The Truth About the Ecommerce Conversion Rate Formula: One Division, Four Choices That Change the Answer

    Orders divided by sessions looks simple until you decide what counts as each

    Published
    9 min read
    By Erik Lornovell
    Updated

    2.1% in the analytics tool, 3.4% in the weekly deck, 2.6% in the store admin. Those were the three conversion rates a mid-sized apparel store reported for the same seven days, and none of them was a mistake. Each one divided a reasonable numerator by a reasonable denominator. They simply made different choices about what to count.

    The textbook ecommerce conversion rate formula fits on a sticky note: conversions divided by visits, times 100. The truth is that the formula hides at least four decisions, and the gap between a careful and a careless version of it is often larger than the improvement a whole quarter of CRO work can deliver. This teardown takes the formula apart piece by piece, then puts it back together with a worked example and a calculator.

    The formula on the sticky note, and what it leaves out

    The common version reads like this:

    conversion rate (%) = conversions / visits x 100
    

    Every term in it is ambiguous. "Conversions" can mean orders, purchasing sessions or customers who bought at least once in the period. "Visits" can mean sessions, users, or unique visitors as defined by a tool that has its own rules for when a session ends. Then there is the question of what you throw out before dividing: bot sessions, staff browsing the store, test orders placed by the payment team, orders cancelled for fraud within an hour.

    A more honest version spells out the choices:

    session conversion rate = purchasing sessions / qualified sessions x 100
    user conversion rate    = purchasers / active users x 100
    order rate per session  = valid orders / qualified sessions x 100
    

    These three are not interchangeable. User-based rates run higher than session-based ones because one person often visits 2 or 3 times before buying, so the store shown above reported 3.4% on users and 2.6% on sessions without anyone doing anything wrong.

    Platforms also differ. Shopify's admin describes its online store conversion rate as the share of sessions that ended in an order. GA4 offers session and user key event rates, and a purchase only counts there if the purchase event fires and is marked as a key event. If the tag misses Apple Pay express checkouts, for example, GA4 undercounts orders while the store admin does not.

    The practical rule: pick one definition, write it in the dashboard subtitle, and never compare numbers built on different ones. Most "conversion dropped 20%" alarms that turn out to be false come from a quiet change in one of these terms, usually made by someone who did not know the old one.

    Teardown of the denominator: which visits deserve to count

    The denominator is where most of the distortion lives, because it is large and full of traffic that could never buy.

    Bots are the first layer. Analytics tools filter known crawlers, but scrapers, price monitors and uptime checks often slip through, especially when they run headless browsers that execute JavaScript. On small stores they can make up 5% to 15% of sessions. If that share changes from one month to the next, your conversion rate moves without a single customer behaving differently.

    Internal traffic is the second layer. Merchandisers checking product pages, support staff looking up orders, agencies testing a new theme: on a store with 30,000 monthly sessions, a team of 8 people can easily add 1,000 sessions, more than 3% of the total. Filtering by office IP addresses stopped working well once teams went remote, so many stores now rely on a staff cookie or a dedicated browser profile.

    The third layer is traffic that is technically human but not shopping. Visitors landing on a blog post from search, customers clicking a tracking link from a shipping email, people opening the returns page. Some teams exclude these and report a "shopping session" rate that only counts sessions viewing at least one product page. It is a legitimate metric for product page work, but it must never be compared with an all-sessions rate from last year.

    Session length rules matter too. GA4 ends a session after 30 minutes of inactivity by default, and a visitor who leaves a tab open over lunch and comes back to pay creates two sessions and one order. On mobile, app switching to check a banking app before paying has the same effect. None of this is wrong, but it pulls a session rate down relative to a user rate.

    Teardown of the numerator: orders, buyers or purchasing sessions

    The numerator looks safer, because orders live in the store database and are hard to fake. It still hides choices.

    Orders and purchasing sessions diverge when a customer places 2 orders in one session, which is common with stores that do not allow editing an order after payment. Someone forgets a size, checks out again, and you now have 2 orders and 1 converting session. Counting orders inflates the rate slightly; counting sessions is closer to what the formula claims to measure.

    Test orders and cancellations are the other trap. A payment integration update can produce 20 or 30 test orders in a day. Fraud screening may cancel 1% to 2% of orders within hours. Partial or full refunds a week later are a different question and normally stay in the conversion rate, because the visit did convert; they belong in a return rate.

