Scope: what this audit covers and what it does not
The question this audit answers is whether an app's marketing brings the right users, gets them to first value and brings them back, and whether its numbers can be trusted enough to decide where money goes next. It treats the app's marketing as a system: the activation event, attribution and install reconciliation, deep links, the store listings and ratings, install campaigns, the routes from the website into the app, the first session and its permission requests, push and in-app messaging, lapsed users and cohort reporting.
Product features and app performance are out of scope; a crash or a slow screen is product work, even when it shows up in reviews. Email to app users is scored by the email programme audit, and the events and reports behind the website are scored by the analytics implementation audit. A single release, launch or send is checked rather than scored: the app store listing checklist covers one listing, the app install campaign launch checklist one campaign and the mobile message send QA checklist one message.
It does not re-teach the work or repeat how it is run. The app marketing operations playbook runs app marketing week to week, and the mobile and app marketing guide sets out a first app growth plan for a team without one. The AARRR framework is a useful frame for the areas below, from acquisition through activation to retention, but the scorecard rates the work, not the funnel. Every scorecard sits in the audit library.
Before you score: access and evidence
Ask for access to each store console, the attribution set-up, the analytics project and the tools that send push and in-app messages. Without the attribution set-up you cannot see where postbacks go, so the measurement and acquisition rows stay provisional until you have it. Fix an evidence window long enough to read several install cohorts past their early weeks, because retention only shows over time, and keep the same window for every rescore.
Collect:
- store console access for each platform, with listing text, screenshots, ratings and review replies;
- the attribution set-up, through a mobile measurement partner (MMPAn independent service that attributes app installs and in-app events across ad networks.) or the platforms' own tools, with its postbacks and the install counts from each source;
- the event taxonomy and the definition of the activation event, with retention by install cohort;
- install campaign exports with optimisation events and spend by network for the window;
- the live push, in-app and onboarding messages with their triggers, caps and permission prompts;
- a test device run through campaign links, email links and web prompts, before and after a fresh install.
Write down what each campaign is meant to produce before you look at results. Where a source is judged on cost per activated user rather than CPIAverage ad cost of each app install., the CPA calculator helps check that the target is consistent with what an activated user is worth to the business.
Criteria and scoring
Rate each row missing (absent or broken), partial (present but incomplete, inconsistent or out of date) or solid (matches the "what good looks like" column). The scale is your judgement against that column, not a number. Rows marked gate decide the band on their own when they score missing. A row that does not apply, such as install campaigns for an app with no paid acquisition, is marked not applicable rather than missing.
Measurement
| Criterion | What good looks like | Score guidance |
|---|---|---|
| Activation event agreed (gate) | One activation event, shown to predict retention, is the goal for campaigns, onboarding and reporting | Missing: no agreed event, or installs treated as success / Partial: an event chosen by opinion, or used in reports but not in campaigns / Solid: tested against retention and used everywhere |
| Attribution set-up (gate) | Install and event postbacks reach the right networks, and privacy-limited or modelled results are read as such | Missing: no postbacks, or networks credited by their own claims alone / Partial: installs posted back but not events, or modelled results read as exact / Solid: installs and events posted back, limits recorded |
| Install numbers reconciled | Store, attribution and network install counts are reconciled on a rhythm and the gaps explained | Missing: one source quoted, the others ignored / Partial: compared when someone asks / Solid: a regular table with each gap explained |
| Deep links tested | Links from campaigns, email and the website open the right screen, including after a fresh install | Missing: links open the store or the home screen / Partial: work when installed, fail after a fresh install / Solid: tested on every route, deferred links included |
The activation event tutorial shows how to find the event from the app's own retention data and why the result is a working definition, and the Activation RateThe share of new users or sign-ups who complete the activation event, the action marking first real value, within a set window. follows from it. Score attribution against the guide to MMPs and SKAdNetwork, which explains postbacks, the privacy limits on iOS and the Android install referrer as Apple's and Google's documentation describe them; postback windows are left to those pages. The install reconciliation tutorial explains why the counts never agree, so the row asks whether the gaps are known, not whether they vanish. The deep link testing tutorial gives the route table and test matrix the last row is scored against.
