Topline CVR cannot tell you what to fix. The behavior underneath it can.
Most stacks track pageviews and orders, then argue about everything in between. We instrument the site down to the module and the position, stitch behavior across sessions, and read tests, promos and merchandising with the rigor a revenue decision deserves.
Instrumentation that can tell a hero from a carousel.
The homepage is where tracking usually gives up: modules with no events, positions nobody records, and a redesign debate with no data on either side. A real tracking plan gives every component its own event and properties, built once with engineering so new modules inherit the schema.
Excerpt of the shape. The full plan covers the whole funnel, with definitions and QA status per event and property.
The same buyers, two opposite conclusions.
A real read from this work: look only at the purchase session and customers seem to skip discovery entirely. Stitch the same people across ~2 weeks and discovery turns out to be doing the selling.
Roughly a third of buying sessions never leave the product page, and most of the rest touch only a page or two. Browsing barely shows up.
Across the full journey, the majority touched homepage, listing pages and multiple product pages. The final session is where the decision lands, not where it is made.
Same warehouse, same buyers, opposite roadmaps. The difference is whether the tracking can stitch a person across sessions. And once it can, personalization stops being a guess: one brand's hero engaged ~20% of visitors overall but ~3% of visitors arriving from its loungewear ads; a hero matched to the ad took that segment to ~60%. [Loungewear brand, client unnamed]
Some products need exposure. Others need a better page. The matrix says which.
Product views against bookings per view. High take, low visibility means the product converts whoever finds it, and the fix is placement: navigation, homepage modules, email, ads. High visibility, low take means the traffic is there and the page, price or positioning is not. Each quadrant has a different owner, which is the point.
The naive read fails quietly. These are built not to.
Every one of these reads is cheap once the instrumentation above exists, and impossible to trust without it.
Instrument first. Then every read gets cheaper.
You do not need all of it, and almost nobody starts with more than two. Everything lives in your repo and your warehouse, so it is yours whether or not we are around.
What happened when we did it.
Different businesses, same approach. The full studies say how each result was measured.
First-product cohorts showed which entry purchase predicts the customers worth buying, and the media plan followed.
Once clicks and sessions reconciled, landing experiences were rebuilt by journey, and the highest-LTV category grew. Part of the 2.8x growth study.
Instrumentation, the experiment program, and what the module analytics changed, with the mechanism shown.
Get in touch.
Tell us what you are trying to figure out. If we can help, we will say how. If we are not the right fit, we will say that too, and point you somewhere better.