Marketing Funnel Explained: Stages, Assumptions, and When the Model Breaks

A marketing funnel is a simplified model that groups potential and existing customers by their relationship to a purchase—usually awareness, consideration, conversion, and loyalty—so a team can choose appropriate messages and measure movement or drop-off. It is useful as an operating map, but it breaks as a literal journey because people enter late, skip steps, loop backward, and influence one another.

The funnel earns its name from an aggregate pattern: many people may know about a category or brand, fewer seriously consider it, and fewer still buy. The model turns that narrowing population into a set of states that a team can plan around. Amazon Ads’ overview of the marketing funnel describes the same broad purpose while acknowledging that the real path is less linear than the diagram.

That modest definition matters. A funnel can tell you how many defined entities occupy or pass between defined states. It cannot, by itself, tell you the route each person took, why someone moved, or whether marketing caused the movement.

Four neighboring terms are easy to blur. A marketing funnel classifies the relationship with brand, campaign, and product signals; a sales funnel classifies the overlapping process through leads, opportunities, and direct sales interactions. A customer journey records the actual sequence of experiences, including loops and skipped steps. AIDA—attention, interest, desire, action—describes a staged response to selling communication, while TOFU/MOFU/BOFU names broad regions of a funnel. Shared labels do not make these models interchangeable.

The common stages are a vocabulary, not a law#

There is no official marketing funnel. Amazon Ads notes that three-, four-, five-, and longer versions coexist. The familiar TOFU, MOFU, and BOFU shorthand compresses the model into top, middle, and bottom; Shopify maps those labels to awareness, consideration, and conversion. A four-stage version adds loyalty so that the model does not stop at the first transaction.

Funnel math is simple; its counting contract is not#

The marketing funnel itself has no defining formula. Its basic movement measures do:

Stage conversion rate = next-stage entrants ÷ eligible current-stage entrants × 100

Stage drop-off rate = (eligible current-stage entrants − next-stage entrants) ÷ eligible current-stage entrants × 100

For a closed funnel in which everyone must begin at the first step:

Overall conversion rate = final-stage completions ÷ first-stage entrants × 100

Here is an illustrative example, not real company data. A team defines four observable events within one fixed window: solution-page view, pricing-page view, trial start, and activation.

The overall closed-funnel conversion rate is 20 ÷ 1,000 = 2%. That arithmetic is correct only under the stated contract. If people can enter at pricing, skip the pricing page, activate after the window, or appear as different users across devices, the same event stream produces a different answer.

Google Analytics makes this dependence visible in its funnel-exploration rules. An open funnel admits a user at any qualifying step; a closed funnel requires the first step. Its counts also depend on step order, whether intervening actions are allowed, and any time constraint. The custom-funnel documentation defines continuation and abandonment from adjacent user counts.

There is therefore no broadly accepted conversion benchmark for an entire marketing funnel. A benchmark would need the same population, stage definitions, channel mix, unit, window, and counting rules. Search Engine Land’s conversion benchmark guidance calls an all-market average limited, recommends filtering by country, industry, and channel, and treats an organization’s own past performance as the better baseline. A stable internal trend is more useful than a borrowed number with a different denominator.

Marketing funnel, sales funnel, customer journey, and AIDA#

These terms overlap, but they answer different questions.

LinkedIn’s marketing-versus-sales comparison describes the distinction as perspective more than substance: marketing emphasizes interactions with content, ads, and brand experiences, while sales emphasizes interactions with people. The handoff is an organizational choice, not a natural law. Some teams end marketing at lead capture; others keep marketing accountable through revenue and retention.

AIDA—attention, interest, desire, action—is a related hierarchy, not a synonym for TOFU/MOFU/BOFU. The Open University’s overview of communication models notes that AIDA began as a staged selling model and was later adopted to explain advertising. TOFU/MOFU/BOFU names broad funnel regions. Neither vocabulary supplies the event definitions required for measurement.

The customer journey is different in kind. The funnel says which state an entity occupies under your model. The journey records the sequence of experiences that may have contributed to that state. One is a classification; the other is a path.

Six assumptions are hiding inside every funnel#

The model becomes trustworthy only when its assumptions are tolerable for the decision at hand.

These assumptions are not defects unique to funnels. Every model simplifies. The error is forgetting which simplifications produced the number and then treating the output as direct observation of buyer intent.

