Conversion rate optimization for ecommerce

We build the funnel from your own analytics: sessions, product views, add to cart, checkout started, order placed, split by device and by traffic source. Then we compare each step against the same step for your other devices and sources, because the useful comparison is you against yourself. Usually one gap stands out, and it is often not the one the team expected. A store convinced its checkout was the problem sometimes finds that most visitors never reach a product page at all.

Numbers say where, recordings say why. We watch sessions on the step that leaks, especially on phones, and pay attention to hesitation, repeated scrolling, pinch zooming on an image or a size chart, taps on things that are not links, and the point at which people leave. Alongside that we read on-site search queries, support tickets and return reasons. By the end there is a list of specific reasons written in the customer's language rather than a list of preferences.

The product page has one job: answer everything that stands between interest and a decision. Sizing with real measurements, materials, what is in the box, compatibility, delivery cost and date, returns terms, and photography that shows scale and use rather than only the product on white. We put the answers where the question occurs on a small screen rather than in a tab nobody opens, and cut anything competing for attention with the add to cart button.

Speed work is worth doing on the templates that carry revenue, which is normally the product page and the collection page, on mobile. We measure on a mid-range phone rather than a fast laptop, then fix in order of damage: oversized images, scripts from apps that block rendering, fonts that hide text while they load, and anything third party running on every page for the benefit of one. A faster homepage nobody enters through changes nothing.

A split test needs enough visitors and enough orders to separate a real difference from noise, and on most stores only a few templates qualify. Where the volume exists, we test properly and let it run to a conclusion instead of stopping at the first encouraging day. Where it does not, we make the change, note the date, and compare a clean period before against a clean period after, accepting that seasonality makes it less certain. Pretending a small test is conclusive is worse than not testing.

Six months of changes with no record is why most stores cannot tell what worked. Every change gets logged with the date, the reason, the page affected and the numbers before it. When something moves, the log explains it. When something breaks, the log shortens the search. It also stops the same idea being retried every year by someone new, and it means the work does not walk out of the door if we stop working together.