Quick answer: Across Tolstoy's benchmark stores, median AOV (items after discounts, without shipping and tax) runs from $60 in Food & Beverages (112 stores) to $217 in Home (120). The median share of orders with a refund runs from 2.1% in Food & Beverages (102 stores) to 5.8% in Fashion (319). These are medians, not targets: compare a store with its own past, then its vertical. Windows and exclusions: see the table.
Why measuring the right thing matters
This post is for an AI agent that builds an online store, and for the person who directs it. Once the store ships, the next question is whether it works, and the agent answers with whatever it measured. With nothing measured, it cannot tell a pricing problem from a returns problem, or a good month from a bad one. With the wrong thing measured, it gives a confident wrong answer, in one of four ways:
- The wrong comparison. The same AOV reads differently by vertical: $95, the 25th percentile of Home stores, is above the Beauty & Wellness median ($62) and below the Fashion median ($122). An agent that judges a store against the wrong vertical, or compares an AOV that includes shipping with a benchmark that does not, recommends the wrong change.
- The wrong window. A refund rate taken on last week's orders is not comparable with the table's, where every order had 30 to 89 days for a refund.
- The wrong unit. A step rate built on event counts instead of shoppers counts every repeated add to cart as if it were another shopper.
- The wrong inference. A feature whose users buy more has not been shown to work; its users may have been more likely to buy anyway.
Each of these produces a number that looks like a result and points the store at the wrong fix.
Benchmarks by vertical: AOV and refund rate
These are medians across stores that send Tolstoy their orders. They are published in Tolstoy's ecommerce-analytics agent skill and were read from Tolstoy's warehouse on 2026-11-16.
| Vertical | Median AOV | AOV middle half | AOV stores | Median refund rate | Refund middle half | Refund stores |
|---|---|---|---|---|---|---|
| Beauty & Wellness | $62 | $42–$105 | 621 | 2.2% | 1.3–3.9% | 575 |
| Fashion | $122 | $76–$181 | 332 | 5.8% | 2.8–11.8% | 319 |
| Accessories | $87 | $54–$160 | 154 | 3.8% | 1.7–6.7% | 136 |
| Home | $217 | $95–$582 | 120 | 4.3% | 2.5–6.4% | 109 |
| Food & Beverages | $60 | $44–$101 | 112 | 2.1% | 1.2–3.8% | 102 |
| Health | $69 | $44–$94 | 84 | 2.6% | 1.5–4.3% | 77 |
| Sports & Outdoors | $129 | $61–$196 | 57 | 2.7% | 1.9–5.2% | 55 |
| Baby & Kids | $89 | $60–$168 | 51 | 3.4% | 2.1–5.2% | 51 |
| Pets | $61 | $33–$140 | 30 | — | — | — |
Population and definitions:
- Stores: stores that send Tolstoy their orders, each store once, with all its orders in USD, at least 30 orders in the applicable window, and a vertical from Tolstoy's classification of the store. Stores that joined after April 2026 have no classification. A vertical with fewer than 30 stores is left out. The AOV and refund columns have separate store counts.
- Middle half: the 25th to the 75th percentile of stores.
- AOV: each store's revenue ÷ its orders, for orders from 2026-08-18 to 2026-11-15 (90 UTC days). Revenue is the order's items after discounts, without shipping and tax.
- Refund rate: each store's share of orders with a refund above zero, partial or full. Orders from 2026-08-18 to 2026-10-16, refunds up to 2026-11-15, so each order had 30 to 89 days for a refund. Stores with no refund in the window (88 of 1,512) are left out, because Tolstoy may not receive their refunds.
- Pets: no refund benchmark is published.
- No conversion-rate benchmark: Tolstoy does not see a store's total visits, so it cannot compute a store's conversion rate.
What the table shows (descriptive reading)
This section is a descriptive reading of the published table: what its medians and quartiles show for these stores and windows. It does not explain why a vertical looks the way it does, it does not describe the full distribution beyond the 25th percentile, median and 75th percentile, and none of it is evidence that any feature or tool changes these numbers.
Inside a vertical, AOV spreads widely; Home most of all. Home has the highest median AOV ($217) and the widest middle half ($95 to $582): a Home store at the 75th percentile has more than six times the AOV of one at the 25th. The Home median alone therefore says little about where one Home store should sit. Health has the narrowest AOV middle half ($44 to $94, a $50 span).
Four verticals have close AOV medians. Food & Beverages ($60), Pets ($61), Beauty & Wellness ($62) and Health ($69) have medians within $9 of each other, and each of these four verticals' middle half contains all four medians. The table does not show how far the store distributions overlap beyond that. It does show that for these four the gap between medians is smaller than the spread inside each vertical, whose middle half spans $50 or more.