The model breaks as a literal path#

The classic picture suggests a person starts with awareness and then progresses downward while alternatives and population counts steadily shrink. Real consideration can expand as well as contract.

McKinsey’s consumer decision journey research examined almost 20,000 purchase decisions across five industries and three continents. It reported that people may add brands during active evaluation rather than only narrow an initial set, and that post-purchase experience feeds future loyalty and consideration.

Google’s messy-middle research offers another bounded model. Between a trigger and purchase, consumers can loop between exploration, which expands a consideration set, and evaluation, which narrows it. Exposure remains an always-on backdrop rather than one completed stage, and product experience feeds back into later exposure.

This does not make the funnel useless. It establishes a boundary: stage totals can summarize a population even when the people inside that population followed different routes. A map of airport arrivals can be accurate without describing every passenger’s trip to the airport.

The model breaks faster in group buying#

A person-level funnel is especially weak when an account buys collectively. One participant may identify the problem, another may research suppliers, finance may define constraints, security may reject an option, and an executive may approve the final commitment. Their work overlaps; the same account can be advanced in one respect and blocked in another.

Gartner’s B2B buying-journey guidance describes a nonlinear set of often-revisited buying jobs: problem identification, solution exploration, requirements building, and supplier selection. It also says diverse stakeholders can work on tasks concurrently and without a consistent order.

The practical consequence is not to add more person-level stages. Use the account as the unit and track whether the buying group has enough shared evidence to complete each job. A contact downloading a comparison guide can be useful activity without meaning that the account has entered “decision.”

A funnel locates change; it does not explain or attribute it#

Suppose pricing-page-to-trial conversion falls. The funnel has located a change between two definitions. Several explanations remain possible: the incoming audience changed, the offer became less relevant, the page introduced friction, a tracking event failed, the comparison window is immature, or the same people completed the action somewhere unobserved.

The funnel cannot choose among those explanations. It also cannot award causal credit to the last campaign touched before the transition. Google’s messy-middle report explicitly recommends robust controlled experiments when the question is the causal effect of behavioral interventions, because revenue and sales are comparatively blunt outcomes.

This is where local optimization becomes dangerous. A team can increase lead capture by weakening qualification and celebrate a better mid-funnel rate while producing fewer viable opportunities. It can force every user through an intermediate page and make the measured sequence look cleaner while making the experience worse. A stage metric is useful only in relation to the final customer and business outcome it is meant to support.

Pair the funnel with the model that answers the missing question#

No single replacement needs to “kill” the funnel. Use a small model stack.

The models can disagree without one being wrong. An account may count as consideration in the aggregate funnel, contain several paths in event data, remain incomplete on one buying job, and belong to an immature cohort. Those are different projections of the same commercial reality.

Write a one-page funnel contract before building the chart#

The most useful takeaway is a short specification that travels with every funnel. If a chart cannot answer these questions, it is not ready to drive a decision.

1. Decision: What action will change when this funnel changes?
2. Entity: Are you counting people, accounts, opportunities, sessions, or orders?
3. Population: Who is eligible to enter, and which exclusions apply?
4. States: What observable condition begins and ends each stage?
5. Movement: Is the funnel open or closed? May entities skip, repeat, move backward, or re-enter?
6. Time: How long may a transition take, and how will immature entrants be handled?
7. Identity: Which devices, channels, offline actions, and systems can or cannot be joined?
8. Breakdowns: Which segments must remain visible so an average cannot hide a material change?
9. Outcome: Which downstream customer and economic result prevents local metric gaming?
10. Companion evidence: Which path, cohort, buying-job, qualitative, or experimental view is needed before acting?

Version that contract with the chart. When a definition changes, start a new comparable series or restate history transparently; do not splice two measurement regimes into one trend line.

Use the funnel when its boundaries match the decision#

The marketing funnel is still useful when the unit is clear, stages correspond to observable states, movement rules are explicit, and the question is where an aggregate population narrows. It gives marketing and sales a shared vocabulary, shows where to investigate, and makes operational gaps visible.

Do not ask it to reconstruct every buyer’s path, represent simultaneous work by a buying group, measure long-run retention, or prove that a campaign caused revenue. Use paths for routes, cohorts for time, buying jobs for collective progress, and experiments for causality.