    Then there is the question of which orders belong to the website at all. Orders placed by phone, created manually by support, or coming through a marketplace integration may appear in the store admin while having no web session behind them. If they land in the numerator while the denominator only counts web sessions, the rate drifts upward every time the phone team has a busy week.

    Draft orders and abandoned payment attempts that a platform records as pending should also stay out until they are paid.

    A clean numerator for an online store conversion rate is therefore: web orders, minus test orders, minus orders cancelled before fulfilment, deduplicated to one per session if you report a session rate.

    Worked example: the same week, three numbers

    Here is an illustrative week for a fashion store, with numbers chosen to show how far apart the versions can land.

    The analytics tool reports 48,000 sessions and the store admin shows 1,250 orders. The quick calculation gives 1,250 / 48,000 = 2.60%. Now apply the cleaning. A bot and staff audit suggests about 6% of sessions are not real shoppers, so 2,880 sessions come out, leaving 45,120. Of the 1,250 orders, 40 were test orders or fraud cancellations, leaving 1,210 valid orders. Those 1,210 orders came from 1,170 distinct sessions, because 40 customers checked out twice.

    The cleaned session conversion rate is 1,170 / 45,120 = 2.59%. The order rate per qualified session is 1,210 / 45,120 = 2.68%. On a user basis, 31,000 active users and 1,090 purchasers give 3.52%. With an average order value of 62 euros, revenue per qualified session is 1,210 x 62 / 45,120 = 1.66 euros.

    The raw and cleaned session rates happen to land close together here, because the bot share and the test orders roughly cancelled out. If next month the bot share rises to 12% and there are no test orders, the raw rate falls while real shoppers behave exactly the same.

    The user rate, at 3.52%, is the one that will end up in an investor deck, and it is 0.9 points higher than the admin figure for reasons that have nothing to do with performance.

    The calculator uses the order rate per qualified session, because most stores can export orders easily but cannot always count purchasing sessions.

    Why the blended rate lies even when the formula is clean

    A perfectly cleaned formula can still mislead, because a single store-wide rate averages very different kinds of traffic.

    Take an illustrative month A: 20,000 organic sessions converting at 3.5% give 700 orders, and 10,000 paid social sessions at 0.8% give 80. The blended rate is 780 / 30,000 = 2.60%. In month B the team improves both channels: organic rises to 3.6% on the same 20,000 sessions, giving 720 orders, and paid social rises to 0.9%. The paid budget also grows, so paid social brings 25,000 sessions and 225 orders. Blended: 945 / 45,000 = 2.10%.

    Both segments got better, total orders rose by 21%, and the headline conversion rate fell by half a point. This is the mix effect, a version of Simpson's paradox, and it is the most common reason a growing store sees its conversion rate drop right after a successful budget increase. Device mix does exactly the same thing: mobile usually converts below desktop, so a shift toward mobile traffic in the autumn drags the average down even when the mobile checkout improves.

    The cure is to report conversion by segment first and blended second. At minimum split by channel group, by device, and by new versus returning visitors. Three cuts with 2 to 6 values each fit on one screen, and they explain most month-to-month swings in a store that spends on paid acquisition. When the blended rate moves, check whether any segment moved or only the weights did.

    A checklist before the number leaves the dashboard, and the verdict

    Most errors in a published conversion rate are caught by the same handful of questions, asked every time and in the same order. Teams that skip them usually discover the problem 2 or 3 months later, when someone tries to explain a trend line that never existed. Run through this list before you publish a conversion rate anywhere:

    • The definition is written next to the number: sessions or users, orders or purchasing sessions.
    • Bot, uptime monitor and staff traffic are filtered, and the filter method has not changed since the comparison period.
    • Test orders and pre-fulfilment cancellations are removed; refunds are reported separately.
    • Phone, manual and marketplace orders are excluded if the denominator only counts web sessions.
    • The purchase tag fires for every payment method, including express wallets, and matches store orders within about 5%.
    • The period comparison uses the same session timeout and consent settings.
    • At least one segmented view sits next to the blended rate.

    The verdict comes down to a trade-off between simplicity and honesty. The raw store admin rate is easy, consistent and good enough for spotting a broken checkout within hours.

    It is also the number your platform support team will quote back to you. The cleaned, segmented rate takes an afternoon to set up and a few minutes each month to maintain, and it is the only version you should use to judge whether a redesign, a price change or a new channel actually worked. Keep both, label both, and do not let the easy number make decisions that only the honest one can support.

    Related Articles