and easy to budget. It is also the shape that creates the most friction, because the franchisee pays the same whether the tool works or not.

The subscription suits infrastructure. Keeping store hours and menus correct across search, maps and AI assistants is not optional and does not scale with sales, so a flat fee matches the cost structure. The question to settle is not the fee but the system of record: whoever controls the source feed controls what an AI assistant tells a customer about a unit.

Shape two: the share of results

Outcome pricing moves the risk to the vendor and reads as the fairer deal. It also introduces the hardest measurement problem in the stack: attribution.

Three clauses decide whether outcome pricing is genuinely outcome pricing.

• Exclusions. Does a repeat order from an existing customer count as generated revenue. If it does, part of the fee is being charged on business the unit already had.
• The floor. If there is a minimum monthly fee, the contract is a subscription with a results label on it.
• Cancellations and refunds. If a cancelled catering order has already triggered a fee, the unit pays for revenue it never received.

The published report on Kutlerri.ai does not specify how existing customers, repeat orders, cancellations or refunds are treated. That is not a criticism of the vendor; it is the clause list any buyer has to ask for.

Shape three: the internal build

The third shape costs development time instead of licence fees, and it owns the output. The benchmarking dashboard is the clearest example, because the underlying data already exists inside the franchisor: royalty reporting already collects unit-level sales.

A franchisor that returns that data to operators in comparative form creates something no vendor can sell, since no vendor has the comparison set. The constraint is not technology. It is whether the franchisor is willing to show an operator where their unit ranks.

The clause that outranks the pricing

Across all three shapes, one term decides long-run value: who owns the data and the customer list the agent produces.

A catering agent builds a list of local corporate buyers. A listings platform holds the canonical record of every unit. A dashboard accumulates years of unit-level comparison. In a franchise system, each of those is a system asset that outlives the vendor contract. If the contract is silent, the vendor keeps it, and the system pays again at renewal from a weaker position.

A short checklist before signing

• Which shape is this, honestly: subscription, outcome share, or build.
• If outcome share: what is excluded, what is the floor, and how are cancellations handled.
• Who is the system of record, and who answers for a wrong AI answer about a unit.
• Who owns the data and the customer list at termination.
• Can the franchisee see the benchmark, or only their own numbers.

What we do not know

We do not know Kutlerri.ai pricing levels, its contract exclusions, or whether the RASA result repeats in another system, because the report gives neither the fee schedule nor an independent verification of the USD 450,000 figure.

We do not know how many of the 80,000 or 85,000 locations in the Deliverect figures are franchised units, nor how the two figures reconcile, since the release states both.

We do not know what RightFocus cost to build, nor whether its benchmarking changed unit performance, because no outcome figures were published.

We do not know how any of these tools performs in Vietnam or elsewhere in Southeast Asia, where delivery platforms, labour cost and menu volatility differ from the United States, and none of the three sources addresses the region.

And we do not know whether outcome-based pricing survives contact with a downturn, when the attribution argument becomes expensive for both sides.

This article compiles published reporting and company releases. Vendor figures have not been independently verified. Nothing here is investment advice.

Get next week's brief by email. Free.