Add up what the work would cost you over a set period: one-time fixes plus monthly upkeep. Divide by what one order contributes after product cost, shipping, fees and expected returns. That’s the number of extra orders the work has to bring in to break even.
Then compare it with the AI-channel orders your own Shopify admin already reports. If it needs a jump you can’t believe, shrink the scope or skip it. If it looks plausible, run a bounded test on a few products, with win, loss and inconclusive rules written down before you start. Every input is yours. No platform publishes conversion rates or traffic for your store, and the figures in our worked example are labelled hypotheticals.
Why your own numbers, not industry ones
You’ll see headline figures for AI shopping everywhere: traffic growth, conversion rates, market size. None of them tells you what your products will sell through an assistant. They come from other stores, other categories, other countries and other measurement methods.
The only outcome numbers you can trust for this decision are your own. Shopify gives you a place to find them. Its Agentic section reports, for each AI channel, sales, orders, online store sessions and online store conversion, over a date range you choose. Shopify says the sales figure combines referral-based sales and sales through direct checkout. Those reports aren’t yet available for headless stores.
Our own research can’t supply an input either. In our supplied-snippet simulation, vague product copy was almost never picked and stated facts got a product considered. That tells you which kind of work is worth doing. It measures picks by models from a shortlist we handed over, not visits or orders, so it can’t tell you how many extra orders the work will earn.
What the work actually costs
Before the arithmetic, list the work. Most of it is time, and the biggest line depends on which route you’re on.
| Work | When you need it | What the documentation says |
|---|---|---|
| Becoming eligible | Your Agentic section shows “Ineligible” | Shopify lists plan, private-mode, B2B-only and account-standing reasons, each with a fix. Account standing builds over time; no threshold is published |
| Fixing product facts | Always, for the products you choose | No platform fee. Time to rewrite, plus the cost of measuring anything you haven’t measured |
| The Shopify Catalog route to ChatGPT | You’re on Shopify and eligible | OpenAI says Shopify product data is already integrated, with “no additional work” required |
| A direct product feed to OpenAI | You want more control than Shopify Catalog gives | OpenAI’s merchant page: apply, then join a waitlist |
| Google’s UCP checkout | You want checkout inside AI Mode and Gemini | Google requires the technical implementation before onboarding, then selection. Select merchants only. Budget it as a development project |
| Upkeep | Every month you stay in | Re-reading records, keeping price and stock agreed, answering new customer questions |
And one cost that’s absent: OpenAI’s merchant page says there are no fees on purchases that start in ChatGPT. Other routes have their own terms. Check them for your store.
For most small Shopify stores, the realistic project is the second row plus upkeep: fixing facts on a handful of products, on the route you already have.
The worksheet
Fill in the right-hand column with your own figures. Where you don’t know one, write a range and run the worksheet twice.
Part A: cost over your horizon
| Line | Your input | How to work it out |
|---|---|---|
| A1. Products in scope | Start with 3 to 5. More products means more cost before you learn anything | |
| A2. Hours per product | Reading the record, rewriting facts with units and conditions, checking the result | |
| A3. Value of an hour | What your time, or a contractor’s, actually costs | |
| A4. Out-of-pocket per product | Measurement, testing, new photos. Zero if you already have the facts | |
| A5. Store-level hours | Policies, eligibility fixes, Catalog mapping review. Once, not per product | |
| A6. One-time cost | A1 × A2 × A3 + A1 × A4 + A5 × A3 | |
| A7. Upkeep hours per month | Re-checking records and keeping data true | |
| A8. Monthly cost | A7 × A3, plus any app or feed fees you’d add | |
| A9. Horizon in months | How long you’ll give it before deciding. Pick it now | |
| A10. Total cost | A6 + A8 × A9 |
Part B: what one order contributes
| Line | Your input | Notes |
|---|---|---|
| B1. Average selling price | For the products in scope, after discounts | |
| B2. Product cost | Landed cost per unit | |
| B3. Shipping you pay | Per order, not per unit, if most orders are one unit | |
| B4. Payment and platform fees | Per order, from your own payouts | |
| B5. Expected return and refund cost | Your return rate × what a return costs you | |
| B6. Contribution per order | B1 − B2 − B3 − B4 − B5 |
Leave advertising out of B. This worksheet asks whether an organic change pays for itself.
Part C: break-even
| Line | Formula | What it means |
|---|---|---|
| C1. Break-even extra orders | A10 ÷ B6, rounded up | Orders the work must add over the horizon |
| C2. Per month | C1 ÷ A9 | The monthly lift you’d need |
| C3. Your current AI-channel orders per month | From Sales channels → Agentic | Your baseline, before any change |
| C4. Required multiple | (C3 + C2) ÷ C3 | How many times your current AI-channel volume you’d need |
| C5. Your AI-channel conversion | From the same report | Sessions that completed checkout ÷ sessions |
| C6. Extra sessions needed | C1 ÷ C5 | The qualified traffic the work would have to create |
Two rules for reading Part C.
