How to use Funnel Analysis
A practical guide to funnel analysis for online stores: how to define stages, calculate step conversion, segment the data, and turn drop-off into prioritised CRO experiments.
Funnel Analysis
Funnel analysis breaks a purchase path into ordered stages and measures conversion at each transition to locate drop-off.
Funnel analysis is the diagnostic method that turns a vague "our conversion rate is low" into a precise "we lose 62% of sessions between product page and add-to-cart on mobile Safari." You define the ordered stages a visitor must complete, count how many enter and exit each one, and compute step-by-step conversion rates.
It sits upstream of every optimisation decision. Before you run an A/B test, redesign a checkout, or rewrite product copy, funnel analysis tells you which stage is actually costing revenue — so test effort lands where it moves the number, not where the loudest opinion in the room points.
Most online stores report a single conversion rate — sessions divided by orders — and stop there. That number tells you the patient has a fever; it does not tell you which organ is failing.
Funnel analysis is the temperature-by-organ readout. It is the diagnostic that lets you walk into a Monday standup and say which stage to fix this sprint, with the drop-off percentage to back it up. Everything in funnel optimization — the experiments, the redesigns, the personalisation — depends on getting this measurement right first.
Defining the stages of your funnel
A stage is any required transition between the visitor landing and the order confirming. On a typical Shopify store the default sequence is: landing → product view → add to cart → begin checkout → shipping → payment → order placed.
Resist the urge to over-fragment. Seven stages is usually the practical maximum — beyond that, each step has too few sessions to draw a reliable conclusion, and you spend more time arguing about definitions than fixing problems. For micro-stages inside checkout (form-field abandonment, shipping selection), drop into session replay rather than expanding the funnel itself.
The stages must be strictly ordered and mutually exclusive. A session that reaches checkout has, by definition, also viewed a product and added to cart — funnel math breaks if a visitor can skip a stage or be counted in two at once. If your store has multiple entry points (PDP from paid social, collection from organic, cart from a back-in-stock email), keep the stage definitions identical and segment by entry point instead.
Closed vs. open funnels
A closed funnel only counts visitors who entered at stage one. An open funnel counts anyone who reached a stage, regardless of where they joined. For ad-driven traffic that often lands deep (a Meta ad to a PDP, a Google Shopping click to a specific SKU), open funnels are usually more honest — closed funnels undercount checkout performance because they exclude legitimate buyers who skipped the homepage.
Calculating step conversion and drop-off
For each transition, step conversion = sessions reaching stage N+1 ÷ sessions reaching stage N. Drop-off is the complement: 1 minus step conversion. Always report both — step conversion makes you feel good ("68% click add-to-cart!"), drop-off makes you act ("we lose 1 in 3 here").
Overall conversion rate is the product of every step's conversion. That is why a 5-point lift on the weakest stage usually beats a 5-point lift on the strongest: multiplicatively, fixing the worst leak compounds through everything downstream. A site that goes from 40% to 60% step conversion at add-to-cart pulls every later stage's contribution up with it.
Typical Shopify apparel funnel: sessions remaining at each stage (per 10,000 entries)
In the example above, the biggest absolute leak is landing → product view (4,200 lost sessions), but the biggest percentage drop is product view → add to cart at 75%. Which one to attack depends on cost-of-fix: messaging and PDP layout changes are cheap experiments; rebuilding category navigation is not.
Segmenting to find the real problem
An aggregate funnel hides more than it reveals. The same store often has a 3.8% mobile conversion rate and a 6.2% desktop rate; a 5.1% returning-customer rate and a 1.4% paid-social rate. Reporting one blended number averages those differences into uselessness.
The four segments worth cutting on day one: device (mobile / desktop / tablet), traffic source (paid social, paid search, organic, email, direct), new vs returning, and landing page type (PDP / collection / homepage). Pull the funnel for each segment side by side — the leaks are almost never uniform, and the segment with the worst drop-off at a given stage is usually where the cheapest win lives.
