Funnel Analysis

A funnel represents a user's journey as a sequence of discrete stages, each with a measurable conversion rate to the next. A common generic template, often summarized by the mnemonic AARRR ("pirate metrics," coined by Dave McClure), is Awareness → Acquisition → Activation → Retention → Revenue → Referral, though the specific stages should always be tailored to the actual product journey being studied, rather than applied mechanically regardless of fit.

Each arrow's percentage represents the conversion rate from one stage to the next — the single most important number for finding where to focus improvement effort, since it isolates each transition rather than only reporting the overall, aggregate conversion from start to finish.

Two different ways of measuring "biggest drop-off" can point to different priorities, and conflating them is a common mistake. Relative drop-off is the percentage of users lost at a given step (in the diagram above, Stage 1→2 loses 30% of users). Absolute drop-off is the raw number of users lost (Stage 1→2 loses 300 users, the largest raw number in this example, even though Stage 3→4's 55% relative loss is a larger percentage).

In this example, Stage 1→2 and Stage 3→4 tie on absolute drop-off (300 users each), but Stage 3→4's relative drop-off (55%) is much higher, suggesting a more severe, specific problem at that step relative to how many users reached it — while Stage 1→2's large absolute number may simply reflect that far more users reach that early stage in the first place. Generally, relative drop-off is more useful for diagnosing where a specific step itself is unusually broken, while absolute drop-off is more useful for prioritizing where fixing a step would recover the most total users — both numbers matter, and a PM should look at both rather than defaulting to just one.

A funnel's aggregate conversion rate can obscure dramatically different underlying realities across segments, a phenomenon related to Simpson's Paradox — where a trend visible in aggregated data reverses or disappears entirely once the data is broken into meaningful subgroups. In a funnel context: an overall signup-to-activation rate might look stable or even improving, while masking the fact that mobile users are converting far worse than desktop users (or vice versa), because a shift in the mix of traffic (more mobile users overall, who convert at a lower baseline rate) is hiding a genuine, segment-specific problem worth investigating separately.

Funnel data reliably tells a PM where users drop off, but rarely tells them why on its own. Once a specific stage transition is identified as the priority (using both absolute and relative drop-off, and checked for segment-hidden variation via Simpson's Paradox), the next step is typically qualitative: session recordings, targeted user interviews, or usability testing focused specifically on that step, to understand the actual user experience causing the drop-off. A PM who treats funnel data as sufficient on its own, without this qualitative follow-up, risks guessing at a fix based on assumption rather than genuine understanding of the underlying cause — precisely the discovery discipline from Lesson 8, applied here to a specific, quantitatively-identified problem area rather than a broad, undirected exploration.