Only extra orders count. If some AI-channel orders would have happened anyway, they don’t pay for the work. That’s why Part D compares changed products with unchanged ones.
If C3 is zero, start somewhere else. A multiple of zero can’t be computed, and zero AI-channel orders usually means the question is eligibility or recommendation, not break-even. Our overview of agentic commerce helps you find which.
Part D: a bounded test
If C4 looks plausible, test before you roll out.
| Element | What to write down before you start |
|---|---|
| Changed products | The 3 to 5 products you’ll fix |
| Unchanged products | Similar products, by price and past sales, that you won’t touch during the test |
| What you change | Product facts only. No price changes, sales or new ads on either group during the test |
| Measure | AI-channel orders and sessions, before and after, for both groups |
| Window | Your horizon from A9, with a baseline period of the same length before it |
| Win | Changed products gain orders at or above C2 per month, and unchanged products don’t gain the same |
| Loss | Changed products gain less than C2, or gain no more than unchanged products |
| Inconclusive | Too few orders in either group to tell (set your own minimum before you start), or something else changed: a stockout, a viral post, a platform change |
Shopify’s help page describes the performance reports by channel. We haven’t verified whether they break down by product. If you can’t separate changed and unchanged products in your reports, the test can only be read at store level, which is much weaker. Say so when you write up the result.
A worked example, every number hypothetical
Every figure below is invented to show the arithmetic. It isn’t a benchmark, an average or a real store. Swap in your own.
The hypothetical store sells one product line at USD 80. Its contribution per order:
| Line | Hypothetical value (USD) |
|---|---|
| B1. Selling price | 80 |
| B2. Product cost | 28 |
| B3. Shipping paid | 9 |
| B4. Fees | 3 |
| B5. Expected returns | 4 |
| B6. Contribution | 36 |
It values an hour at USD 50, gives the test three months, and its Agentic report shows a hypothetical 4 AI-channel orders a month at a hypothetical 2% conversion. It compares three scopes:
| Line | Lean | Base | Broad |
|---|---|---|---|
| Products × hours each | 2 × 1.5 | 5 × 2 | 12 × 2.5 |
| Out-of-pocket per product | 0 | 30 | 60 |
| Store-level hours | 2 | 4 | 8 |
| One-time cost (A6) | 250 | 850 | 2,620 |
| Monthly upkeep (A8) | 50 | 100 | 200 |
| Total over 3 months (A10) | 400 | 1,150 | 3,220 |
| Break-even extra orders (C1) | 12 | 32 | 90 |
| Per month (C2) | 4.0 | 10.7 | 30.0 |
| Required multiple of 4 a month (C4) | 2.0× | 3.7× | 8.5× |
| Extra sessions at 2% (C6) | 600 | 1,600 | 4,500 |
Read across the bottom rows. The lean scope needs the store to double its AI-channel orders for three months. The broad scope needs eight and a half times as many. With nothing but this store’s own baseline to go on, the broad scope is a bet, not a plan. The lean scope is a test it can afford to lose.
The horizon matters as much as the scope. Stretch the base scope to 12 months and the total becomes USD 2,050: 850 once, plus 100 a month. That needs 57 extra orders, about 4.75 a month instead of 10.7. But over 12 months the upkeep (USD 1,200) costs more than the one-time fixes, so a longer horizon only keeps helping while upkeep stays small.
The decision this hypothetical store should make: run the lean test first. Fix two products well, leave similar ones alone, and read the result after three months against the rules in Part D. Scale only if the changed products clear 4 extra orders a month and the unchanged ones don’t.
What the worksheet leaves out, on purpose
Benefits outside AI channels. Clear facts with units and conditions also help shoppers on your own product page, and may reduce returns from people who bought the wrong thing. That’s plausible, and we haven’t measured it. Treat it as a tie-breaker, not an input. If you want to count it, measure it separately.
Visibility tools. A tool that reports how often an assistant mentions you isn’t measuring orders, and Shopify itself says its own search preview is “a directional signal, not an exact prediction” because assistants re-rank results. Use mentions to find problems. Use orders to judge payback.
Checkout programs. In-assistant checkout may change conversion for some stores. Whether it’s available to you depends on the route, your market and your enrolment. Our guide to AI shopping carts maps what you control there. Don’t put a checkout uplift into C5 until you’ve seen it in your own report.
Your next step
Open Sales channels → Agentic in your Shopify admin. Set the date range to the last three months and write down orders, sessions and conversion for each AI channel. That’s line C3 and line C5. Then fill in Part B for one product line from your own payouts, and Part A for the lean scope only. If the required multiple in C4 makes you laugh, don’t start. If it doesn’t, pick your two products and write the Part D rules down today.