Which of your paid channels actually brings buyers?

The conversion event is a purchase. That part is rarely in dispute, which is why ecommerce teams assume this is a solved problem for them. It usually is not, for three reasons that all look like details.

• Gross versus net. Firing on the confirmation page counts orders that get refunded, cancelled or fail capture. In a category with a thirty percent return rate, a channel ranking built on gross orders is describing a business you do not have.
• First order versus every order. For ranking acquisition, the unit should usually be the first purchase from a new customer. Counting repeat orders credits the acquisition channel for retention it did not produce, and systematically over-rates whichever channel your existing customers happen to arrive through — which is almost always brand search and email.
• Orders versus revenue. Ranking by order count treats your smallest basket and your largest one identically. For a business with a wide basket range that inverts the ranking outright.

None of these is technically hard. All of them are decisions someone has to make deliberately, and the default in every tool is the version that is easiest to instrument rather than the version that is true.

Three failures show up in almost every ecommerce measurement setup we look at.

The channel-to-revenue mismatch. Paid social reliably tops the sessions table and reliably underperforms on orders, because it acquires against a broad interest signal rather than purchase intent. Meanwhile email and returning-customer traffic sit near the bottom of the traffic report and near the top of the revenue one. Ranked by sessions, the obvious action is to spend more on the channel producing the fewest orders.

Mobile and desktop diverging. Aggregate conversion rate hides the most actionable number in ecommerce. Mobile typically carries the majority of sessions and a minority of revenue, and the gap is usually a specific, fixable checkout problem — a payment method that fails silently, a form that fights the keyboard, a shipping step that requires typing an address on a phone. Nobody finds it, because the blended figure looks acceptable.

The checkout that leaves the site. Where checkout completes on a hosted payment page or an off-site wallet flow, the purchase event frequently never fires at all, or fires only on a return redirect that a meaningful share of buyers never complete. The revenue is real and the measurement is missing, and it is missing unevenly by payment method — which means it is missing unevenly by device and by audience.

Deployment for an ecommerce business is shaped around the purchase definition and the paths that reach it.

• The purchase event, defined explicitly as gross or net, first-order or all-orders, count or value — decided with you rather than defaulted.
• Every checkout path walked and verified, including hosted payment pages, wallet flows and any route that leaves the site and returns.
• Terminal event delivery confirmed, so purchases fired immediately before a redirect are not lost to the navigation.
• Channel, campaign and entry-page ranking against the purchase event, with device held as a first-class split rather than a filter.
• Behavioural scoring across the pre-purchase sessions, so multi-visit consideration is visible instead of collapsing into last click.

The top of the page is source ranking against purchases, with rate and value alongside so that a high-volume low-value channel cannot hide behind order count.

Immediately below it, checkout progression split by device — because that is where the money leaks and because a blended number will never surface it. Then entry-page performance against purchase, which is how you find the category or landing page that is drawing traffic that never had purchase intent.

Session trend sits low. It moves for reasons nobody acts on, and putting it at the top is how a reporting system trains its readers to look at the wrong thing first.