Foundations of Growth Engineering: From Marketing Funnels to Closed-Loop Systems

The Definitional Crisis of "Growth"

The term *growth* has acquired a definitional plasticity that now threatens its analytical utility. Within a single growth-team standup, it is not unusual to hear "growth" used to denote paid acquisition, product-led onboarding, retention engineering, monetization, and—more rarely—the compound effect of all of these. This semantic overload is not a harmless colloquialism; it is a structural obstacle to treating growth as a system that can be designed, measured, and improved with predictive accuracy. When a metric is defined by the metric owner's current quarter, no engineering discipline can be erected on top of it. The first task of growth engineering, therefore, is not to optimize a funnel but to recover a precise vocabulary: what *is* a user population, what *is* a flow, what *is* feedback, and what *is* a loop.

The difficulty is exacerbated by an institutional reality. Growth, as a functional role, was invented less than two decades ago, when Sean Ellis's 2010 coinage of the term "growth hacker" implicitly reclassified a set of activities that had previously been distributed across product management, marketing analytics, and engineering. Since then, three distinct organizational species have emerged: the *growth marketing* team, which treats growth as a paid-media optimization problem; the *product growth* team, which treats it as a feature-experimentation problem; and the *growth engineering* team, which treats it as a systems problem. Each species tends to define growth in its own image. The result is that when an executive asks "what is our growth rate?", the answer depends on who is in the room.

The engineering framing introduced in this chapter does not reject the legitimate concerns of the other framings; it subsumes them. Paid acquisition is a flow into a stock of users. Onboarding optimization is a time-constant reduction. Retention engineering is a feedback gain adjustment. Monetization is a stock-and-flow relationship between a user base and a revenue base. The claim of growth engineering is that none of these problems can be solved in isolation, because they are coupled. A retention improvement changes the optimal acquisition rate; a monetization change alters the referral loop's gain; an onboarding change compresses the activation time constant and thereby increases the loop's asymptotic growth rate. The chapter is an argument that these couplings are not metaphorical: they are quantitative, and once quantified, they can be engineered.

A Critical Genealogy: From AIDA to Growth Loops

A rigorous defense of the loop-based framing requires an honest accounting of what preceded it. The intellectual lineage of modern growth thinking runs through four major artifacts, each of which captured part of the truth and obscured the rest. Understanding their failure modes is as important as understanding their contributions, because most contemporary teams still operate inside a conceptual residue of the older models.

AIDA and the St. Elmo's Fire of Late Nineteenth-Century Advertising

The AIDA model—Attention, Interest, Desire, Action—was codified by E. St. Elmo Lewis in 1898 to describe the psychological progression of a consumer encountering a print advertisement. Its structural assumption is that of a *unidirectional sequence*: an external stimulus moves a passive prospect through discrete cognitive states, culminating in a purchase event. The model is essentially a hydraulic metaphor in which attention is a fluid poured into a funnel, and each subsequent stage leaks a fraction of the previous stage's volume.

AIDA's contribution was to formalize the insight that not all prospects are equal in their proximity to conversion, and that the conversion probability is roughly stage-conditional. Its limitations, however, are severe and worth enumerating because they recur in every funnel model that descends from it. First, AIDA assumes a homogeneous population: the model does not specify how to handle prospects who have already been exposed to prior stimuli, who have partially converted, or who exit and re-enter the funnel. Second, AIDA is memoryless with respect to its own outputs: a converted customer does not, in the model, generate any new attention. The model has no concept of compounding. Third, AIDA has no time axis: there is no specification of how long a prospect is allowed to remain in the "interest" stage before being classified as lost. These omissions are not minor; they are the structural reasons why AIDA, taken literally, cannot describe any growth phenomenon that involves retention, virality, or repeat behavior.

The Marketing Funnel and the Industrialization of Demand

The mid-twentieth-century marketing funnel—typically rendered as Awareness, Consideration, Preference, Action, and sometimes Loyalty—retained AIDA's unidirectional-sequence structure while adding two refinements: an explicit appeal to mass-media channels and a tacit admission that not all conversion events are final. Loyalty, when included, gestures toward repeat behavior, but it is treated as a separate downstream stage rather than as a feedback path that re-injects users into the awareness pool.

The funnel model was enormously productive for the broadcast advertising era in which it matured. It aligned with the operational structure of mass-media buying: a brand purchased impressions, those impressions produced awareness, and the awareness stock decayed at a measurable rate. The model also aligned with the financial structure of the era: a campaign had a defined budget, a defined flight, and a defined cost-per-impression. The growth problem, in this framing, reduced to a media-mix optimization. Vanity metrics—reach, frequency, gross rating points—became the lingua franca because they were the controllable inputs of the system.

The funnel model's blind spot, visible only in retrospect, was its treatment of the customer as a terminal node rather than as a node with outgoing edges. The loyalty stage, where it existed, was an afterthought. The consequence was that customer lifetime value, referral, and word-of-mouth were systematically underweighted in marketing budgets, because they did not appear in the model. This is the historical reason why "growth" historically meant "acquisition" and why the growth function in many firms is still org-charted under marketing. The model produced the institutional arrangement, which then reinforced the model.

AARRR and the Software-as-a-Service Reframing

Dave McClure's 2007 AARRR framework—Acquisition, Activation, Retention, Referral, Revenue—marked a genuine conceptual advance. The acronym's five stages are not a strict sequence; they are five *rates* that jointly determine the trajectory of a user base. McClure's implicit insight was that a software product's growth is governed by a system of coupled flows, not a single conversion funnel. A user is acquired once, activated zero or more times, retained over a window, referred zero or more times, and monetized zero or more times. The model thus began to escape AIDA's memoryless trap by including Retention and Referral, both of which acknowledge that past users continue to influence current growth.

AARRR's contribution to the discipline is hard to overstate. It is the first model in the marketing canon in which the *output of one stage*—referred users—appears as the *input of another stage*—acquisition. This is the conceptual seed of the loop. Yet AARRR is still predominantly *prescriptive* and *categorical*: it tells practitioners which metrics to measure, but it does not specify the functional relationship among them. The model does not say how retention rate and referral rate combine to determine steady-state growth rate, nor does it specify the time constants that govern how quickly a cohort's behavior propagates through the system. AARRR is to the growth loop what Ohm's law is to a SPICE simulation: a necessary but