How to Optimize Your eCommerce Funnel Using GA4 Insights

Funnel Exploration, not Path Exploration. Funnel Exploration lets you define the steps yourself, then reports completion and drop off at each one, which is what funnel optimisation requires. Path Exploration answers a different question, showing where people went next from a chosen point, and on a store with many templates it produces a lot of noise. Use Funnel Exploration when you know the journey you want to measure, and Path Exploration only when you genuinely do not know what route visitors are taking.

Use the reserved ecommerce events as steps: view_item, add_to_cart, begin_checkout, add_payment_info and purchase. Set the funnel to open rather than closed if you want to include people who entered mid journey, which is common for shoppers arriving from a saved cart or an email. Then segment it, because a blended funnel averages away the finding. Mobile and desktop, new and returning, and paid and organic visitors usually drop off at different steps and need different fixes. Keep the step count low, because a ten step funnel produces percentages so small at the end that normal variation looks like a crisis.

Because two months is GA4's default data retention and most properties never change it. Explorations, including funnels, run on the user and event level data governed by that setting. Standard GA4 properties can be raised to 14 months, with 26, 38 and 50 month options only on Analytics 360. The change is not applied retroactively, so every day left at the default permanently discards history. Raise it to 14 months now, then link the BigQuery export if you need year on year comparison.

Less than the headline figure suggests, but the recoverable share is substantial. Baymard Institute puts average documented cart abandonment at 70.19 percent across dozens of independent studies, though much of that is browsing rather than purchase intent. Among shoppers who abandoned for reasons other than browsing, extra costs such as shipping, tax and fees were the largest cause at 48 percent, followed by forced account creation at about 25 percent and delivery being too slow at about 24 percent. All three are fixable without custom development.

Most often it means shipping cost is discovered at the cart, which matches the research showing unexpected extra costs as the leading abandonment reason. It can also mean the cart page is a dead end with no obvious next action, or that stock or delivery information appears only at that stage. Before redesigning anything, watch a handful of session recordings at that step and check the mobile layout specifically, since the checkout button often falls below the fold on smaller screens.

Three causes dominate. Forced account creation, which the research identifies as a leading reason for abandonment and which guest checkout resolves outright. Payment friction, where a preferred method is missing or a card is declined without a clear message. And form burden, where the checkout asks for information the order does not require. Instrument each checkout step as its own event so the funnel shows exactly which screen loses people, because a single begin_checkout to purchase gap tells you almost nothing actionable.

It is a funnel problem at every step. Google and Deloitte's Milliseconds Make Millions study found that improving mobile site speed by 0.1 seconds lifted retail conversion 8.4 percent and average order value 9.2 percent, and Portent found conversion drops an average of 4.42 percent per additional second of load time. GA4 will not report Core Web Vitals, so measure them in Search Console against the thresholds of 2.5 seconds for Largest Contentful Paint, 200 milliseconds for Interaction to Next Paint and 0.1 for layout shift.

Always, because it is usually where the finding is. Mobile and desktop shoppers behave differently at every stage, and mobile typically shows a wider gap between cart and purchase. A blended funnel averages the two and hides the problem. Add new against returning visitors as a second segment, since returning customers skip steps and make the aggregate look healthier than it is for first time buyers. Two segmentations turn a vague drop off number into a specific, fixable observation.

Compare the same funnel across matched periods and hold everything else steady, or better, run the change as a controlled test with a holdout group. Before and after comparisons are contaminated by seasonality, campaigns, product mix and price changes, which is why most reported funnel wins do not survive scrutiny. Whatever method you use, define the success metric before you ship. Deciding afterwards which number improved is how teams convince themselves that neutral changes worked. Give the test long enough to cover a full purchase cycle, since shoppers researching a considered purchase often convert days after the session where they first arrived.

Because GA4 sees only what the browser sent. Ad blockers, declined consent, network failures and shoppers closing the tab before the purchase event fires all cause undercounting, while duplicate purchase events from a refreshed confirmation page cause overcounting. Your platform records what was actually charged, including offline and phone orders. Send transaction_id with every purchase event so GA4 deduplicates, then treat the platform as financial truth and the funnel as a measure of relative drop off between steps.

Only partially, and it is worth being clear about the limit. GA4 measures visitors once they arrive, so it captures the consideration and conversion stages well and awareness barely at all. With more search results now ending without a click, impressions matter as much as sessions, and those live in Search Console rather than GA4. Read the two together: Search Console for whether people are seeing you, GA4 for what happens once they arrive. Neither answers the other's question.

Verify the ecommerce events fire correctly, because a funnel built on broken tracking will send you after imaginary problems. Raise data retention to 14 months and link BigQuery, since neither is retroactive. Build the funnel, segment it by device and visitor type, and find the largest single drop. Fix that one thing, measure it properly, then move to the next. Teams that change five things at once get a result they cannot attribute and learn nothing they can reuse.