Product funnel definition

A funnel is a sequential model mapping the steps between a user entering a product context and completing a defined goal. The name comes from the shape: volume at the top narrows at each step as users drop off. In PM interviews, “funnel” refers to two distinct things that candidates often conflate. The first is the acquisition funnel (macro, from first touch to revenue). The second is the in-product micro-funnel (onboarding completion, feature activation, checkout, referral). Knowing which one you are analyzing, and why, is the first grading criterion.

The AARRR funnel and its metrics

AARRR (Acquisition, Activation, Retention, Referral, Revenue) is the most cited funnel in PM interviews. Dave McClure coined it. Also called Pirate Metrics. Each stage has a metric that belongs to it:

Acquisition: new users or signups per period; CAC; channel conversion rate (ad click to signup)
Activation: percentage of new users who reach a defined first-value event within a set window (24-72h for consumer, 7-14 days for PLG B2B); see activation
Retention: Day-7, Day-30, or cohort-level return rates; churn rate as the anti-metric
Referral: viral coefficient (invites sent per retained user times invite-to-signup rate); Net Promoter Score as a proxy
Revenue: conversion to paid; expansion MRR; LTV/CAC ratio

Top PLG companies sustain activation rates of 20-40% for freemium products and 40%+ for free-trial products. Below 20% is an optimization target, not a nuance. Interviewers expect you to know these baselines.

Funnel analysis as prioritization

The skill interviewers are testing is not whether you can name the stages. It is whether you can find the highest-leverage leak and explain what you would do about it.

Calculate drop-off rate at each step: users lost divided by users who entered that step. The step with the highest absolute drop-off is usually the fix priority, not necessarily the lowest percentage conversion, because it affects the most users.

A canonical interview scenario: a freemium SaaS onboarding funnel with 1,000 signups. 600 complete profile setup. 180 create their first workflow. 40 share a workflow. The leaky step is workflow creation (70% drop-off on a large base), not sharing (78% drop-off on a base of 180). You investigate workflow creation first: is the builder confusing, are users hitting a load error, are they landing on the wrong template? You validate with session replay, support ticket tagging, or a quick user call. Then you test a fix. The interviewer expects this sequence: locate the leak, hypothesize root cause, validate, test.

Conversion rate is always computed as: users who completed the goal divided by users who entered at the top. Always specify the window (session-level, 7-day, 30-day) or the number is meaningless. Interviewers will press on this.

The adjacent terms and where they live on the funnel

Activation is the funnel step where a user first hits the aha moment, between Acquisition and Retention in AARRR. Conversion is completing a monetization or goal step, usually at the Revenue stage. Churn is exit from the Retention stage and functions as a re-entry signal: churned users either re-engage (re-enter Retention) or must be re-acquired (re-enter the top of the funnel). These are not separate concepts. They are positions on the same user journey.

One structural distinction most candidates miss: the acquisition funnel is linear (users pass through stages once). Product engagement is a loop. Retention feeds back into Activation for new features, and churn reopens the Acquisition stage. Treating the funnel as strictly one-directional produces wrong diagnoses.

When funnel analysis is the wrong model

Not all user behavior is linear. If users access features in non-sequential order (a content discovery product, a marketplace with variable entry points), a funnel model will mismatch the actual behavior and produce misleading drop-off numbers. Use a funnel when there is a genuine sequential dependency between steps. Otherwise, use cohort analysis or event-flow visualization.

For AI/agentic products, the human AARRR funnel breaks down at Activation. An AI agent is configured by an operator, not self-onboarded in the way a consumer user is. The relevant funnel becomes: Configured (operator sets up the agent) to Executed (agent receives a task) to Completed without escalation (containment rate) to Repeated (operator volume trends, week-over-week). Task completion rate and containment rate replace activation rate. Prompt volume decline week-over-week is an early signal of eroding operator confidence, the agentic equivalent of churn. If you blend human user events and agent traffic into one funnel, the bottom-of-funnel numbers inflate artificially and you will optimize for the wrong thing.

The 2026 viability and lovability lens

In 2026, AI collapses build time, so feasibility is close to free. Funnel leaks now expose viability failures or lovability failures almost exclusively. A high Acquisition drop-off is usually a viability problem: the product is not solving something the right users will pay for, or the channel is mismatched to the audience. An Activation or Retention drop-off is almost always a lovability failure: the product is not meeting users where they work, its proactive help is absent or obnoxious, or the core value is not landing in the user’s actual workflow.

A PM who diagnoses a 60% Activation drop-off and proposes adding a tooltip is solving the wrong problem. The right question is whether this is a use case misalignment (viability) or a delivery failure (lovability). That distinction determines whether you iterate on the onboarding flow or revisit the product’s core premise.

Weak vs strong interview answer

weak

"A funnel is just AARRR." This conflates a specific framework with the broader concept. AARRR is one instance of a funnel. Defining funnel as Awareness to Interest to Desire to Action (AIDA) is a marketing model, not a product-analytics model. Listing stage names without specifying the metric that belongs to each stage signals surface knowledge. Proposing to fix all leaky steps simultaneously misses the core skill: identify the highest-leverage step first. Stating a conversion rate without a defined time window will get pressed on immediately.

strong

"A funnel maps the sequential steps between a user entering a product context and completing a defined goal. I distinguish two types: the macro acquisition funnel, where I track CAC, activation rate, and LTV/CAC across AARRR, and the in-product micro-funnel, like onboarding or checkout, where I calculate step-level drop-off rates to find the leaky step. The leaky step is the one with the highest absolute user loss, not just the lowest percentage conversion. I would hypothesize root causes, validate with session replay or user research, and test a fix. For a 2026 AI product, I would split the funnel: human users follow a standard AARRR path, while agent traffic follows Configured to Executed to Completed to Repeated, with containment rate as the key activation equivalent. Mixing those two funnels produces inflated metrics and wrong priorities."

For the full AARRR model with stage-by-stage worked examples, see AARRR pirate metrics. For how funnel leaks connect to the metrics interview question type, see activation and churn.