Conversion Funnel Explained: Assumptions, Drop-Offs, and Nonlinear Journeys

A conversion funnel is a measurement model that orders selected events or states from an entry population to an outcome and reports how many eligible units continue or drop between steps. It is not a literal map of every journey. The result depends on identity, eligibility, step definitions, ordering, time window, open-versus-closed entry, and treatment of repeated or nonlinear behavior.

A funnel is a contract over a sequence#

Google Analytics’ funnel exploration and Amplitude’s funnel documentation both require configured steps and ordering behavior. The chart is therefore an answer to a designed question, not a neutral picture of all customer behavior.

The core calculations are:

step conversion rate = unique eligible units completing step B within the window
                       / unique eligible units at step A × 100

overall conversion rate = unique eligible units completing the final step
                          / unique eligible units entering the funnel × 100

The arithmetic is exact after the records are chosen. The hard work is choosing records correctly. “Unique unit” might mean a browser, user, account, opportunity, or order. “Within the window” could start at first exposure, first step, or a reporting boundary. Two tools can show different correct values because their contracts differ.

A worked example shows why denominators matter#

Illustrative only—not a benchmark. Suppose 800 eligible accounts enter a funnel, 320 start setup, and 160 reach the defined activation state within 14 days.

• Entry-to-start conversion is 320 / 800 = 40%.
• Start-to-activation conversion is 160 / 320 = 50%.
• Overall conversion is 160 / 800 = 20%.

If the first report counted sessions and the second counted accounts, the comparison would be invalid. If 40 accounts activated after 15 days, they are non-converters under this contract but may be successful under a longer one. If account identity was attached only after signup, earlier anonymous activity may be missing or joined imperfectly.

The example also shows why “the biggest drop” is not automatically the best priority. A step can have high numerical loss because it is intentionally broad. Another step can have lower loss but block the highest-value eligible segment. Prioritization needs consequence, reach, evidence quality, and feasibility, not only a percentage.

Open and closed funnels answer different questions#

Google Analytics defines a closed funnel as one in which users must enter at the first step. An open funnel permits entry at any step. A closed funnel is useful for analyzing a specified path from a common exposure. An open funnel can include people whose earlier behavior occurred elsewhere or was not observed.

Neither is inherently more accurate. The choice depends on the decision:

Ordering matters too. Strict order excludes a unit that completes steps in another sequence. An indirect sequence can allow intervening events. Exact-order rules answer a narrower process question. Document the choice rather than relying on a chart label.

Drop-off is a measurement outcome, not a diagnosis#

Funnel drop-off means an eligible unit did not satisfy the next defined condition under the configured rules. It does not prove abandonment, confusion, objection, or loss.

A unit can drop because it:

• took another valid path;
• completed the step on another device or identity;
• returned after the conversion window;
• was intentionally disqualified;
• encountered a technical or tracking failure;
• needed another stakeholder to act; or
• never had the underlying need.

Investigate with event-quality checks, path analysis, segmented comparisons, interviews, usability research, support records, and operational outcomes. Each method has limits. A session recording can show behavior but not the person’s complete reasoning; an interview can explain context but may not represent the population.

Real journeys can loop around the model#

McKinsey’s consumer decision journey proposed a circular model in which evaluation and loyalty do not follow one simple narrowing sequence. That older consumer study should not be imported as proof of B2B SaaS behavior. It does establish a useful caution: linear funnels are simplifications.

B2B journeys can involve several people completing different tasks. Product journeys can loop through learning, value, failure, and reactivation. Customers can enter through support, community, integrations, referrals, or direct product use that the selected web funnel does not capture.

Keep the funnel when a bounded transition matters. Add path, cohort, state-transition, journey-research, or pipeline views when the decision involves loops, multiple actors, or long intervals. More diagrams do not automatically produce more truth; each needs its own contract.

Publish the assumptions beside the chart#

Every production funnel should state:

• decision and owner;
• eligible population and exclusions;
• counting unit and identity stitching;
• event or state definition for every step;
• open or closed entry and ordering rule;
• conversion window and timezone;
• handling of repeats, skips, late events, and deletions;
• data-quality tests and known blind spots; and
• action available for supported, weakened, or inconclusive evidence.

Version the contract when a step changes. Recomputing history under a new definition can be useful, but it should not be presented as a continuous series without disclosure. Preserve the old contract long enough to explain the break.