Accelerate your
app's growth, pixel by pixel.

Key metrics for measuring performance marketing success

A team optimises for the metric it puts on the dashboard. After several years of operating our own traffic across more than 180 geographies, we have come to a straightforward conclusion: most growth dashboards measure the wrong things, in the wrong order, at the wrong cadence. The following five layers represent the metric stack we now apply to every campaign we run.

The first observation worth making is that installs, on their own, contain almost no information. An install is the moment a user agrees to download a binary onto their device. It says nothing about whether the user intended to open the application, whether the application matched their expectations, or whether they will return. A campaign that delivers one hundred thousand installs and a campaign that delivers one hundred thousand engaged users may share an identical headline figure and yet have entirely different economic outcomes.

This is not a novel observation, but it is one that most performance marketing programmes structurally fail to act on. The metric the dashboard surfaces shapes the decisions the team makes; the decisions shape the budget allocation; and the allocation shapes the outcome. If installs is the first number on the page, installs is what gets optimised, even when everyone in the room would agree, in a meeting room, that retention is what matters.

The framework below replaces the headline number with five layered measurements, each of which constrains the layer above it.

01Day one retention, not day one installs

Installs is a count; retention is a verdict. If a campaign delivers 100,000 installs of which 20,000 reopen the application the following day, the campaign has produced 20,000 candidate users; the remainder were transient. Day one retention serves as the first credibility test on any traffic source. A channel that fails to clear a basic day one threshold cannot be rescued by anything further down the funnel.

02Day seven cohort retention curve

Single point retention figures, taken in isolation, can be misleading. The diagnostic value lies in the shape of the curve from day one through day seven. A healthy application typically shows a steep decline between day one and day two followed by a flattening of the curve. An unhealthy one continues to lose users at a roughly constant rate, which usually indicates that the original install was driven by friction (incentivised flows, misleading creative, fraudulent attribution) rather than authentic product interest. The curve, not the point, is the data.

03Day thirty return on ad spend, broken out by cohort

Return on ad spend (ROAS) is the metric most teams designate as their primary performance indicator. The principle is sound; the execution is usually flawed. Blended ROAS, reported as a single weekly figure, conceals more than it reveals: it averages over channels with very different decay characteristics and over cohorts at very different points in their payback curve. A more useful approach is to track ROAS by acquisition cohort and by source independently, and to compare each cohort against the same cohort six months earlier. This is the only reliable way to detect a channel whose efficiency is gradually degrading before the deterioration becomes structural.

04Payback period rather than blended lifetime value

The statement "this user is worth fourteen dollars over their lifetime" is comforting and operationally useless. The relevant question is not what the user is eventually worth in expectation, but how long the firm waits before recovering its acquisition cost. A ninety day payback at a multiple of 1.4 is, in almost every realistic capital structure, preferable to a 365 day payback at a multiple of 2.5, because cash recovered earlier can be redeployed and cash that may yet be recovered cannot. We benchmark every campaign against a sixty day payback target, with the curve plotted weekly.

05Net new users that compound

The fifth layer is the one most performance programmes never reach. Are the users the campaign acquires themselves becoming a source of further acquisition? The relevant indicators are the referral coefficient, the k-factor, and the organic uplift attributable to paid cohorts. If paid users bring in roughly 0.3 organic users each, the effective customer acquisition cost is approximately twenty three per cent below the reported figure. If they bring in zero, the programme is running on a treadmill: each install must be paid for in full, with no compounding contribution. The distinction is the difference between renting growth and owning it, and it should be measured monthly for every campaign in flight.

The cheapest install is rarely the best install. The best install is the one that, sixty days later, remains active on the user's device.

None of these five measurements is conceptually new. What is uncommon is the discipline of looking at all five together, in this order, every week. Most teams report one or two; almost none report all five. The gap between the first two and the full set is where most growth budgets quietly disappear.

Beyond the install: building a sustainable UA strategy

It is entirely possible to run a user acquisition programme that appears successful for six months and quietly destroys its own unit economics by the ninth. The pattern is common enough that we have developed a working hypothesis about what separates programmes that scale durably from programmes that decay. The distinction, in our experience, is structural rather than tactical.

