Against the funnel: a long argument for a different mental model.
The marketing funnel has, since the 1920s, been the dominant mental model for how customers move from awareness to purchase. It is, in the mobile app context, almost completely wrong. What we propose to use instead.
The marketing funnel — that simple downward-tapering diagram with "Awareness" at the top and "Purchase" at the bottom — has been the most durable mental model in marketing for the better part of a century. The original version, attributed to the American advertising theorist E. St. Elmo Lewis in 1898, was called the AIDA model: Attention, Interest, Desire, Action. The funnel shape was added later, by which point the basic structure was already embedded in how marketing departments understood their work.
I want to argue, in this essay, that the funnel is now an actively harmful model for thinking about how mobile-app customers come to use mobile apps. It was a reasonable approximation of the relevant dynamics for most of the twentieth century. It became a worse approximation as the internet matured. It became a substantially misleading approximation when mobile apps became the dominant consumer software category. In 2026, the funnel is, in my view, doing more damage to marketing teams' ability to think clearly than any other inherited mental model in the field.
I will spend the rest of this essay explaining what is wrong with the funnel as a model of mobile-app customer behaviour, and then proposing a different mental model that I think more accurately describes what actually happens.
What the funnel assumes
The funnel assumes a one-way, broadly linear, broadly time-bounded journey. The customer starts at the top, in a state of unawareness. They acquire awareness, then interest, then desire, then they take action. The journey takes a finite, knowable amount of time. The customer's relationship with the product, at the end of the journey, is broadly the relationship the funnel was designed to produce.
This was a reasonable model when the action at the bottom of the funnel was, for instance, the purchase of a specific physical product on a specific occasion. The Cadillac advertising of the 1920s, for instance, was reasonably well-modelled by the funnel: a customer encountered the advertising, considered the purchase over a period of weeks or months, and either bought a Cadillac or did not. The transaction was discrete. The journey was finite. The model approximately matched the underlying dynamic.
The funnel is a much worse model of what happens when the action is "downloads a mobile app and starts using it." The action is not a single discrete event. The customer's relationship with the product, after the action, is not the relationship the funnel was designed to produce. The journey is not finite. The journey is, in many cases, not even one-way. The funnel, applied to mobile-app customer behaviour, is a model that has been stretched well beyond the conditions it was designed for.
What actually happens with a mobile-app customer
The mobile-app customer's relationship with a product does not follow a funnel. It follows, in most cases, something closer to a cycle. The customer encounters the product (sometimes by accident, sometimes by deliberate search), decides to try it, uses it for a period, lapses or churns, encounters it again, returns to it, uses it differently the second time, lapses again, returns, and so on. The cycle can repeat many times. The customer's relationship with the product is shaped not by a single linear journey but by the accumulated experience of many encounters.
This is not, I want to stress, a marginal phenomenon. The published lifetime-value data from any reasonably mature mobile app shows it clearly. The single largest source of "new" monthly active users in most apps is not actually new users at all — it is returning lapsed users. The single biggest predictor of long-term value is not the depth of the first session but the number of return cycles. The funnel, with its single arrow from awareness to action, has no way of representing any of this.
"The funnel is a model of how a customer buys a Cadillac in 1925. It is a poor model of how a customer relates to a mobile app in 2026. We have been forcing the wrong shape onto the data for years."
What is the right mental model?
I want to be careful here, because the temptation is to replace the funnel with something else that is equally tidy and equally misleading. The honest answer is that the right mental model for mobile-app customer behaviour does not have a tidy diagram. It looks more like a network of states, with the customer moving between the states in a non-linear way, sometimes repeatedly, sometimes for different reasons.
The states themselves are, in my proposed model: unaware, aware, trialling, active, lapsed, and returned. The transitions between states can happen in any direction. A customer can move from active to lapsed and back to active, possibly several times. A customer can move from aware directly back to unaware if enough time passes without re-engagement. A customer can move from returned to active in a way that is qualitatively different from their original move from trialling to active.
The model is uglier than the funnel. It is also, in my view, considerably more honest about what actually happens. It captures the cyclical nature of the customer relationship in a way that the funnel does not. It does not pretend that the customer's journey is finite. It does not pretend that the relationship is the same the second time round as it was the first. It allows the marketing team to think about which transitions they are trying to influence, in which order, for which segments of customers.
What this implies for how we work
Adopting a non-funnel model changes the work of marketing in three concrete ways.
The first change is in measurement. The standard funnel metrics — conversion rate, abandonment rate, time-in-funnel — are no longer the most useful metrics. The more useful metrics, in a state-based model, are transition rates between specific states (active-to-lapsed, lapsed-to-returned) and the population sizes at each state at a given point in time. These metrics are, in honesty, less easy to put on a dashboard. They are also, in our experience, considerably more useful when you are actually trying to make decisions.
The second change is in budget allocation. The funnel model encourages a particular kind of budget allocation: spend at the top to fill the funnel, optimise the middle, measure the bottom. The state-based model encourages a different allocation: spend disproportionately on the transitions that are both common and high-leverage. The transition from lapsed back to returned is, in most mobile apps, far more cost-effective to influence than the transition from unaware to aware. Marketing teams that have adopted state-based thinking are, almost without exception, reallocating budget towards the middle of the customer cycle and away from the top.
The third change is in narrative. The funnel encourages a narrative of growth: more at the top, more at the bottom, bigger overall. The state-based model encourages a narrative of relationship health: how many active users do we have, how is the rate of lapse looking, are the lapsed users returning. This is a much more honest narrative for a mature mobile app. It is also a narrative that fits the kind of business that most mobile apps actually are.
What I would not say
I do not want to argue that the funnel is useless. It is, for some kinds of one-time transactional business, still a reasonable approximation. If you are selling a single product in a single transaction with no realistic possibility of a repeat purchase, the funnel will not lead you badly astray. The problem is that most mobile-app businesses are not that kind of business. Most mobile-app businesses depend, structurally, on repeat engagement over time. The funnel is the wrong shape for those businesses, and the wrong shape produces wrong decisions.
I also do not want to argue that the state-based model I have sketched is the final word on this. There are several variants of state-based models in the academic marketing literature, and at least one of them is, I think, more sophisticated than what I have proposed. The point is not the specific model. The point is that the funnel needs to be retired from its position as the default mental model for mobile-app marketing, and that something better — almost any better thing — should replace it.
The marketing teams that have made the switch are, in our observation, making better decisions and producing better business outcomes than the teams that are still working from the funnel. The switch is not easy. It is, however, increasingly important. The longer we keep using a model designed for 1925 to think about a business that runs in 2026, the more time we spend optimising for the wrong things.