Store presence
| Criterion | What good looks like | Score guidance |
|---|---|---|
| Store listing | Title, subtitle, screenshots and preview show the value people search for, localised where the app is offered | Missing: feature lists or outdated screens / Partial: a strong default listing, untranslated or untested elsewhere / Solid: value-led, localised and tested |
| Ratings prompts and replies | Rating prompts appear at moments of success within store rules, and reviews are answered and routed to the team | Missing: prompts at launch, or reviews unread / Partial: prompts placed well, replies occasional / Solid: prompts at success, every review read, tagged and routed |
| Campaign store listings | Campaigns with a distinct promise send people to a matching custom listing | Missing: every campaign lands on the default listing / Partial: custom listings built but not linked or not read / Solid: a listing for each promise that earns one, read against the default |
The listing row is scored against the app store optimisation guide, the ratings row against the rating prompts and review replies tutorial, which keeps prompts within Apple's review guidelines and Google Play's rules as it states them, and the last row against the custom store listings tutorial. Character limits and review rules sit on the stores' own pages, and this scorecard does not repeat them.
Acquisition
| Criterion | What good looks like | Score guidance |
|---|---|---|
| Install campaign optimisation | Campaigns optimise towards in-app events once volume allows, with install quality and fraud checked | Missing: optimised to installs indefinitely, quality never checked / Partial: event optimisation on some networks, fraud checks informal / Solid: deliberate event choice per network, quality and fraud reviewed |
| Web-to-app routes | App prompts on the website appear where they help, link to the matching screen and keep people signed in | Missing: a blanket banner to the store / Partial: placed with care, but links open the home screen or sign people out / Solid: placed, linked and signed in |
The app install campaigns guide covers the choice between installs and in-app events, pacing and fraud; the web-to-app journeys guide covers who to move, where prompts belong and how to measure each route against the web visit it interrupted.
First session
| Criterion | What good looks like | Score guidance |
|---|---|---|
| Path to first value | New users reach the app's core value in the first session before being asked for anything | Missing: account forms and tours before anything useful / Partial: value reached, but after avoidable steps / Solid: the shortest path, with everything else deferred |
| Permission requests | Permissions are asked when their value is clear, with soft prompts that tell the truth | Missing: every prompt at launch / Partial: some moved to the moment of need, soft prompts vague / Solid: each asked at its moment, with an honest soft prompt |
Both rows rest on the onboarding and permission requests guide, which also explains why onboarding changes are read by install cohort. A notification prompt shown before value is the most common reason the Push Opt-In RateShare of app users shown the notification permission prompt who allowed notifications. disappoints, which is why this area sits ahead of messaging in the priorities.
Messaging
| Criterion | What good looks like | Score guidance |
|---|---|---|
| Push programme | Message types, triggers, frequency caps and quiet hours are set, and each push lands on the right screen | Missing: broadcasts with no caps / Partial: triggers in place, caps or landing screens inconsistent / Solid: typed, triggered, capped and deep linked |
| Notification preferences by topic | Users can choose topics instead of a single on or off switch | Missing: one switch / Partial: topics shown but not enforced in every sending tool / Solid: topics people recognise, enforced everywhere |
| In-app messages | In-app messages have a job, a trigger and a cap, and do not block the task the user came to do | Missing: modals on open / Partial: triggered, but no shared cap with push / Solid: a job, a trigger and a shared cap each |
| Lapsed user returns | Re-engagement is measured against a holdout, and there is a rule for when to stop | Missing: campaigns credited with every return / Partial: a holdout on some campaigns / Solid: holdouts throughout, and a stop rule |
Score the push row against the push notifications and in-app messaging guide, preferences against the notification preferences tutorial, in-app messages against the in-app message design guide and returns against the guide to re-engaging lapsed app users. Every message that links into the app depends on the Deep LinkA link that opens a specific screen inside a mobile app rather than its home screen. row above, so a broken link scores there, not twice.
Reading retention
| Criterion | What good looks like | Score guidance |
|---|---|---|
| Event taxonomy | Events and user properties follow a written taxonomy that marketing and product share | Missing: events named by whoever added them / Partial: a taxonomy written but not followed / Solid: written, shared and kept current |
| Cohort reporting | Retention is read by install cohort and source, never by totals alone | Missing: active-user totals only / Partial: cohorts read, but not split by source / Solid: by cohort and source, feeding budget decisions |
The deep links and app analytics guide sets out an event taxonomy built for apps, and the cohort analysis tutorial builds the table that lets each CohortA group of customers who share a starting event and period, tracked together over time. be compared with customers of the same age.