Step conversion rates by device and traffic source — apparel & beauty stores, AOV €40-€120
| Segment | Landing → PDP | PDP → ATC | ATC → Checkout | Checkout → Order |
|---|---|---|---|---|
| Desktop, organic | 62% | 28% | 68% | 52% |
| Desktop, paid search | 58% | 31% | 71% | 55% |
| Mobile, organic | 55% | 22% | 61% | 41% |
| Mobile, paid social | 48% | 16% | 54% | 33% |
| Returning customers | 71% | 38% | 78% | 64% |
| Email campaigns | 66% | 34% | 74% | 58% |
The mobile / paid-social row is where most apparel stores hemorrhage revenue: cold visitors on small screens with low intent. If your funnel looks healthy in aggregate but mobile paid-social is 30-40% below the desktop baseline, that single segment is your highest-leverage experiment queue — landing-page speed, social-proof above the fold, simplified PDP.
Common mistakes that invalidate the analysis
The most frequent error is comparing this week's funnel to last week's without controlling for traffic mix. A 0.4-point conversion drop after a big paid-social push is not a UX regression — it is a composition shift. Always pin your comparison to the same segment slice, not the blended total.
The second is reading a single week of data on a low-volume stage. If only 280 sessions reached the payment step, the confidence interval on that step's conversion rate is roughly ±5 points. Wait for sample, or roll up to 28-day windows. The third is forgetting that a fresh GA4 property has no historical baseline — if you're starting cold, import the prior 12 months from GA4 first so day-one funnel analysis includes seasonality, not just this week's noise.
Drop-off is a symptom, not a cause
Funnel analysis tells you WHERE visitors leave, never WHY. Pair every identified leak with a qualitative layer — session replay, on-exit survey, or heatmap on that specific page — before you write the test brief. Optimising blind, based purely on funnel numbers, is how teams ship five rounds of CTA-colour tests on a page where the real problem is a broken shipping calculator.
Funnel analysis FAQ
Funnel analysis is the measurement step — defining stages, computing drop-off, segmenting cohorts. Funnel optimization is what you do with that information: prioritising experiments, redesigning pages, running A/B tests. Analysis tells you where to fix; optimization is the fixing.
Four to seven for an e-commerce store. Fewer than four and you can't distinguish PDP problems from checkout problems; more than seven and individual stages get too thin for reliable analysis. Use session replay for micro-step diagnostics inside a single page.
Sessions for short-cycle funnels (visit-to-purchase on a sub-€100 SKU), users for considered purchases that span multiple visits (electronics, furniture). GA4 defaults to users; most CRO platforms default to sessions. Pick one and document it — mixing the two across reports is the fastest way to lose stakeholder trust.
In a closed funnel they're excluded; in an open funnel they're counted at every stage they reach. For paid traffic landing deep in the site, open funnels are almost always more accurate — closed funnels make checkout performance look artificially weak.
Roughly 1,000 sessions per stage per segment per week as a working minimum. Below that, your step conversion confidence intervals are wide enough to drown the signal. Lower-traffic stores should widen the time window to 28 days rather than try to slice weekly data thin.
Yes, but with two funnels. Run an acquisition funnel (first-order, identical to single-purchase stores) and a retention funnel (order N → order N+1, with stage definitions like "replenishment email opened → cart restored → reorder placed"). Blending them hides the fact that returning-customer economics are usually 3-4x stronger.
GA4's Funnel Exploration handles basic linear funnels, but it struggles with open funnels, deep segmentation, and historical comparisons longer than 14 months. Most CRO teams use GA4 for the data layer and a dedicated tool for the analysis layer.
A full segmented review monthly, plus a top-line check weekly. After every major site change (replatform, theme update, checkout edit) rerun within 7 days — regressions hide easily in aggregate numbers and only show up when you compare the affected segment pre- and post-change.
Rules of thumb for apparel / beauty: landing → PDP 50-65%, PDP → ATC 8-15%, ATC → checkout 60-75%, checkout → order 45-60%. Higher-AOV verticals (electronics, furniture) run lower at every stage. Use these as a sniff test, not a target — your real benchmark is your own funnel trended over time.
Once the funnel and segments are computed, AI is useful for pattern recognition across hundreds of segment-stage combinations and for proposing testable hypotheses tied to specific drop-off points. It does not replace judgement about which experiment to run — but it does eliminate the hours spent manually scanning pivot tables looking for the worst leak.
Get an AI expert review of your site
Paste your URL — Metricuno's AI runs the same heuristic checks a senior CRO consultant would, scoring your page and prioritising the fixes that'll move conversion fastest.