Funnel Segmentation
Funnel segmentation breaks aggregate conversion into device, source, geo, and visitor-type cohorts — surfacing the steps where lift is actually available.
Funnel Segmentation
Splitting funnel conversion by traits like device, traffic source, geography, or visitor type to expose where drop-off is concentrated.
Funnel segmentation is the practice of recomputing each funnel step's conversion rate across cohorts — paid vs organic, mobile vs desktop, returning vs new, country, campaign, landing page — instead of relying on a single blended number. The blended rate is an average of averages, and averages routinely hide the segment that is actually broken.
It sits underneath Funnel Analytics as a diagnostic technique and feeds Funnel Optimization as a prioritisation tool: you cannot fix what the aggregate has averaged out, and you should not test against cohorts where the leak does not exist.
The classic failure mode looks like this. Your store converts at 2.4% overall and that number has been flat for two quarters. You assume the funnel is healthy enough. Segment by device and you find desktop converts at 4.1% while iOS Safari sits at 1.3% — and iOS Safari is 58% of your sessions.
Now the question changes from "how do we lift conversion?" to "what is broken on iOS Safari checkout?" That is the entire point of segmentation: it converts a vague optimisation goal into a specific, testable hypothesis with a clear addressable audience.
segment_lift = (CR_segment - CR_blended) / CR_blended
CR_segment
Segment conversion rate
Conversion rate for the isolated cohort (e.g. mobile, paid social, Germany).
CR_blended
Blended conversion rate
Overall funnel conversion rate across all traffic.
segment_lift
Segment lift
Relative gap between the segment and the blended baseline. Negative = underperforming cohort.
A Shopify apparel store measures checkout-completion rate across devices for the last 30 days.
Blended checkout CR: 2.4%
iOS Safari mobile CR: 1.3%
iOS Safari share of sessions: 58%
→ segment_lift = (1.3 - 2.4) / 2.4 = -45.8%
iOS Safari converts 46% below the blended baseline and represents the majority of traffic. Fixing this single segment to parity would lift store-wide CR by roughly 30%, dwarfing any sitewide copy test.
Pick segments that are both actionable and large enough to test. Device, traffic source, country, and new-vs-returning are the workhorses. Campaign and landing page matter when paid is your main acquisition lever. Avoid slicing so fine that no cohort has the volume to reach significance — sub-1% sessions per cohort is usually a dead end.
Typical checkout-completion rate by device and source — Shopify apparel and beauty stores, €2M-€10M GMV
| Segment | Add-to-cart → checkout | Checkout → purchase | Session → purchase |
|---|---|---|---|
| Desktop, organic search | 58% | 72% | 3.8% |
| Desktop, paid search | 54% | 70% | 3.2% |
| Mobile, organic search | 46% | 61% | 2.1% |
| Mobile, paid social | 38% | 54% | 1.4% |
| Mobile, email (returning) | 62% | 78% | 4.6% |
| Tablet, all sources | 48% | 65% | 2.3% |
Two traps to avoid. First, Simpson's paradox: a variant can win every segment but lose in aggregate (or vice versa) if traffic mix shifts mid-test — always check segment-level lifts before declaring a result. Second, post-hoc slicing: if you cut the data twenty ways after the experiment ends, one segment will look significant by chance. Pre-register the two or three segments you actually care about.
Frequently asked questions
Funnel analytics measures step-by-step drop-off across the whole audience. Funnel segmentation is one layer deeper: it recomputes that drop-off separately for each cohort so you can see which audience is responsible for the aggregate number.
Device first (mobile vs desktop), then traffic source (organic, paid, email, direct), then new vs returning. Those three cuts surface 80% of the meaningful variation on a typical Shopify or WooCommerce store. Geography and campaign come next.
For directional reads, 500-1,000 sessions per segment per week. For A/B testing against a segment, you need enough conversions to hit your minimum detectable effect — usually 200+ conversions per variant in that cohort.
Yes, GA4's Funnel Exploration supports breakdowns by dimension, but you are limited to one breakdown at a time and sampling kicks in on high-volume properties. Most teams export to BigQuery or a dedicated analytics layer for multi-dimensional segmentation.
Source is the channel bucket — google / organic, facebook / cpc, klaviyo / email. Campaign is one level down: a specific Meta ad set or a specific abandoned-cart flow. Segment by source to find broken channels, by campaign to find broken creative.
Segmentation tells you where conversion fails; attribution tells you which touchpoint deserves credit. A paid-social cohort can convert poorly on the landing page (a segmentation finding) while still being a profitable assist (an attribution finding). Use them together, not interchangeably.
Yes — it is one of the highest-signal cuts on Shopify and Magento. Logged-in shoppers typically convert 2-3x guest rates, and the gap quantifies the value of pushing account creation or one-click checkout (Shop Pay, Apple Pay) earlier in the flow.
Yes, if you slice the data post-hoc. Each additional segment cut multiplies the false-positive risk. Either pre-register the segments you will analyse, apply a Bonferroni correction, or only segment for diagnostic reads outside the primary test result.
Usually no at the country level, but city or region can still matter — shipping cost expectations, language variants, or device mix often vary regionally. If you ship internationally via Shopify Markets, country segmentation is essential because currency and shipping economics change conversion.
A full cut once a quarter, and after any major traffic-mix change — new channel, new market, big paid push. Between those, monitor a small dashboard of segment-level conversion rates so you spot regressions in a specific cohort before they drag the blended number down.
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.