Conversion Funnel Optimisation: Finding and Fixing the Leaks

Conversion funnel optimisation is the narrow, high-leverage part of CRO: instead of improving a single page in isolation, you instrument the whole sequence from landing to conversion, find the step where the most users (and the most revenue) fall out, work out why, and fix that step first. Done properly it is a measurement exercise before it is a design exercise, and the measurement part is where most teams go wrong: they look at percentage drop-off instead of absolute users lost, they trust funnel numbers distorted by consent banners, and they redesign pages without ever watching a single abandoner’s session. This guide covers the full loop: defining the funnel as real event steps rather than a marketing metaphor, building the report in GA4, diagnosing the why, prioritising fixes by revenue, and what the 2026 tool landscape actually looks like now that Google Optimize is gone and Hotjar has been folded into Contentsquare.

A funnel is event steps, not AIDA

Most ranking guides frame the funnel as awareness, interest, desire, action. That is a useful way to think about marketing messages and a useless way to find leaks, because none of those stages corresponds to anything you can measure. For optimisation purposes, your funnel is a sequence of instrumented events on your own site. For an ecommerce store that is typically:

1. Landing page or product listing view
2. Product page view
3. Add to basket
4. Begin checkout
5. Add payment details
6. Purchase

For a SaaS product it might be pricing view, signup start, signup complete, first key action. The difference matters: “interest is dropping” is not fixable; “62 percent of mobile users who begin checkout never reach the payment step” is. If you have not yet defined what conversion means for your site or set a baseline, start with our conversion rate optimisation guide and come back.

A sales funnel and a conversion funnel are the same idea at different altitudes: the sales funnel is the whole journey from first touch to revenue, often spanning channels; the conversion funnel is the on-site, instrumented subset of it that you can actually measure step by step.

Step 1: Build the funnel report in GA4

GA4 has no funnel in its standard reports, which surprises almost everyone migrating from Universal Analytics. Funnels live in the Explore section as the Funnel exploration technique. To build one: Explore, new Funnel exploration, then define each step as an event (page_view with a path condition, add_to_cart, begin_checkout, purchase). Google’s official documentation for Funnel exploration covers the settings in detail; the ones that change your numbers are:

• Open vs closed funnels. In a closed funnel, users must enter at step one, and anyone who skips a defined step falls out of all subsequent steps. An open funnel lets users enter at any step. Closed is the right default for diagnosing a linear checkout; open is better for content sites where entry points vary.
• Abandonment rate per step. GA4 shows the percentage abandoning at each transition. This is your leak map.
• Elapsed time. Turning on elapsed time between steps surfaces hesitation: a step where the median time balloons is often a step where users are confused, comparison shopping, or hunting for a discount code.
• Breakdowns. Add device category as a breakdown dimension immediately. Mobile cart abandonment runs around 80 percent against roughly 66 percent on desktop in current Baymard-derived figures, so a blended number routinely hides a mobile-specific disaster.

One UK-specific caveat that no major guide mentions: your GA4 funnel only contains consented users. Under UK GDPR and PECR, analytics cookies need opt-in, so every step count excludes the visitors who declined the banner, and decliners are not a random sample (they skew more privacy-conscious and often more mobile). GA4’s consent mode can model some of the gap, but treat absolute funnel counts as directional and step-to-step ratios as the reliable signal.

Step 2: Find the leak that matters (not the biggest percentage)

In most mature funnels, one or two steps account for the majority of total drop-off. But “biggest percentage drop” is the wrong sort order. A 90 percent drop on a step that 200 users reach matters less than a 40 percent drop on a step that 20,000 users reach. Weight each leak by absolute users lost multiplied by the revenue those users represent, then by how hard the fix is. A worked example for a store with 50,000 monthly sessions, estimating recoverable orders as users lost multiplied by how many would plausibly complete after a fix:

The product-page drop is the biggest percentage by far, but much of it is window shopping you will never convert. The two checkout leaks are smaller in raw users and far larger in recoverable revenue, because users that deep in the funnel have declared intent. This is also the answer to “top or bottom of funnel first”: fix the bottom first, both because intent is highest there and because every top-of-funnel improvement you make later flows through the repaired steps. For scoring frameworks (PIE, ICE, PXL) and a full worked prioritisation, see how to build a CRO programme.

Step 3: Diagnose why, with a real process

The funnel tells you where; it never tells you why. Three evidence sources, in order of effort:

Known abandonment causes first. Baymard Institute’s long-running research puts average cart abandonment at 70.22 percent across 50 studies, and its cart abandonment statistics list the reasons shoppers actually give. Excluding “just browsing”: extra costs too high (39 percent), delivery too slow (21 percent), card security concerns (19 percent), forced account creation (19 percent), checkout too long or complicated (18 percent), unsatisfactory returns policy (15 percent), and site errors or crashes (15 percent). Baymard estimates that a large share of lost orders across the US and EU is recoverable through better checkout design alone. Check your leaky step against this list before inventing exotic theories: hidden delivery costs and mandatory account creation are boring, common and cheap to fix.

