A conversion funnel is a model of the path to a completed action, divided into ordered stages, with a count of how many people remain at each one. The shape supplies the name: many enter at the top, few emerge at the bottom. Its function is arithmetic rather than descriptive. Counting the loss between adjacent stages turns one aggregate conversion rate into a series of stage-to-stage rates, which locates where the loss happens rather than merely confirming that it happened.

The conversion is defined by whoever builds the funnel: a purchase for a retailer, a demonstration request for a software vendor, a subscription for a publisher. The stages above it are whatever actions reliably precede that outcome.

Stages and the arithmetic of drop-off

Two vocabularies coexist. The strategic version labels stages by mental state: awareness, consideration, intent, conversion, and in longer variants loyalty, compressed by marketers into upper, mid and lower funnel. The operational version labels them by observed event: a product list viewed, an item opened, a basket filled, a checkout begun, a payment submitted. Analytics software counts the second.

The arithmetic is unforgiving. If 100,000 sessions yield 4,000 product page views, 1,200 add-to-cart events and 360 purchases, the aggregate rate is 0.36 percent, but the stage rates are 4 percent, 30 percent and 30 percent. The aggregate cannot distinguish a traffic problem from a checkout problem. The stage rates can.

One setting changes the result more than any other. A closed funnel admits only people who entered at the first step; an open funnel admits entry at any step. Google Analytics 4 documents both, defaults to closed, and permits up to 10 steps and four segments per funnel exploration. Open funnels report higher completion counts because they include people who arrived midway, so comparing an open funnel in one period against a closed one in another produces a change existing only in the configuration.

Event naming determines whether any of this works. Google's ecommerce documentation defines the recommended sequence as view_item_list, select_item, view_item, add_to_cart, view_cart, begin_checkout, add_shipping_info, add_payment_info and purchase, each carrying an items array. The built-in checkout report uses a fixed subset that cannot be edited, custom event names do not populate those reports, and a missing step registers as silence rather than as an error. Two GA4 and server-side Google Tag Manager defects documented in September 2025 inflated counts through service worker duplication, distorting exactly the ratios funnel analysis depends on. Microsoft Clarity added a Funnels feature in June 2024 joining step drop-off to session recordings and heatmaps, pairing a rate with footage of the failure.

Origin and evolution

The stage model predates the funnel drawing. Edward Kellogg Strong Jr, writing in 1925, credited E. St. Elmo Lewis with formulating the sequence attract attention, maintain interest, create desire in 1898, with get action appended later; C. P. Russell coined the acronym AIDA in Printers' Ink in 1921. That attribution is disputed. A doctrinal review in the Proceedings of the Conference on Historical Analysis and Research in Marketing found no Lewis writings from 1898 supporting the claim and credited Frank H. Dukesmith and Arthur Frederick Sheldon with much of the formulation.

The metaphor is traceable to William W. Townsend, whose 1924 book Bond Salesmanship described the sales task as forcing a broad set of facts through a funnel toward one specific consideration. For most of the twentieth century the model carried no measurement. Web analytics changed that, turning a diagram into a report.

The first serious challenge came in June 2009, when McKinsey published research by David Court, Dave Elzinga, Susan Mulder and Ole Jorgen Vetvik covering roughly 20,000 consumers across five industries and three continents. Buyers frequently added brands during active evaluation rather than narrowing steadily, inverting the funnel's defining assumption; McKinsey proposed a circular consumer decision journey with a loyalty loop instead.

Google followed in 2020, when Alistair Rennie and Jonny Protheroe of its UK consumer insights team published Decoding Decisions, a 98-page report describing the space between trigger and purchase as a messy middle where buyers cycle between exploration, an expansive mode, and evaluation, a reductive one, under six cognitive biases including social proof, scarcity and authority.

Rejection of linearity then became a platform talking point. At Google's NewFronts presentation on 23 May 2025, executives introduced four behaviours - streaming, scrolling, searching and shopping as a replacement for stage thinking, citing data that roughly 80 percent of consumers skip paths on the way to purchase; Katie Klein of Omnicom USA said marketers needed to rethink the traditional funnel entirely. MiQ went further in April 2026 with a study drawing on 700 trillion signals across 53 million households and surveys of more than 4,000 consumers and 600 advertisers, reporting that 72 percent of consumers under 34 had completed entire purchase journeys inside social apps.

How platforms encode the funnel

Despite the criticism, the funnel remains the organising structure of nearly every advertising interface. Campaign objectives are grouped by stage, budget authority delegated by stage, reporting segmented the same way. TikTok launched Brand Consideration Ads on 13 May 2026 as an explicitly mid-funnel objective.

Signal standards encode it too. IAB Tech Lab's ECAPI 1.0 specification, finalised in 2026, was named to cover full-funnel activity rather than terminal conversions alone, classifying page_view, ad_impression and search as upper-funnel events alongside a six-stage lead sequence from generate_lead to close_convert_lead. Bidding systems consume the same taxonomy: Google product manager Kristina Park described Journey Aware Bidding in March 2026 as feeding intermediate events into the model when a purchase takes seven to fourteen days from click, weighting signals before the terminal event occurs.

Reporting has followed. Amazon extended conversion path reporting worldwide on 13 November 2025, showing the top five observed sequences of ad touchpoints within a 30-day pre-purchase window, with historical data from 15 September 2025.

Why it matters for marketers

Stage rates decide where money goes. A funnel failing at the top implies a targeting or creative problem; one failing at the bottom implies a site, price or fulfilment problem, and the two have different owners and different budgets.

Checkout data illustrates the point. Baymard Institute, aggregating 50 studies, puts average cart abandonment at 70.22 percent, a figure barely moved in a decade. DHL research published in August 2026 found 67 percent of shoppers abandoning over delivery terms while only 52 percent of sellers identified that cause, with the payment page named as the most common drop-off point at 32 percent. Media budgets optimised into a checkout that fails on shipping terms buy traffic into a leak.

