Marketing Funnel: Stages, Assumptions, and Where It Breaks
A marketing funnel groups potential and existing customers by their relationship to a purchase—often awareness, consideration, conversion, and loyalty—so a team can see where a defined population narrows. That makes it a useful operating map. It is a poor literal account of how buying happens: people enter late, skip stages, loop backward, pause, 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.
Marketing-funnel guides commonly use the model to categorize relationships from awareness toward purchase and loyalty. According to Shopify’s funnel guide, neither source presents its stage labels as a universal standard.
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.
Shopify and LinkedIn explain that common sources distinguish the marketing and sales views by their interaction focus, define TOFU/MOFU/BOFU as broad funnel regions, and define AIDA as attention, interest, desire, and action, as Open University also documents.
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.
| Practical stage | Relationship state | Decision the stage helps with | Evidence that could represent entry |
|---|---|---|---|
| Awareness | The relevant audience recognizes the problem, category, or brand | Where and how to become mentally available | A defined reach event, aided or unaided recall study, or first qualified visit |
| Consideration | A prospective buyer is actively learning, comparing, or validating | Which questions, risks, and alternatives need an answer | A qualifying research action, return visit, comparison interaction, or accepted handoff |
| Conversion | The buyer completes the commercial action the funnel was built to explain | Which friction, proof, or approval blocks the decision | Purchase, signed agreement, qualified signup, or another explicitly named outcome |
| Loyalty | A customer repeats, renews, expands, or advocates after experiencing the offer | Whether the relationship creates durable demand after acquisition | Renewal, repeat purchase, expansion, referral, or another observed post-purchase event |
The examples in the last column are not interchangeable. An ad impression is not proof of awareness; a page view is not proof of consideration; a form submission is not necessarily a qualified opportunity. The team has to choose an observable proxy and state what it does—and does not—mean.
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.
| Defined event | Eligible people | Conversion from prior event | Drop-off after this event |
|---|---|---|---|
| Solution-page view | 1,000 | — | 75% |
| Pricing-page view | 250 | 25% | 80% |
| Trial start | 50 | 20% | 60% |
| Activation | 20 | 40% | — |
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.
In GA4, entry rules, sequence requirements, step conditions, and time constraints are part of the funnel definition. Changing them can change membership and reported continuation even when the underlying events do not change.
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.
| Model | Useful question | Primary object | Main limitation |
|---|---|---|---|
| Marketing funnel | What relationship state is the market or audience in? | People or accounts interacting with brand, campaign, and product signals | Compresses many paths and influences into a few states |
| Sales funnel | What state is a lead or opportunity in relative to a commercial close? | Leads, opportunities, or accounts interacting with sales | Can mistake seller activity for buyer progress |
| Customer journey | What did a customer actually experience across time and touchpoints? | A person, account, or buying group moving through a path | Becomes unwieldy when it tries to represent every possible route |
| AIDA | What response should a message help produce: attention, interest, desire, or action? | The audience response to selling or advertising communication | Implies a staged hierarchy that behavior may not follow |
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.
A funnel is a ledger of defined states, not a biography of a buyer.
Every funnel quietly assumes how buying works
The model becomes trustworthy only when its assumptions are tolerable for the decision at hand.
| Hidden assumption | When it is tolerable | Signal that it has broken | Repair |
|---|---|---|---|
| One counting unit represents the buyer | A purchase is made by one identifiable person | Several contacts influence one account, or one person creates several records | Count the account or buying group and preserve person-level activity underneath |
| A signal represents a real state | The event is a close proxy with a written entry rule | Stage counts rise while downstream quality falls | Tighten the state definition and retain the raw event separately |
| States have a meaningful order | The task really requires a stable sequence | People enter late, skip, repeat, or move backward | Use an open funnel, allow explicit re-entry, or add path analysis |
| Identity and observation are sufficiently complete | Important actions occur in instrumented systems with durable identifiers | Cross-device, offline, privacy, or integration gaps create unexplained loss | Label observable scope and reconcile known blind spots instead of imputing certainty |
| The time window fits the decision cycle | Most entities have enough time to reach the next state | Recent cohorts look worse only because they are immature | Compare cohorts at equal age and publish the conversion window |
| Aggregation does not hide decisive differences | Segments share similar intent, economics, and process | A blended rate improves while a key market or channel deteriorates | Segment by material dimensions before choosing an intervention |
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.