Refund medians differ by vertical, and Fashion's quartiles are the highest. Fashion's median refund rate (5.8%) is more than twice Beauty & Wellness's (2.2%) and Food & Beverages's (2.1%). Fashion's 25th percentile (2.8%) is above the medians of Food & Beverages, Beauty & Wellness, Health (2.6%) and Sports & Outdoors (2.7%), and its 75th percentile (11.8%) is the highest in the table; its middle half, 9.0 points wide, is also the widest. Refund rate here counts orders with any refund, partial or full; it does not say how many dollars were refunded.
AOV and refund rate rank verticals differently. Home has the highest median AOV but a lower median refund rate (4.3%) than Fashion (5.8%). Sports & Outdoors has the second-highest median AOV ($129) and the fourth-lowest of the eight median refund rates (2.7%). Food & Beverages has the lowest median of both ($60 and 2.1%). In this table, one metric does not stand in for the other, so track both.
Some rows rest on few stores. Beauty & Wellness has 621 stores for AOV. Pets has 30, the minimum for a vertical to be published, and no refund benchmark; Baby & Kids (51 stores for each metric) and Sports & Outdoors (57 and 55) are also small. Quartiles of 30 to 57 stores are less settled than those of several hundred; read them as rough.
The windows and exclusions are part of the numbers. AOV uses orders from 2026-08-18 to 2026-11-15; refund rate uses orders from 2026-08-18 to 2026-10-16, with refunds counted to 2026-11-15, so every order had 30 to 89 days in which a refund could be recorded. A refund rate on orders that had less time is measured over a shorter window: on the same orders it can only count the same or fewer refunds, so it is not on this table's definition. Compare refund rates on orders of the same age. The 88 of 1,512 stores that reported no refund are left out; removing the lowest values can keep a median in place or raise it, never lower it. Every store here sells in USD, had at least 30 orders in the window and did not join after April 2026; a new store with a handful of orders is outside this population.
How an agent should use these numbers
Compare the store with its own previous period first, of the same length, with the same weekdays, and with no sale or launch in only one of the two. Then match the definitions before placing the store in its vertical's middle half: AOV on items after discounts, without shipping and tax; refund rate as the share of orders with any refund, on orders that had 30 to 89 days; USD only. A number outside the middle half is worth a look at the definitions first, then at prices and returns. It is not a verdict, not a target, and not proof that anything worked.
What to measure, and why
Revenue = sessions × conversion rate × AOV, where conversion rate is orders ÷ sessions and AOV is revenue ÷ orders. A change can raise revenue only through one of those three terms, so name the term a change is meant to move before shipping it. Refund rate sits beside them: it is the share of orders that were partly or fully refunded, not the share of revenue refunded. For refunded dollars, use the refund amounts in the store's order data.
| Metric | Formula | Take it from |
|---|---|---|
| Orders, revenue | Count and sum of orders in the period | Store platform |
| Average order value (AOV) | Revenue ÷ orders | Store platform |
| Conversion rate | Orders ÷ sessions | Store platform |
| Buyer rate | Shoppers who ordered ÷ shoppers who visited | Event counts (shoppers) |
| Orders per buyer | Order events ÷ order shoppers | Event counts |
| Step rate | Shoppers at a step ÷ shoppers at the step before | Event counts (shoppers) |
| Refund rate | Orders with any refund ÷ orders | Store platform |
Each of these rates is meaningful because its numerator and denominator are defined units over the same period. The trap is mixing them. Conversion rate divides orders by sessions from the store platform; buyer rate divides shoppers by shoppers from event counts. They answer different questions, so neither substitutes for the other, and a change in one says nothing exact about the other. Raw event counts are not shoppers: add-to-cart events ÷ product-view shoppers is not a step rate, because one shopper can add to cart many times. A ratio also needs its numerator and denominator to match in definition, population, period and currency; whether they come from the same tool matters less. Orders from the store's order data divided by sessions from its analytics are a valid conversion rate when both cover the same storefront and the same days in the same time zone. The ratio breaks on a mismatch: orders that include sales the sessions never saw, such as another sales channel, orders for one set of days against sessions for another, or revenue in two currencies added as one. And a revenue figure needs its scope stated: with or without discounts, shipping and tax.
For the storefront, a small event plan is enough: page_view when any page shows, product_view when a product page shows, add_to_cart only after the cart accepts the item, checkout_started when checkout opens, and order_completed once per order on the confirmation page. Give each name one meaning, write it down in the store project, and never rename it, because a new name starts a new count. If checkout is hosted and runs no store code, the store cannot send order_completed; take orders and buyers from the store platform instead.
What event counts can and cannot tell you
The E-commerce Analytics SDK (the TolstoySDK package) sends storefront events from the browser, and Tolstoy's report gives two numbers per event name for a period of whole UTC days, up to 180 days: events, how many times it was sent, and shoppers, how many different shoppers sent it. Every rate is computed by the agent from these counts, and each one is a ratio of two counts, not the share of one confirmed group of shoppers. That has consequences:
- It is not a strict funnel.
add_to_cartshoppers ÷product_viewshoppers does not check that the same shoppers did both. A buyer can skip a step, return in another browser, or take the first step before the period started. Ask for a period much longer than the time between the two events so that these edges stay small. - Identity is per browser, and only with consent. The SDK keeps one shopper ID per browser. Where the store has a consent tool, it sends no event until the shopper allows analytics and drops the events from before that. Counts cover only browsers that allowed analytics, and one person on two devices is two shoppers. Both sides of a ratio share these limits.