How to score and prioritise
Score the gates first. Without an agreed activation event and working attribution, campaigns optimise towards the wrong thing and no source can be compared with another. If either scores missing, mark the acquisition and retention rows provisional: a network that looks cheap may simply be the one claiming installs that never became users, and an ASOImproving an app's visibility in app store search and the share of store visitors who install it. test on the listing may be judged on the wrong outcome.
Then rank the remaining weaknesses in this order:
- Whatever loses users already won: broken deep links, a first session that asks before it gives and permission prompts that are refused. These waste every channel's work at once.
- The store listing and ratings, because every channel lands there: paid campaigns, web prompts and search alike.
- Messaging, re-engagement and campaign refinement next.
- Listing localisation and copy polish last, unless a market is new and the listing is its first impression.
Within each group, rank by the users the weakness touches. A broken link on the route carrying most new installs outranks the same fault on a seasonal campaign.
Reading the result: what to do with each band
Red means a gate scored missing. Hold budget increases and new networks, agree and document the activation event or repair postbacks, and point campaigns at the agreed event before trusting any source comparison. Rescore once an install cohort has been measured in full on the repaired set-up.
Amber means the gates hold at least partially but several other rows are partial or missing. This is the usual outcome. Work a short ordered list, one area at a time, and read each change by install cohort so its effect can be seen.
Green means the gates are solid and most rows are too. Move effort into listing and onboarding tests and the monthly cohort read that moves money between sources, and rescore each quarter or after a major release to catch drift.
From findings to a prioritised fix list
Write each partial or missing row as a finding in plain words: what you saw, the evidence and why it matters. A finding from the test device run is the most persuasive kind, so record the link, the app state and the screen it opened. Then turn the findings into one fix list with a line per fix: the finding, the fix, the owner, the guide that explains it and the rescore date. Order the list by the priorities above, not by how easy each fix is.
Group the lines by who does the work, because the activation event and the taxonomy sit with the analyst and product, postbacks and campaigns with the user acquisition manager, deep links and permission prompts with the developers, the listing and replies with the listing owner, and messages with whoever runs the messaging tools. Present the band, the gate findings and the first few fixes to the budget holder, and hand the rest to the playbook's rhythm so each fix has an owner after the audit ends.
Illustrative example
A meal-planning app whose installs rose while active users stayed flat
Nimrazo Meal Planner, an invented meal-planning app spending about £9,000 a month on install campaigns, runs the audit over the last twelve weeks after installs rose by a third while weekly active users stayed flat. The activation event scores missing: campaigns on both networks optimise for installs and the team's reports count installs as success, although the analyst's cohort table shows that users who build their first weekly plan are far more likely to return. Attribution scores partial, because installs are posted back but no in-app events are. Deep links score missing: links in the welcome email open the store page instead of the plan screen, even for people who already have the app. Permission requests score missing, because the notification prompt appears on the second screen, before any plan exists. The store listing scores solid and replies partial. The band is red on the activation gate. The fix list puts agreeing "first plan built" as the activation event and posting it back to both networks first, moving the notification prompt to after the first plan second, repairing the email links third and the review routine fourth, and holds a planned new network. Eight weeks later the gates score solid, deep links and permissions score partial and the band moves to amber.
Common mistakes
- Scoring campaigns before the gates. A network that reports cheap installs is not a good source until its users are measured against the activation event. Score the gates first.
- Treating a mismatch between install counts as a failure. The counts never agree; the row asks whether the gaps are known, stable and explained.
- Testing links only on a phone that already has the app. The fresh-install route is where deferred links fail. Run the test device both ways.
- Blaming the listing for a first-session problem. If users install and leave, the listing did its job. Score the first session before rewriting screenshots.
- Crediting re-engagement with every return. Some lapsed users come back anyway. Without a holdout, score the row partial at best.
- Reading retention from totals. A growing base hides falling retention in newer cohorts. Read by install cohort and source, and rescore on the same evidence window.
