Combining multiple entry paths hides how any one path actually converts

What it is

People can enter the same business flow from home, search, email, ads, or a deep link. Merging those entries into one funnel yields a traffic-weighted average. A high-volume, decently converting path drowns a low-volume path that barely works: an email chain can fail almost completely while the blended funnel still looks healthy. The blend answers how many people finished overall. It does not answer whether each entry path is usable. The hidden path is often exactly the experience a campaign or a segment actually had.

Why it happens

Overall conversion weights each path by how many people entered it. Those weights move with campaigns, placements, and season, so the headline can move while every path’s conversion stays put. A high-friction path with little traffic cannot pull the average enough to alarm anyone; when a campaign enlarges it, overall conversion drops and looks like a regression. Merging also packs different intents into one step: comparison shoppers and ready buyers share “entered checkout,” emptying that step of a single meaning. Cross-path returns create double counts if left raw, and erase origin if naively deduplicated.

Studying it

Enumerate the main entries into the flow, build a funnel per entry, then report the blend as capacity, not diagnosis. Make version comparisons inside a path. Inspect the entry-by-step table for “only this entry dies at this step.” When traffic is badly unbalanced, read small paths with absolute drop-off and rate together so a tiny denominator does not dominate. If entry can change mid-session, pre-commit an attribution rule—first entry, last entry, or step-wise credit—and keep it fixed.

Where it stops holding

Over-fine entry cuts produce many unstable mini-funnels that should not be ship gates. Missing or blocked entry labels dump the worst traffic into an “unknown” bucket, which then looks like a bad source when it is a measurement hole. Cross-session continuation—ad yesterday, home today—makes single-session path analysis undercount delayed completion. Blends remain useful for capacity: cashier load and peak traffic need overall throughput, not path diagnosis.

Applying it

• Compute a funnel for each major entry; put blended conversion in an appendix, not as the verdict on whether a path works.
• Before launching a campaign or placement, give that entry its own baseline; do not substitute the site-wide funnel.
• When blended conversion moves, split by entry first: a single path worsened, weights shifted, or every path moved together.
• For a path that stays at the bottom, fix it or stop sending people there; do not let other paths’ success vouch for it.

Related

• Same group: Q3.12.1 Funnels locate drop-off, not causes · Q3.12.2 Retention curve shape outweighs a single-day number · Q3.12.3 Cohort analysis keeps new and returning users apart · Q3.12.4 Inconsistent step definitions make conversion incomparable · Q3.12.6 Survivor bias overstates typical long-term experience · Q3.12.7 Rolling and classic retention are not interchangeable
• Adjacent: Q3.05 Multivariate testing · Q1.04 Sampling and representativeness
• Search terms: path-level funnel · aggregation bias · entry-point conversion