Both research programs reject systematic one-way narrowing as a complete description of consumer decisions. McKinsey & Company research reports that they describe expansion, reevaluation, feedback, and post-purchase effects that a linear funnel collapses.
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.
Gartner’s public model treats B2B progress as collective completion of buying jobs rather than one buyer moving through an ordered sequence.
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.
Funnel counts are observational state summaries. GA4 and Google’s Messy Middle report show that sequence data can show that one recorded event preceded another, but causal effect requires a design that addresses what would have happened without the intervention.
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.
| If the question is… | Use… | What it adds |
|---|---|---|
| Where does a defined population narrow? | Funnel analysis | Comparable state counts and transition rates |
| Which routes, loops, and skipped steps occur? | Path analysis or journey mapping | Observed sequences and experience context |
| How does progress or retention change with time? | Cohort analysis | Equal-age comparisons and maturation visibility |
| What must several stakeholders accomplish to buy? | Account-level buying-job map | Collective progress without forcing one-person stages |
| Did an intervention cause an incremental outcome? | Randomized experiment or defensible causal design | A counterfactual rather than sequence-based credit |
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.
Put the counting rules next to 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.
- Decision: What action will change when this funnel changes?
- Entity: Are you counting people, accounts, opportunities, sessions, or orders?
- Population: Who is eligible to enter, and which exclusions apply?
- States: What observable condition begins and ends each stage?
- Movement: Is the funnel open or closed? May entities skip, repeat, move backward, or re-enter?
- Time: How long may a transition take, and how will immature entrants be handled?
- Identity: Which devices, channels, offline actions, and systems can or cannot be joined?
- Breakdowns: Which segments must remain visible so an average cannot hide a material change?
- Outcome: Which downstream customer and economic result prevents local metric gaming?
- 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.
Use the funnel as a ledger of observable states, not a law of buyer behavior. Keep the counts while they remain useful, but change the explanatory model when paths, cohorts, group decisions, or causal questions exceed what narrowing stages can show.
Frequently asked questions
What is the difference between an MQL and an SQL?
An MQL has met marketing’s documented fit and interest criteria, while an SQL has passed the additional qualification needed for direct sales attention. Salesforce distinguishes the two as different readiness states, but the threshold is not universal. Write the handoff rule as observable evidence—such as role, account fit, stated need, and accepted follow-up—and track rejected MQLs so marketing can repair a weak definition rather than merely increase volume.
When is a flywheel more useful than a funnel?
A funnel is better for locating loss between defined states; a flywheel is better for asking how customer success, advocacy, and operational friction affect continuing growth after the first conversion. HubSpot’s flywheel model treats customers as a source of force rather than the end of a narrowing process. Keep the funnel for denominator-based stage counts, then add a flywheel view when renewals, referrals, and service friction materially influence new demand.
Is a funnel time limit the same as an attribution window?
A funnel time limit sets how long an entity may take between defined steps, whereas an attribution window sets how long after an ad interaction a later conversion remains eligible for credit. GA4 applies time constraints to step sequences, while Google Ads defines conversion windows around ad interactions. Record both settings independently; shortening either one can lower the reported count for a different reason.
Does GA4’s average time between steps include people who abandoned the funnel?
GA4’s elapsed-time value is calculated for users who reached the next displayed step, using the first qualifying repetition when a step repeats. The GA4 funnel exploration documentation describes that calculation. Read the elapsed-time average beside the abandonment count and a time distribution, because people who never advanced are absent from that average and a small group of slow completers can still distort it.