← All briefs
Analysis · Unit economics

Costing an AI tool per unit: three lines for your worksheet, and one caution about who sold it to you

Published 4 Nov 2026

AI tools for restaurants are now sold at published monthly prices, which means they can finally be modelled rather than guessed. Here is how to put a tool on a per-unit worksheet, and why the recruitment pitch that introduced it needs separate scrutiny.

For two years the cost of AI in a restaurant was unmodellable, because vendors quoted bespoke. That has changed. Published per-month pricing means an AI tool is now an ordinary line item, and ordinary line items can be tested against the unit they sit in.

The published numbers to anchor on

According to Nation's Restaurant News on 7 October 2026, Toast launched AI agents under its Toast IQ layer in two tiers. The Team tier, covering labour, starts at USD 60 per month plus USD 9 per employee per month. The Grow tier, covering marketing, starts at USD 699 per month.

The same report gives one performance figure with a number attached. Restaurants using the beta Scheduling Agent, which builds a labour schedule from employee availability, expected sales and overtime risk, averaged 41% fewer overtime hours per employee than non-users. Other agents listed include an SEO agent, a reviews agent, an offers agent, and in early access a menu agent, an answering agent and configurable agents built from templates.

Take the labour tier as a worked example, because it is the one with a measurable claim.

Line one: the gross cost per unit per year

For a unit with 25 employees, the Team tier at USD 60 plus USD 9 per employee is USD 285 per month, or USD 3,420 a year. The employee count is the sensitive variable, not the base fee. At 40 employees the same tier is USD 420 a month, or USD 5,040 a year.

Write the cost as a function of headcount, not as a flat number. A seasonal unit that doubles staff in peak months pays the higher figure in exactly the months its labour cost is already highest.

Line two: the offsetting saving, stated conservatively

The 41% overtime reduction is a vendor beta figure, so it should be discounted heavily and modelled as a range, not a point.

Build it as follows. Take your actual overtime hours per employee per month from payroll. Multiply by your overtime rate. Multiply by headcount. That is your annual overtime spend, which is the only pool the tool can draw from. Then run three scenarios against it: no reduction, half the claimed reduction, and the full claimed reduction. If the tool does not pay for itself at half the claim, treat the purchase as a bet rather than a saving.

The discipline matters because the claim is a per-employee overtime-hours reduction, not a total labour-cost reduction. Those are different quantities and a vendor is entitled to quote the first.

Line three: what happens at renewal and at scale

A per-employee monthly fee grows with the business. If your three-year plan takes a unit from 25 to 40 employees, the tool cost rises 47% over the same period with no change in the contract. Model the fee at your year-three headcount, not today's.

For a multi-unit operator, multiply by units and ask whether the price is per location or negotiated across the group. The published figure is a starting price for one site.

The caution: the pitch that introduced the tool

The same technology is being used on the other side of the table, to recruit you. According to Franchise Times on 29 May 2026, Hot Palette America, the parent of Pepper Lunch, has used AI for marketing copy for about two and a half years and works with a vendor to target multi-unit franchisee owners with individualised landing pages. Its chief executive Troy Hooper said the interactive prospect journey is weeks away from being driven by bots without a human lead.

The same report describes Senior Helpers, which has operators in more than 420 US territories with about 500 still available, and quotes Franchise Ventures president Hunter Stokes describing AI as an amplifier rather than a replacement for people.

An individualised landing page is a document written for you alone. It is also, in substance, a representation by the franchisor. Save it. Screenshot it. If a figure that influenced your decision appeared only on a page generated for you, you want a copy of it before the page changes.

The worksheet in short

  • Cost line, expressed as base fee plus per-employee fee at year-three headcount.
  • Saving line, expressed as three scenarios against your actual overtime pool.
  • Decision rule, written before you see the demo: the minimum saving at which you would buy.
  • Evidence file, holding every personalised page, email and figure used to sell you the system.

What we do not know

The 41% overtime figure is a beta average reported by the vendor, with no sample size, no restaurant profile and no independent verification in the source. It should not be used as an expected outcome.

The Nation's Restaurant News report places the USD 699 starting price immediately after the marketing-tier paragraph, so the tier it applies to is inferred from position rather than stated unambiguously. Confirm the tier with the vendor before modelling it.

Neither source gives a franchise-system-wide cost for any AI stack, and neither reports what proportion of these tools is paid by the franchisor versus the franchisee. That split is the single most important unknown for a candidate, and it is not in these reports.

This article is information, not investment advice.

Sources

  1. https://www.nrn.com/restaurant-technology/toast-unveils-an-array-of-new-ai-agents-for-restaurants
  2. https://www.franchisetimes.com/franchise_technology/franchisors-talk-practical-uses-of-ai-to-reach-prospective-franchisees/article_84eddac9-93da-464f-8a1b-b91934d15902.html

Written with AI research assistance and published with the sources it was built from. Not investment, legal or financial advice.