Why a web conversion tree does not fit an app store
A website tree runs from visits to conversions on pages the team owns and measures end to end. An app journey has a gap in the middle. The person sees an advert or a search result, lands on a store listing designed within the store's own rules, installs, and only then opens something the team controls. The store reports some of that journey, the advertising platforms report another part, and the app's own analytics begin at first launch. Each source counts differently and none sees the whole path.
Two further breaks matter. Store ranking responds to install velocity, so paid installs can lift organic ones and a featuring can make paid campaigns look cheaper; branches that a web tree would treat as independent push on each other. And on part of the traffic, platform privacy frameworks report campaign results in aggregate and after a delay, so some leaves are modelled rather than counted. The app marketing hub maps the channels involved; the tree has to show how they combine and how far each number can be trusted.
Decisions that change for app installs
| Tree step | Generic tree | App install tree | Decision it forces |
|---|---|---|---|
| Root | Conversions or revenue in the period | Activated users, defined by an early in-app milestone | Which first-value event counts and within what window |
| Landing step | Landing page owned by marketing | Store listing governed by the store's rules | Who owns the store branch and its experiments |
| Channel split | Channels are additive and independent | Paid and organic linked through store ranking | How to read one branch when the other moves |
| Data quality | Leaves measured directly | Some leaves aggregated or modelled under privacy rules | Which leaves are labelled estimates |
| Guardrails | Cost ceiling | Early uninstall and cost per activated user | Which signal stops a cheap install from counting as success |
Root the tree in activated users
Pick a milestone inside the app that shows the person got what they downloaded it for: a first saved item, a first booking, a completed setup. The test is the same one used for any activation event, namely that people who reach it go on to stay far more often than people who do not. Install counts then become the first driver beneath the root, multiplied by an activation rate. The consequence is deliberate: a campaign that doubles installs and halves activation has not moved the outcome, and the tree says so on its face. The AARRR framework is a useful prompt for choosing the milestone, since its activation stage asks the same question. To pick the milestone from the app's own retention data rather than by opinion, follow the tutorial on finding the app activation event.
Give the store page its own branch
Between the advert and the install sits the listing. Draw it as a short chain: store impressions, product page views and page conversion to install. These leaves belong to whoever runs app store optimisation, because screenshots, preview video, ratings and localisation are what move them. The decision is how to source the leaves: the store consoles report impressions and page views by traffic source, which lets the tree separate search visitors from browse visitors and from people sent by paid campaigns. A falling page conversion rate (CVRShare of visits or clicks that convert.) on paid traffic alone usually points at a mismatch between advert and listing, not at the listing itself.
Draw paid and organic as linked branches
In most trees, sibling branches add up and do not affect each other. Here they do. Heavy paid spend can push an app up category rankings and lift organic installs; a featuring or a press mention can make paid campaigns look cheaper for a week because some people who saw the advert would have installed anyway. The tree should keep the two branches separate in the arithmetic but carry a note on the link, and any judgement about paid efficiency should be checked against a period without unusual organic visibility. The app install campaigns guide covers the bidding and optimisation choices on the paid branch; the tree only needs to stop a reader crediting one branch for the other's movement.
Label the modelled leaves
Some campaign leaves on some platforms arrive aggregated, thresholded or delayed under privacy rules, and the figures shown in a measurement partner's dashboard are part counted and part modelled. Mark each such leaf with its source and method. The guide to app attribution, MMPs and SKAdNetwork explains why the numbers differ between tools; the tree's job is to keep a modelled figure from being compared one-for-one with a counted one. Treat AttributionAssigning conversion credit to touchpoints. on the paid branch as an estimate with a range, and lean on leaves from the app's own analytics, which count activation directly, wherever the decision allows.
Guard against installs that never become users
Two guardrails sit beside the root. The first is early uninstall: the share of new installs removed within the first day or so, which rises when traffic is poorly matched or onboarding asks too much too early. The second is cost per activated user, the paid spend divided by the activated users it produced, which replaces cost per installAverage ad cost of each app install. as the efficiency figure for the paid branch. Set against cost per install (CPACost of each conversion.), it shows at once whether a cheaper source is truly cheaper. The deep links and app analytics guide covers the event set that feeds both.