Most user acquisition (UA) programmes decay over time. They begin with a strong campaign, ride a creative or geographic configuration for several months, encounter a plateau, and then slide downward as fatigue and audience saturation set in. The typical response is to add more spend, more creative variants, or additional channels, all of which buy time without addressing the underlying issue. By the point at which the financial reporting catches up, the programme is already three quarters into a budget that no longer recovers its cost.

Programmes that compound look qualitatively different. Drawing on our own experience operating traffic at scale, we have identified four structural decisions that consistently distinguish the two categories. None of them is a tactic; each is a commitment to how the function is constructed.

01Acquire users, not installs

The distinction between an install and a user is more than rhetorical. An install is a discrete event: it occurs once and is then complete. A user is a relationship characterised by a specific channel of origin, a specific motivating intent, and a specific point in the application's lifecycle. The context surrounding the install determines almost everything that follows.

The first question on a campaign brief should not be the target cost per install. It should be the following: what type of user is the campaign attempting to recruit, and what does the entry path need to look like for that user to remain engaged? A casual gamer recruited through a misleading playable advertisement is indistinguishable on day one from a casual gamer recruited through an honest one. By day seven the difference is measurable, and by day thirty it is decisive. The cost per install premium associated with honest creative is, in our data, recovered many times over by retained value.

02Cap concentration risk

The most common failure mode we observe is not a poor campaign; it is over-reliance on a single supply source. A team identifies an efficient combination on a particular advertising network, scales it aggressively, and arrives at a state in which seventy per cent of monthly installs originate from that single source. When the network changes its bidding algorithm, saturates the available audience, or restricts a creative format, the entire UA function loses its primary engine in a matter of weeks.

Our internal rule is straightforward: no single channel should represent more than forty per cent of monthly install volume, even when it is the most efficient available channel. We deliberately allocate spend to second and third tier sources at marginally worse efficiency, on the principle that optionality has a measurable price and that price is consistently lower than the cost of a forced reconstruction. When a primary channel inevitably falters, the programme retains functioning alternatives into which it can reallocate.

03Treat creative as infrastructure rather than output

Most UA teams produce creative in the manner of a kitchen producing meals: one batch at a time, in response to a specific request, against a near term deadline. This production model is the proximate cause of creative fatigue at month two and the recurring quarterly scramble for fresh assets that follows.

A more durable approach is to treat creative the way a manufacturing operation treats raw material: continuous, parallel, and surplus. Our internal pipeline ships in excess of two hundred variants per advertiser per month, organised into structural families (hook, demonstration, social proof, gameplay, problem and solution) that can be recombined indefinitely. Any individual creative may exhaust its useful life within two weeks; the pipeline producing creatives does not.

04Re-acquire users before they are lost

The least expensive user is the one who already uses the application. The next least expensive is the user who used to. Most UA programmes treat retargeting as an adjacent activity, run it through a separate team with a separate budget, and underspend on it by a factor of approximately five.

A more sustainable structure incorporates retargeting into the UA plan from inception. Every cohort acquired should have a re-engagement schedule defined at days fourteen, thirty and sixty. The arithmetic is consistent across categories: a user re-engaged at day thirty costs roughly forty per cent of the equivalent cost of a fresh user with comparable expected lifetime value. Underinvesting in re-engagement leaves the most efficient growth lever in the toolkit unused.

Sustainable user acquisition is not the search for the cheapest install. It is the construction of a system in which every install acquired makes the next install acquired more profitable.

The teams whose UA functions have continued to scale cleanly through the second quarter of 2025, past the iOS signal loss, past creative fatigue, and past the algorithmic shifts that have rendered older playbooks obsolete, share these four structural decisions in some recognisable form. None of the four is a campaign tactic. Each is a commitment to how the function itself is built.

An UA programme that appears to be working without an obvious explanation for why it is working is, in our experience, generally about three months from finding out.

Play, profit, repeat: mobile game monetization demystified

There are three architectures for monetising a mobile game and a great many opinions about which is correct. Our position, after working across the full range, is that the right architecture is determined by a single variable, namely the retention curve, and that most studios select the wrong architecture because they reason from genre instead.

Every quarter brings a fresh argument that hybrid monetisation is the inevitable future, that pure in-app purchase has been exhausted, or that hyper-casual advertising has run its course. The arguments are typically articulate and almost always misdirected, because they engage the question at the wrong level of abstraction.

