Mobile app marketing agency, measured past the install

How is app marketing different from marketing a website?

Six structural differences: a store listing sits between your ad and the user, the install starts the funnel rather than ending it, iOS attribution is deliberately limited, the two platforms behave like separate markets, fixes ship on review cycles, and your public rating compounds into every future install.

The first two together explain most of the money wasted in this category. An ad platform reports an install and calls it a conversion. Between your ad and that install sits a store listing you may never have tested, and after it sits every question that decides whether the user is worth anything. Optimising a campaign toward installs while ignoring both ends is the standard arrangement and it is why so many app budgets produce numbers that look fine and revenue that does not move.

What happened to mobile attribution, and what replaced it?

Apple replaced user level tracking with a privacy preserving model: install and re-engagement data arrives aggregated, delayed and withheld below certain thresholds. Campaign reporting on iOS is an estimate by design, and published opt-in figures disagree because vendors measure against different denominators.

The mechanics matter because they determine what you can and cannot know.

Since App Tracking Transparency, an app must ask permission before tracking a user across other companies' apps and websites. Where permission is not given, attribution runs through Apple's own framework instead. That framework, now AdAttributionKit and previously SKAdNetwork, sends a postback to the ad network after an install, carrying a small coded conversion value rather than user level data. The postback is delayed deliberately, and where too few installs share the same ad and conversion signature the detail is coarsened or withheld entirely so individuals cannot be identified.

Three practical consequences, which most case studies in this category quietly ignore.

Small campaigns report least. The privacy threshold means low volume campaigns receive the least detailed attribution. The campaigns you most need to learn from are the ones the system tells you least about.

Reporting is delayed. Same day optimisation in the way web marketers expect is not available. Decisions move to a weekly rhythm whether you like it or not.

Numbers will not reconcile. Your ad platform, your attribution provider and your own analytics will disagree, permanently, by design. Teams waste months trying to close that gap. The workable answer is to pick one source as the decision maker, use the others as directional, and move judgement toward cohort based measures that do not depend on per user attribution.

On opt-in rates, be careful which figure you are quoting. Adjust's 2025 reporting puts the industry average around 35 percent, measured among users who were actually shown the prompt, roughly flat across the preceding two years. AppsFlyer has reported around half of users consenting globally on a different basis. Those are not contradictory measurements, they are different denominators, and a plan built on the higher number without checking which one it is will assume visibility you do not have.

What does app store optimization involve?

App store optimization covers the metadata that decides where you rank and the creative that decides whether a visitor installs. Title and subtitle carry the most ranking weight, screenshots carry the most conversion weight, and store listing conversion rate is the number that makes every other channel cheaper.

We start here on nearly every engagement for a simple arithmetic reason. If your listing converts visitors to installs at 18 percent and you raise it to 25, every acquisition channel just became about a third cheaper, permanently, without raising a budget. Very few media optimisations produce that, and the work costs a fraction of a month of spend.

Two cautions. The stores index metadata differently, so treat iOS and Android as separate work rather than translating one into the other. And test screenshots deliberately rather than by opinion, because the first two are what most visitors see and the internal debate about them is rarely settled by argument.

What does it cost to acquire an app user?

Published cost per install figures for 2026 run roughly 4 to 6 dollars on iOS and around 2 dollars on Android, with wide variation by category and source. Treat these as a starting range only, because the sources disagree and none of them is your app.

Two numbers worth holding, with the caveats attached. Android installs are consistently cheaper than iOS, frequently by half or more, which is a real and stable pattern rather than a reporting artefact. And category matters more than platform: a hypercasual game and a financial services app are not in the same market for attention and should not be planned against the same figure.

The caveat is larger here than in most channels. These numbers come from attribution vendors reporting across their own customer bases, with different definitions and different mixes of apps, and none of them publishes in a way that lets you reproduce the figure. We use them to set an expectation before the first test and we stop using them the moment there is real data from your own campaigns, which takes a few weeks.

The more useful question is not what an install costs but what an install is worth, which is the next section. A 6 dollar iOS install is expensive for an ad supported app and cheap for a subscription app with decent retention. Cost per install on its own cannot tell you which situation you are in.

Why are installs the wrong headline metric?

Installs are the cheapest outcome to buy and the weakest predictor of revenue. Published cross category day 30 retention sits in the mid single digits, so the large majority of any install cohort is gone within a month and a report built on install volume is measuring something that mostly evaporated.

Work the implication through. If day 30 retention for your category runs around 5 percent, then a thousand installs is roughly fifty people still present a month later. Whether that was a good purchase depends entirely on what those fifty are worth and on whether the other nine hundred and fifty churned for a reason you could have fixed.

Retention also varies by business model more than by category. Reporting from attribution vendors puts subscription apps far above ad supported ones at day 30, which makes sense: a subscription creates a reason to return and a payment that signals genuine intent. If you are comparing your retention against a published average, compare against your model, not just your vertical.

The failure pattern we are called in to fix looks like this. Acquisition is switched on before activation is understood. The install numbers rise and the team reports progress. Retention is measured as an aggregate rather than by cohort, so the decay is invisible for months. Revenue stays flat, budget rises to compensate, and the eventual conclusion is that the channel does not work. The channel was fine. It was filling a container with a hole in it.

Which is why we will decline user acquisition work for apps with weak and unexplained retention. Taking that budget would be straightforward and it would make your problem more expensive.

What should a mobile app marketing agency measure?

Store listing conversion rate weekly, retention and activation by cohort, revenue per install by source, payback period by channel, and the ratio of organic to paid installs. Cost per install is a diagnostic and should never be the headline number in a report.