Watch abandoner sessions, properly. “Watch session recordings” is hand-waved everywhere; here is a process. Filter recordings to users who reached the leaky step and did not reach the next one, on the worst-performing device. Watch 10 to 15 of those sessions, no more to start. Code what you see into patterns versus one-offs: rage clicks on a dead element, the same form field abandoned repeatedly, a coupon field triggering an exit to Google, an error message flashing. Two or three sessions showing the same behaviour is a hypothesis; one session showing something weird is an anecdote.

Ask the people leaving. A one-question exit survey on the leaky page (“what stopped you completing this today?”) converts guesses into quotes. Run it for a fortnight and you will usually have your answer in the first 30 responses.

Check page speed on the leaky step too; it is the diagnosis major guides skip entirely. Google’s Core Web Vitals thresholds are LCP within 2.5 seconds, INP within 200 milliseconds and CLS within 0.1 (INP replaced First Input Delay as a stable Core Web Vital in 2024). A checkout step that fails INP on mid-range Android phones behaves exactly like a design problem in your funnel data, and is often the real reason mobile converts so much worse than desktop.

Step 4: Fix, test, and know when testing is not viable

With a diagnosed leak, ship the fix as an A/B test where traffic allows. Google Optimize was sunset on 30 September 2023 with no Google-built replacement, so testing now means a third-party tool integrated with GA4; our best A/B testing tools comparison covers the current options, and note that VWO’s old free 50,000-visitor plan is being withdrawn, so the genuinely free testing route in 2026 is PostHog’s experiments on its free tier.

Whether you can test at all depends on volume at the step you are fixing, not site-wide traffic. As a rough rule, if the step converts fewer than about 500 times a month, a classic A/B test will take months to reach significance on a realistic uplift; run the numbers in our sample size calculator before committing. If you are below the line, do not fake it with underpowered tests. Ship the fix directly when the evidence is strong and obvious (Baymard-list problems like surprise costs or forced accounts rarely need a test to justify), measure before and after on the step conversion rate, and accept the weaker evidence honestly.

Then re-run the funnel report, find the new worst leak, and go again. Funnel optimisation is a loop, not a project.

The free diagnostic stack in 2026

Tool advice in ranking guides is widely out of date, so here is the current position:

The claim that Clarity cannot do funnels is simply stale: Microsoft’s own Clarity funnels documentation describes the feature, and because each funnel step links straight to the matching recordings, it collapses the where and the why into one screen for free.

What a good funnel converts at

Treat any single benchmark with suspicion. IRP Commerce’s live ecommerce market index put the average conversion rate at 1.70 percent in April 2026 (down from 1.81 percent a year earlier), while Shopify’s vertical figures run from 1.41 percent (home and furniture) to 6.22 percent (food and beverage), with fashion at 3.06 percent. The spread between sources and verticals is wider than most “good vs bad” judgements, so benchmark against your own funnel’s history and your specific vertical; our breakdown of what counts as a good conversion rate goes deeper on the sector numbers.

Frequently asked questions

How is conversion funnel optimisation different from CRO? CRO is the whole discipline of increasing the share of visitors who convert. Funnel optimisation is the subset that maps the full multi-step journey, finds the step losing the most revenue, and fixes leaks in priority order rather than improving pages in isolation.

Why is there no funnel report in standard GA4 reports? GA4 moved funnels into the Explore section as the Funnel exploration technique. You build one manually by defining each step as an event; standard reports do not include it.

Why do customers abandon at checkout? Baymard’s research across 50 studies puts average abandonment at 70.22 percent. The top stated reasons after browsing are extra costs (39 percent), slow delivery (21 percent), card security worries (19 percent), forced account creation (19 percent) and overlong checkouts (18 percent).

Should I fix the top or the bottom of the funnel first? Bottom first. Users deep in the funnel have declared intent, so recovered users there are worth the most, and any later top-of-funnel gains then flow through the repaired steps instead of leaking out.

How many session recordings should I watch to diagnose a drop-off? Filter to users who abandoned at the leaky step on the worst device, watch 10 to 15 sessions, and separate repeated patterns from one-offs. Two or three sessions showing the same struggle is a testable hypothesis.

Why is my mobile conversion rate so much worse than desktop? It usually is for everyone: mobile cart abandonment runs around 80 percent versus roughly 66 percent on desktop. Check Core Web Vitals on mobile (LCP 2.5s, INP 200ms, CLS 0.1), form length, and payment options before assuming a design problem.