The lower stages carry legal exposure too. The Hessian data protection authority ruled abandoned-cart emails unlawful in June 2025, holding that no contractual relationship exists where a purchase was never completed, making such messages advertising that requires consent under the General Data Protection Regulation.

Limitations and disputes

Three objections recur. The first is empirical: measured journeys are not linear, and a closed funnel discards everyone entering midway, which is most people in a fragmented media environment.

The second concerns the upper stages, which resist measurement precisely where budget defence is hardest. Meta removed the 7-day and 28-day view-through attribution windows from its Ads Insights API with effect from 12 January 2026, eliminating reporting many awareness campaigns used to demonstrate value, a mechanism examined in PPC Land's explanation of lookback windows. Google withdrew Display and Video support from Performance Planner on 9 March 2026, removing impression share as a planning metric. Countermeasures run the other way: Google reported an average 31 dollar sales gain per additional attributed branded search on 29 June 2026, and Walmart Connect routed purchase data into Display and Video 360 for YouTube campaigns that month. Both attach a terminal number to upper-funnel spend, and both rely on the media seller grading its own work.

The third objection is that stage rates are correlational. A funnel records what happened, not what advertising caused, the separate question addressed by incrementality testing. Agentic commerce adds a structural problem: when an artificial intelligence agent negotiates checkout through backend interfaces, merchants receive completed orders or explicit failures rather than partial progress, so browse-to-cart and cart-to-purchase measurement loses its subject.

What it is not

A sales funnel or pipeline tracks opportunities through stages assigned by salespeople in a customer relationship management system, while a conversion funnel counts anonymous or pseudonymous behavioural events. The two are often merged in business-to-business reporting.

A customer journey map is descriptive, cataloguing touchpoints and emotions without counting volume at each. A conversion funnel is quantitative and ordered.

A conversion path is the observed sequence of advertising touchpoints preceding a purchase, as in Amazon's report. The funnel is an imposed model; the path is retrieved data. A conversion rate, meanwhile, is one ratio; a funnel is that ratio decomposed.

Recent developments

Instrumentation continues to shift server-side. Google routed Shopify purchase events into Google Analytics 4 through a server-to-server connection from July 2026 and added four more ecommerce events on 17 August 2026, though only purchase travels server-side, leaving the upper stages browser-dependent. New platforms build the lower funnel last: OpenAI added a pixel and conversions interface to ChatGPT Ads in mid-2026, having launched on awareness and traffic objectives alone.

Vendors are bridging the stages rather than choosing. Cint merged brand lift and sales lift into a single dashboard in June 2026, addressing the case where upper-funnel spend produces lower-funnel outcomes no single report captures. The pattern across PPC Land's coverage of digital marketing measurement is a market that has rejected the funnel as a description of behaviour while keeping it as the interface for buying, reporting and budgeting.

Timeline

  • 1898: E. St. Elmo Lewis credited, by Edward Strong in 1925, with formulating attention, interest and desire as sales stages
  • 1921: C. P. Russell coins the acronym AIDA in Printers' Ink
  • 1924: William W. Townsend applies the funnel metaphor to the sales process in Bond Salesmanship
  • June 2009: McKinsey publishes the consumer decision journey, based on roughly 20,000 consumers, challenging linear narrowing
  • 2020: Alistair Rennie and Jonny Protheroe publish Decoding Decisions, describing the messy middle
  • 23 May 2025: Google NewFronts introduces the streaming, scrolling, searching and shopping framework
  • 21 July 2025: Google Analytics adds Lead Acquisition and Lead Disqualification and Loss reports
  • 13 November 2025: Amazon extends conversion path reporting worldwide
  • 12 January 2026: Meta's removal of 7-day and 28-day view-through windows takes effect
  • 9 March 2026: Performance Planner drops Display and Video campaign support
  • April 2026: MiQ publishes research across 53 million households questioning funnel rigidity
  • 13 May 2026: TikTok launches Brand Consideration Ads as a mid-funnel objective
  • 29 June 2026: Google publishes attributed branded search figures for YouTube brand campaigns
  • 17 August 2026: Shopify sends four additional ecommerce events to Google Analytics 4

Summary

Who: Marketers, analysts and conversion optimisation specialists build and read funnels. The measurement layer is supplied by Google Analytics 4, Microsoft Clarity, Contentsquare and comparable tools; the activation layer by advertising platforms including Google, Meta, Amazon and TikTok, which organise campaign objectives by funnel stage. IAB Tech Lab standardises the event vocabulary through ECAPI. Baymard Institute, McKinsey and MiQ supply the benchmark and critique literature.

What: An ordered model of the path to a defined conversion, counting the population remaining at each stage so that drop-off can be located rather than merely totalled. Stages are named either by mental state, from awareness through consideration to conversion, or by observed event, from a product list view through checkout to purchase.

When: The stage logic dates to formulations credited to 1898 and the funnel drawing to Townsend in 1924, though the 1898 attribution is disputed in the historical literature. Measurement made it operational with web analytics. McKinsey challenged its linearity in June 2009, Google's messy middle research followed in 2020, and platform-side rejections arrived in May 2025 and April 2026 even as the same platforms retained funnel stages as their buying interface.

Where: Inside analytics products, tag management implementations, advertising platform reporting, and the conversion signal specifications that connect them.

Why: A single conversion rate identifies that loss occurred without identifying where. Stage rates assign the problem to an owner, which is why the model persists commercially despite substantial evidence that real purchase journeys are neither linear nor ordered, and despite the fact that funnel counts describe correlation rather than advertising's causal contribution.