Cohort measurement is not a preference here, it is a workaround for the attribution limits described earlier. When per user tracking is unavailable or unreliable, comparing groups of users by the week they arrived still works, because you are measuring behaviour inside your own app rather than trying to follow a person across companies. It is the measurement approach that degrades least as privacy rules tighten further, which they will.

The last row is the one that tends to change strategy. If paid spend doubles and organic installs stay flat, the paid channel is not building awareness, a store ranking or word of mouth, and you should price it as rented traffic that stops the day you stop paying.

Which paid channels work for apps?

Apple Search Ads for intent on iOS, Google App campaigns for automated volume mostly on Android, Meta as the largest consumer acquisition channel, short video where your audience is younger, plus web to app and referral. Creative production rate is the main lever on the automated channels.

On the automated channels, understand what you are actually buying. You are not buying targeting decisions, you are buying access to a system that makes them. What you control is budget, the conversion signal you send back and the creative you supply, and of those three the creative is where most of the variance lives. A team that can produce and test new concepts weekly will outperform a team with better targeting instincts and a quarterly creative cycle.

Web to app is the one we push hardest and the one most often missing. Routing people through a web page before the store gives you a measurable step you fully control, a chance to capture an email address, and a first party record that survives whatever the platforms change next. It also lets organic search contribute to installs, which is otherwise difficult. Our search work covers that side.

How do you stop renting all of your growth?

Build the sources that keep working when spend stops: store ranking earned through retention, a web route that organic search can feed, lifecycle messaging that brings existing users back, and deep links that survive the install. Paid acquisition should be the accelerator, not the engine.

The test from the measurement section applies here. If paid spend doubles and organic installs stay flat, everything you have is rented. Five things change that, roughly in order of how much work they take.

Store ranking, which retention feeds. The stores rank on engagement and retention signals as well as metadata, so an app people keep using climbs on its own. This is the loop worth understanding: better retention produces better ranking, which produces cheaper installs, which makes the next cohort cheaper again. It is also why acquisition into a leaky app is doubly expensive, since the churn suppresses the free channel too.

A web route that search can feed. A real web presence that routes to the install gives organic search something to rank, and gives you an email address and a first party record before the store takes over. For anything with a subscription or an account, this is usually the highest return structural change available.

Lifecycle messaging that is actually wanted. Push and email bringing existing users back is far cheaper than acquiring replacements. It is also the fastest way to make people uninstall if you treat it as a broadcast channel, so trigger on behaviour rather than on schedule.

Deep links that survive the install. Someone who taps a link to specific content, installs, and lands on a generic home screen has been made to start over. Preserving that intent through the install is a small engineering job with an outsized effect on activation.

Editorial featuring. Unreliable as a plan and worth being prepared for. The stores feature apps that use new platform capabilities well and look good doing it, so the preparation is a well presented listing and a reason for an editor to care.

None of these replace paid acquisition in a competitive category. They change what paid acquisition is buying, from the whole of your growth to the part that compounds faster because of it.

What does the measurement setup actually require?

A defined activation event, cohort reporting by install week, revenue attached to those cohorts, and a sane event taxonomy. Most apps we audit have hundreds of tracked events, no agreed definition of activation, and no way to compare one install cohort against another.

Everything recommended on this page depends on this existing, and it usually does not. The minimum that makes the rest possible:

One activation event, written down and agreed. The specific moment a new user has experienced what the product is for. Not registration, which only proves they filled in a form. It is different for every app and arguing it out is half the value of defining it, because the conversation forces a decision about what the product is actually for.

Cohorts by install week, not aggregate charts. An aggregate retention number hides whether things are improving, because new installs constantly refill the top. Comparing the cohort that arrived in week one against week six is the only way to see whether a change worked.

Revenue attached to the cohort. Subscriptions, purchases or ad revenue tied back to the group that arrived together, so you can answer what a cohort was worth rather than what the app earned that month.

An event taxonomy small enough to use. Hundreds of events with inconsistent names is the normal state and it is worse than fifteen good ones. Name them consistently, document what each means, and delete the rest.

Server side where it matters. Purchases and subscription state should be recorded from your own backend rather than relying solely on client events, which get blocked, dropped and duplicated.

This is typically one to two weeks of engineering at the start of an engagement, and we would rather spend it than report confidently on numbers nobody can reproduce. It is also the part a pure media agency has no reason to do, which is a large part of why the problem persists.

What does it cost to work with us?

We bill hourly against an agreed scope rather than taking a percentage of media spend, so our fee does not rise when your budget does. Store listing work, activation analysis and measurement run at our senior rate, and campaign operation at our execution rate.

A first engagement covering the listing, the cohort measurement, an activation review and two paid tests typically runs 50 to 100 hours depending on the state of the analytics, then a lighter ongoing month.

Indicative until we have seen the app and its analytics. Media spend is paid by you directly to the platforms and never passes through us. Engagements start at ninety days with written outcomes, then month to month with thirty days notice.

When should you not spend on user acquisition?

Do not spend when retention is weak and unexplained, when you cannot say what a user is worth, when the store listing has never been tested, when you are buying installs to hit a number, or when your real audience is small enough to reach directly.

The first is the one we turn down most. It is also the one where being a company that builds apps is genuinely useful rather than a line in a pitch: when the problem is that first time users cannot get to the point where the product works, the fix is in the product, and we can do that part instead of selling you media against it.

If what you need is the app itself, see mobile app development. If the business behind it is a subscription product, the economics of retention and payback are worked through on what is B2B SaaS, and the same arithmetic applies to consumer subscriptions with different numbers.

• Mobile app developmentWhen the retention problem is in the product
• SaaS PPC agencyPaid search for software, measured on pipeline
• SaaS SEO agencyThe web to app route, and what still ranks
• Landing pages and CROThe web step before the store takes over
• Subscription economicsRetention, payback and why they decide everything
• B2B SaaS marketingThe whole function, not one channel