Decision points in a worked app install tree
An invented recipe app, Larder Lane, whose installs rise after a featuring while activated users stay flat
Larder Lane defines an activated user as someone who saves a recipe and opens it again within 3 days of install. In the week it was featured on a store's editorial page, installs rose from 12,000 to 30,000, yet activated users barely moved, from 4,770 to 4,800. Paid spend was £9,000 in both weeks.
| Branch and leaf | Week before | Featuring week | Owner | Measured or modelled |
|---|---|---|---|---|
| Store impressions | 400,000 | 1,500,000 | Store optimisation | Measured, store console |
| Product page views | 40,000 | 120,000 | Store optimisation | Measured, store console |
| Page conversion to install | 30% | 25% | Store optimisation | Measured, store console |
| Paid installs | 5,000 | 5,000 | Paid acquisition | Part modelled, privacy aggregation |
| Organic installs | 7,000 | 25,000 | Store optimisation | Measured, store console |
| Activation, paid installs | 45% | 44% | Product and lifecycle | Measured, app analytics |
| Activation, organic installs | 36% | 10.4% | Product and lifecycle | Measured, app analytics |
| Activated users (root) | 4,770 | 4,800 | Growth lead | Measured, app analytics |
| Day-one uninstall | 18% | 35% | Guardrail | Measured, store console |
| Cost per activated paid user | £4.00 | £4.09 | Guardrail | Part modelled |
Read from the root down, the tree places the gap in one cell. The store branch worked: impressions almost quadrupled and page views tripled, and although page conversion dipped, organic installs more than tripled. The activation leaf for organic installs collapsed from 36% to just over 10%. Session recordings showed why: editorial visitors were browsing, and the app demanded an account before showing a single recipe. Paid activation held, so the paid branch was not the problem, and the cost per activated paid user confirms it barely moved. The uninstall guardrail doubling is the same story told by the store.
Decisions: product moves account creation to after the first saved recipe; store optimisation adds a screenshot showing recipes without sign-up; the next featuring is reported on activated users, not installs.
Each decision in the tree did a specific job here. With installs at the root, the week would have been celebrated. Without a separate store branch, the dip in page conversion and the surge in impressions would have cancelled into a vague "more traffic". Without paid and organic split, the paid team might have been asked to explain a problem that sat entirely with organic newcomers. And the labels told the team which cell to trust most: the activation figures, counted in the app, rather than the modelled paid installs.
Measuring an app install tree: KPIs and cadence
Daily or weekly, watch the store branch and installs by source, because featurings, ranking changes and campaign launches move them within hours. Weekly, read activation rate by source and the early uninstall guardrail, since both settle within days of install. Monthly, review activated users at the root, cost per activated user on the paid branch and the trend in organic share, which reveals whether paid spend is lifting ranking or merely replacing installs that would have come anyway. Leaves from the store consoles and the app's analytics are counts; leaves from paid platforms on privacy-restricted traffic are estimates that settle late, so judge paid campaigns on a longer window than organic ones. The calculators for cost per acquisition and conversion rate help check each split.
Where an app install tree misleads
The arithmetic balances whether or not the activation milestone is the right one, so a milestone chosen for convenience will make the root look healthy while retention erodes; revisit it once longer-term data exists. Featurings and seasonal peaks bring populations that behave unlike normal traffic, and a tree compared across such weeks will blame the wrong branch. Cross-device and reinstall behaviour inflates install counts in ways no leaf separates cleanly. Modelled leaves can be revised after the fact, so a number read early may change. And the tree stops at activation: whether those users stay and pay is a retention question that needs its own tree, read by cohort.
Common mistakes
- Installs at the root. The cheapest download looks like the best result.
- Treating the store page as marketing's landing page. Its owner, rules and data are different.
- Adding paid and organic as if independent. Ranking links them, and credit leaks between branches.
- Comparing modelled leaves with counted ones. An estimate and a count are not the same kind of number.
- Cost per install as the efficiency measure. Cost per activated user is the figure that tracks the root.
- Judging a featuring on its install spike. The early uninstall guardrail tells the rest of the story.
