What an Agent Ready Catalog Costs and How to Make the Budget Case

The cost to make a catalog agent ready for e-commerce AI agents is mostly people's time. In our worked model, labor makes up 87% to 95% of the first-year total and the feed subscription covers the rest. A hypothetical 5,000-SKU store on a hosted platform lands near $15,000 in year one and about $5,300 a year after that. The tooling figures come from one vendor's public price list. Every labor figure is a labeled assumption, because we found no published per-SKU benchmark for this work in the documentation and pricing pages we read on October 7, 2026. Our position: fund the catalog work now, and treat a custom checkout build as a separate, later request.

What are you paying for?

Five lines make up the budget, and only one of them is a subscription. Data cleanup is the hours spent fixing titles, identifiers, variant groupings and missing attributes until every row in the feed is complete. Feed tooling is the software that maps and exports that data. Public list prices from DataFeedWatch are $64 a month for 1,000 SKUs, $84 for 5,000 and $239 for 30,000 on monthly billing, with a custom quote at 100,000 SKUs and a 17% discount shown for yearly billing. Schema work is the structured data on the product page. Checkout integration is optional and gets its own section below. Upkeep never ends, since the get-started guide from OpenAI says, "It is generally recommended to provide the entire feed once a day via file upload, and then send updates throughout the day via the API."

What does a worked estimate look like by catalog size?

A cost estimate for making a product catalog agent-ready is mostly an estimate of AI feed engineering and catalog hours, so it needs an hourly rate before anything else. The Bureau of Labor Statistics reports 2025 median pay of $64.44 per hour for software developers, quality assurance analysts and testers in the US. Everything past that figure is our assumption:

• Engineering time costs $90 an hour, the BLS median multiplied by 1.4 for benefits and overhead, then rounded.
• Catalog specialist time costs $45 an hour.
• Cleanup takes two minutes of manual time per SKU on average, after bulk rules have done what they can.
• Schema work takes 16, 24 and 40 hours across the three catalog sizes, and upkeep takes 4, 8 and 20 hours a month.

The table is a hypothetical example for a store on a hosted platform that needs no custom checkout build.

Tooling is the smallest line at every size, between 5% and 13% of the year-one total. A team that spends its evaluation time comparing subscription prices is negotiating over the smallest number on the page. Before you copy any vendor's price into your own version, confirm that the tool exports to the destinations you plan to use.

What does that come to per SKU?

Cost per SKU for structured data and feed optimization, the figure an ecommerce engineering lead will be asked to defend, falls as the catalog grows: $5.87 at 1,000 SKUs, $3.00 at 5,000 and $2.08 at 30,000 in year one of this model. Cleanup is the only line that grows in step with the catalog, at $1.50 per SKU under our two-minute assumption. Schema lives in the product template, so its cost follows the number of templates. Tooling gets cheaper per unit on the published tiers, from 6.4 cents per SKU per month at 1,000 SKUs to under a cent at 30,000. From year two the figure drops to $2.93, $1.07 and $0.46 per SKU.

The two-minute assumption carries the whole model, so test it before anyone signs off. Time a specialist on 100 representative SKUs, then multiply. If the sample runs at five minutes a row, the cleanup line at 30,000 SKUs becomes $112,500, and that is a finding worth having before the budget meeting.

Where do the engineering hours go?

Agentic commerce catalog engineering cost, in this model, is the schema line plus any feed plumbing your tool does not cover. The merchant listing documentation from Google requires three properties on a product (name, image and offers) and a price with its currency on the offer. Its recommended list is where the hours go, because availability, shipping details, return policy, GTIN and brand often have to be pulled from systems that disagree with each other. Google says a page with Product markup can be eligible for display in merchant listing experiences on Google Search, and we make no claim that markup changes what an assistant recommends. Our Schema Generator is a further resource for drafting it.

What does a checkout integration add?

AI agent shopping catalog integration cost depends on which API and schema you have to implement yourself. On Shopify the answer may be none. The help center from Shopify calls its ChatGPT channel "a discovery-focused referrer platform", says it is "active by default for eligible stores" and states, "There are no fees associated with selling in ChatGPT." Shoppers finish the purchase on the store's own checkout.

A custom stack is a build wherever a program offers checkout inside the assistant. The Agentic Checkout Spec from OpenAI lists five REST endpoints for creating, updating, completing, cancelling and retrieving a checkout session, plus order events published to a webhook, and it tells merchants, "Authenticate every request, verify signatures, enforce idempotency, validate inputs, and support safe retries." Confirm that spec's status before you scope it, because the merchant page from OpenAI now says, "We're moving away from a standalone Instant Checkout experience in ChatGPT and prioritizing better shopping discovery and merchant-owned checkout experiences." Purchases, it adds, are completed on merchant-owned websites or apps. The native checkout guide from Google for the Universal Commerce Protocol also lists five checkout session endpoints. OpenAI's merchant page states, "There are no fees on purchases that start in ChatGPT." We found no platform fee for merchants stated in Google's guide. Our assumption for one protocol is 240 engineering hours, or $21,600 at $90 an hour. It belongs on its own line because it is larger than the entire year-one catalog budget at 5,000 SKUs.

How do you turn the cost into a budget case?

An agentic commerce budget business case is easier to defend as a break-even order count than as a revenue forecast. Take a hypothetical 5,000-SKU store with 40,000 orders a year, an $80 average order and a 30% contribution margin, which leaves $24 per order. The $14,988 year-one cost needs 625 incremental orders to pay back, about 52 a month, or 1.6% of current volume. Year two needs 222. Add the $21,600 checkout build and the first-year bar rises to 1,525 orders, or 3.8% of volume.

That framing keeps the scenario conservative. It counts only new orders that arrive through agent referrals and gives no credit for any benefit the cleaned feed brings to channels you already run. It also hands the approver one number to accept or reject. To see whether those orders are arriving, pair order data with a view of how often assistants mention you: AI share of voice is the metric, our notes on where prompt-based tracking breaks cover its limits, and the AI Visibility Checker is a further resource for a first read. Monitoring is its own cost, which we model in the budget behind an AI visibility program.

What changes the case in 2026?

For an agentic commerce budget business case in 2026, access rules decide when the money should be spent. OpenAI's get-started page carries the notice, "Onboarding product feeds in ChatGPT is currently available to approved partners." The Universal Commerce Protocol page from Google invites merchants to join a waitlist, tells them to use their existing Merchant Center shopping feeds and states, "You remain the Merchant of Record."

So the two halves of the budget carry different risk. Cleanup, schema and tooling improve a feed and a product page you already operate, whatever any single assistant decides about access. A custom checkout build is a bet on a specific program admitting you and keeping its checkout route open. Ask for the first amount now, on the break-even above. Bring the second request when you hold an approval and can show agent-referred orders in your own data. If you want a second pair of eyes on the model, our AI search optimization team does this work.

Sources

• DataFeedWatch, Pricing
• OpenAI, Get started
• OpenAI, Agentic Checkout Spec
• OpenAI, Power product discovery in ChatGPT
• Bureau of Labor Statistics, Occupational Outlook Handbook, Software Developers, Quality Assurance Analysts, and Testers
• Google, Merchant listing structured data
• Google, Native checkout REST API implementation
• Google, Universal Commerce Protocol
• Shopify, Selling on ChatGPT