- Shoppers do not add up across reports. A shopper is counted once per report, so daily shopper counts do not sum to a weekly one. Ask for the whole period in one report.
- Zero and missing differ.
0is a measured zero. A missing name was not sent in the period, or is not among the 120 most frequent names in the report. - There is no revenue in the events. The report does not show, group or sum parameters, so an order value sent as a parameter never becomes revenue or AOV. Those come from the store's orders, as do the benchmarks above, not from the event report.
Correlation is not effect. Shoppers who choose to open a size guide often were more likely to buy anyway, so a higher buyer rate among them is a correlation, as is a before-and-after comparison around a launch. To estimate an effect, show the feature to a random half of shoppers, compare the halves, and judge the 95% range of the difference rather than each rate; when that range includes zero, there is no shown effect. The ecommerce-analytics skill gives the formula and how to credit an order to a feature.
Adding it to a store
The docs state that sending events and reading the report are free on every Tolstoy plan, the free plan included, and that events never count toward Shoppable Videos view billing. No Tolstoy widget is needed. Two facts shape the plan: sign-up and the approval of the CLI or MCP sign-in happen in a browser, so an agent without one needs its person for those steps; and the report covers whole UTC days before today, with today's events showing from about 06:30 UTC tomorrow. The SDK takes the public app key, not the secret API key. Setup, consent handling and the same-day delivery check are in the Tolstoy e-commerce analytics guide, with reference in the Tolstoy SDK, Tolstoy CLI and Connect your agent pages. Installing the ecommerce-analytics skill in the store project, and recording the event plan in its AGENTS.md, lets the next agent read the numbers with the same definitions.
Frequently asked questions
Which e-commerce metrics should a new store measure?
Measure the numbers a decision depends on: orders and revenue; average order value (revenue ÷ orders); conversion rate (orders ÷ sessions) or buyer rate (shoppers who ordered ÷ shoppers who visited); step rates such as add-to-cart shoppers ÷ product-view shoppers; and refund rate (orders with any refund, partial or full ÷ orders). Write down each denominator and period, and say whether revenue includes discounts, shipping and tax.
Are event counts and shoppers the same?
No. In the Tolstoy report, events is how many times an event was sent, and shoppers is how many different shoppers sent it. One shopper can send an event many times. The E-commerce Analytics SDK keeps one shopper ID per browser, so one person on two devices counts as two shoppers. Step and buyer rates divide shoppers by shoppers; orders per buyer divides the order events by the order shoppers.
Does Tolstoy provide a funnel, conversion rate, or revenue for custom events?
No. The report counts each event name on its own and does not link your events to orders. It gives no strict funnel, no conversion rate and no revenue for your events. Tolstoy keeps event parameters, but the report does not show, group or sum them. Take sessions, revenue, AOV and refunds from the store platform.
Is Tolstoy analytics free without a widget?
Yes. The Tolstoy developer docs state that sending events and reading the report are free on every Tolstoy plan, the free plan included, and that the E-commerce Analytics SDK and the report work with no Tolstoy widget on the store. Your events never count toward Shoppable Videos view billing.
When do today's events appear?
The report covers whole UTC days before today. Events that you send today show from about 06:30 UTC tomorrow. On the day that you send, the only check is that the browser's event requests return HTTP 200.
What is a good AOV or refund rate for my vertical?
There is no universal good number, only a typical range per vertical. These figures come from stores that send Tolstoy their orders (USD, at least 30 orders in the window, read 2026-11-16). AOV is each store's revenue divided by its orders, with revenue as order items after discounts, without shipping and tax, on orders from 2026-08-18 to 2026-11-15. The middle half of stores had AOVs of: Fashion $76–$181 around a $122 median (332 stores), Beauty & Wellness $42–$105 around $62 (621 stores), Home $95–$582 around $217 (120 stores). Refund rate is the share of a store's orders with any partial or full refund, on orders from 2026-08-18 to 2026-10-16 with refunds counted to 2026-11-15; stores that reported no refund (88 of 1,512) are left out of the refund medians. The middle half of stores had refund rates of: Fashion 2.8–11.8% around 5.8% (319 stores), Beauty & Wellness 1.3–3.9% around 2.2% (575 stores), Home 2.5–6.4% around 4.3% (109 stores). The table in this post covers all nine verticals. Inside the middle half is typical for the vertical; outside it is worth a look, not a verdict.
Do these benchmarks prove lift?
No. The benchmarks are store-level medians that describe a sample of stores that send Tolstoy their orders. They do not show the effect of any feature or tool. Only a randomized comparison, with half of shoppers who see a change and half who do not, can estimate an effect.
Create with Tolstoy
Put these ideas into practice
Turn your product ideas into AI images, videos, and shopping experiences.