Why the default north star and OKR set does not fit app installs
The framework's general advice, choose a metric of value delivered and decompose it into inputs, assumes the team will resist the easy candidate. App teams face an unusually easy one: the store and the ad networks report installs every day, the number goes up with spend, and leadership often frames the goal in those terms. Adopting it creates three problems. The north star stops reflecting value; teams with very different levers, store optimisation and paid acquisition, end up sharing one number they cannot separate; and the data behind paid installs is modelled and delayed in ways a quarterly grade has to allow for. The app marketing hub treats installs as the start of an install-to-retention loop, and the decisions below keep the OKRs consistent with that view.
Decisions that change for app installs
| Element | In the generic framework | For an app-install goal | Decision it forces |
|---|---|---|---|
| North star | A count of value delivered | A value moment inside the app, never the install | Which in-app event marks value, and in what time frame |
| Inputs | Breadth, depth, frequency, efficiency | Installs, activation rate, early retention | Where install volume sits in the tree |
| Key results by team | Shared inputs with co-owners named | Store page conversion for the store team; cost per activated user for paid | How the two teams' contributions are separated |
| Data and scoring | A named source and baseline | Modelled, delayed paid data with stated caveats | Which reading counts and when it is final |
| Cycle | Usually a quarter | A quarter with a settlement window for activation data | When the grade is taken |
Set the north star at a value moment inside the app
Pick the first action that shows the user got what they came for: a completed lesson, a first match, a first order delivered, a conversation held. Then add a frequency, such as weekly, so the north star counts people receiving value repeatedly. The AARRR framework helps identify which activation event best predicts retention. The choice involves a trade-off between speed and meaning: an early event like completing sign-up is quick to read but weakly linked to value, while a later one is meaningful but takes longer to observe. Prefer the earliest event that clearly separates retained users from those who leave. The tutorial on finding the app activation event runs that comparison step by step, from candidate actions to one written definition.
Keep install volume as a key result beside activation
Installs belong in the tree as an input, and a growth objective can carry an install key result. What must not happen is an install key result on its own. Pair it with an activation key result under the same objective, so that buying cheap, low-intent installs to hit one number visibly drags down the other. The KPI tree framework shows the arithmetic: activated users equal installs multiplied by activation rate, so either factor can move the north star input and both have to be watched.
Separate store and paid key results
The store team controls the listing: screenshots, video, ratings and keywords. Its natural key result is store page conversion, the share of page visitors who install, which the app store optimisation guide covers in depth. The paid team controls bids, creative and targeting, and its natural key result is cost per activated user, a CPACost of each conversion. measured at the value moment rather than at download, as described in the app install campaigns guide. The two overlap, because paid traffic lands on the store page, so each key result should name the shared dependency. Choose this split whenever the two teams have separate budgets or leaders; a single small team can share one key result.
Write privacy caveats into how key results are scored
Paid install data, especially on iOS, is often aggregated, delayed or modelled, and in-app events may be reported only partly. The guide to app attribution, MMPs and SKAdNetwork explains the mechanics. For OKRs, the decision is procedural: each key result built on paid data states its source, whether modelled numbers count, and the date after which the reading is final. A blended measure, such as total activated users divided by total paid spend, is often a sturdier key result than a channel-level one, because it does not depend on which network claims which user.
Match the quarter to activation data maturity
If activation is measured within a set number of days of install, installs from the final days of the quarter cannot be judged until after it ends. Either shift the measurement window so the quarter covers installs that have had time to activate, or grade the activation key result a few weeks after quarter end. Lifecycle messaging, such as push notifications and in-app messages, can shorten time to activation and is worth its own input once the north star is set.
Decision points in a worked north star and OKRs for app installs
An invented language-exchange app, Parlo, sets a north star and one quarter of OKRs
Parlo pairs learners for spoken practice. Leadership asked for 400,000 installs in the third quarter. The growth lead proposed a north star instead: weekly learners who complete at least one 10-minute voice exchange, then at 58,000.
| Element | Parlo's choice |
|---|---|
| North star | Weekly learners completing a voice exchange (58,000; target 70,000 by quarter end) |
| Inputs | Installs; activation (first exchange within 7 days of install, currently 24%); week-4 retention of activated learners (46%) |
| Objective | Bring more new learners to their first real conversation |
| KR1 (growth, committed) | Installs from 290,000 to 380,000 in the quarter |
| KR2 (product and lifecycle, committed) | 7-day activation from 24% to 30% |
| KR3 (store team, aspirational) | Store page conversion from 27% to 33% across both stores |
| KR4 (paid team, committed) | Blended cost per activated learner from £6.80 to £5.50, read from MMP data settled 21 days after quarter end |
| Cycle | Installs counted 1 July to 23 September so every install has its 7-day activation window inside the quarter |
End of quarter: installs 362,000 (graded 0.8), activation 29% (0.8), store conversion 30% (0.5), cost per activated learner £5.70 settled on 21 October (0.85). The north star reached 67,500. A mid-quarter creative test had pushed installs up with a cheaper network, but activation for those users was 11%, so the paid team cut it in week 8 because KR2 and KR4 made the cost visible.
Each decision showed up in the result. The value-moment north star meant the leadership install number was treated as an input, not the goal. Pairing installs with activation caught the cheap network. Separate store and paid key results let the store team's partial grade be read on its own, rather than being blurred by paid volume, and the settlement date stopped a debate over provisional attribution numbers.
Measuring app-install OKRs: KPIs and cadence
The north star is read weekly, since it is defined weekly. Installs and store page conversion can be read daily or weekly from the store consoles. Activation needs its window to close, so read it weekly by install cohort, never for the most recent days. Cost per activated user is best read weekly as a provisional figure and graded monthly or at quarter end from settled data. Retention of activated users, the input after activation, is monthly. Every KPIA metric tied directly to an objective. in the set should name its source and whether it is modelled. The CPA calculator helps check cost per activated user from spend, installs and activation rate.
Where the framework misleads for app installs
A value-moment north star still reacts slowly to acquisition, so a team can grow installs well and see little movement at the top in the same quarter. Activation events can be gamed, for example by prompting a trivial version of the action. And modelled attribution can shift after the fact, so a key result graded on early data may change once the data settles; publishing provisional grades invites confusion.
Common mistakes
- Installs as the north star. It counts downloads, not value, and rewards cheap volume.
- An install key result on its own. Without activation beside it, low-quality installs look like success.
- One key result for store and paid together. Neither team can see its own contribution.
- Grading on provisional attribution data. Decide the settled reading before the quarter starts.
- Counting late-quarter installs in activation. Their window has not closed, so the rate is understated.