Monetisation architecture is not a matter of strategic preference. It is a function of how long users remain in the application. When the retention curve is correctly characterised, the appropriate monetisation model becomes substantially self-evident. When the curve is misread, no amount of pricing sophistication will recover the lifetime value left on the table.

Drawing on direct experience scaling games across all three architectures, the following framework summarises the recommendation we now apply before any campaign is constructed.

01Day thirty retention below eight per cent: in-app advertising

Games with short retention profiles, including most hyper-casual and many casual arcade titles, lack the temporal window required to develop a monetisable relationship with the user. Requesting payment on day three from a user who will be gone by day five is, in effect, requesting capital from a stranger, and the response rate is correspondingly poor.

What functions in this regime is the monetisation of attention while attention is available. Rewarded video, interstitials and banners all play a role; density and pacing matter more than format. A reasonable benchmark, in our data, is at least one advertisement every sixty seconds of active play, with rewarded placements positioned at meaningful friction points. Executed competently, average revenue per daily active user typically lands between five and twenty cents, which is sufficient to recover acquisition cost at scale even when day thirty retention is in single digits.

02Day thirty retention above twenty five per cent: in-app purchase

Games with long retention profiles, including most match-three, role-playing, strategy and social casino titles, exhibit the inverse property. A user retained at day thirty is statistically likely to be retained at day three hundred, and the upper five per cent of such users are willing to pay substantial sums for advantages, content and customisation. Showing those users banner advertisements forfeits an order of magnitude of recoverable lifetime value.

The relevant architecture is what the industry refers to as whale economics, although it is more accurately described as standard commerce: a small fraction of users contributes the majority of revenue, and the design objective is to ensure those users encounter regular and meaningful reasons to spend. The relevant indicator is not conversion rate but spending frequency among converted users. A well-designed in-app purchase economy can sustain monthly average revenue per paying user above forty dollars within its top tier; the median user contributes nothing, and that distribution is correct.

03Day thirty retention between eight and twenty five per cent: hybrid, with caution

This intermediate band is the territory in which most monetisation failures occur. The intuitive response is to combine the two prior architectures and accept whatever revenue results. The practical outcome is more often a configuration in which advertising cannibalises the most engaged users (whose tolerance for friction is the lowest) while in-app purchase underperforms (because the population of high-intent payers is structurally too small).

A functional hybrid architecture is not, in our view, advertising plus in-app purchase. It is segmented monetisation: advertising for non-payers, in-app purchase for payers, and a clear operational boundary between the two cohorts. The mechanism that supports this segmentation is the "remove advertisements" purchase, typically priced at a low entry point. It serves a modest revenue function and, more importantly, identifies which users belong in the in-app purchase funnel and which do not. Once that segmentation signal is available, the experience can be tuned independently for each group.

04Three principles that hold across all three architectures

Three further observations apply uniformly across the three monetisation models, and are the principles we observe being violated most frequently:

• Do not switch architectures after launch. Migrating from in-app advertising to a hybrid model six months into operation typically degrades retention, often severely. The architecture should be selected before soft launch and committed to.
• Monetisation design is creative design. The placement of an offer, the timing of a rewarded advertisement and the language of an in-app purchase prompt have a larger effect on revenue than the underlying price points. The design of these surfaces deserves the same attention as core gameplay.
• The arithmetic should be reviewed weekly, not quarterly. Retention curves drift over time. A title that occupied the in-app purchase regime six months ago can migrate into hybrid territory as its audience broadens. The monetisation architecture should follow the curve, not resist it.

Select the monetisation architecture appropriate to the retention curve. Then select the user acquisition strategy appropriate to the monetisation architecture. Reversing the order of those two decisions is the most reliable way to ensure that neither will work.

None of this analysis is theoretical. Every title we have personally taken to scale has followed this sequence: characterise the curve, select the architecture, design the campaign last. Studios that attempt to bypass the first step and begin with the campaign typically have to reconstruct the campaign within two quarters, and the cost of that reconstruction reliably exceeds the cost of measuring the curve in the first instance.

Once the architecture is correctly selected, monetisation ceases to be the most difficult problem in the business and becomes the most tractable.