The next paid dollar should go to whichever job produces more incremental contribution inside a window you can actually observe — not to whichever campaign reports the lower cost per action. For most app teams with a small dormant base, no working deep links, or an unreliable in-app event, that dollar belongs to net-new acquisition, because paid re-engagement has no qualified audience to buy. Once a behaviour-defined dormant cohort exists, has a real reason to return, and can be measured against a holdout, re-engagement becomes a genuine competitor for budget.
Both campaign families exist in the major platforms. Google separates App campaigns for installs from App campaigns for engagement, which target people who already have the app installed (Google Ads Help). TikTok's App Promotion objective likewise splits App Install from App Retargeting (TikTok Ads Help Center). Meta trains advertisers on app campaigns that both acquire new customers and re-engage existing app users, with the signal and measurement setup those require (Meta Blueprint). The platforms will happily run both. The allocation decision is yours.
What are the two jobs, precisely?
Acquisition buys a net-new user: someone who does not have the app, has to see a store listing, decide, download, open, onboard, and only then reach a value moment. The paid job ends at the install; the economic job ends much later, at a retained and monetising user.
Re-engagement buys attention from someone who already installed. The user has an app icon, a possible account, some history, and — critically — a state: dormant, lapsed, subscribed-but-inactive, cart-abandoned, trial-expired, onboarding-incomplete. The paid job is to interrupt that state with a credible reason to return and to land the tap on the exact screen where the reason pays off.
Platform definitions of the second job differ, and that matters operationally. Google's engagement campaigns require an audience segment of existing users plus deep links, and Google documents both requirements explicitly (Google Ads Help — engagement campaign setup, deep links for install and engagement campaigns). TikTok's App Retargeting is a distinct product from its acquisition-side optimisation models, and TikTok publishes the differences between App Retargeting and its app-event optimisation and value-based options (TikTok Ads Help Center). Privacy constraints also apply asymmetrically: matching a known installed user to an ad account depends on identity, consent and platform eligibility in ways that a cold acquisition auction does not.
A comparison matrix for the allocation decision
Dimension Paid acquisition (install) Paid re-engagement (retargeting) --- --- --- Eligible audience Effectively the addressable market minus current installers Only your installed base, minus non-matchable and non-eligible users User state No product knowledge, no account, no history Known state: dormant, incomplete action, lapsed payer, inactive subscriber Destination Store listing, then cold open and onboarding A specific in-app screen via a deep link (Google Ads Help; Apple universal links; Android App Links) Optimisation signal Install, or a post-install event the SDK reports (FBSDKAppEvents) An in-app return action the platform supports as an event (TikTok supported in-app events) Creative job Explain the product and earn a download decision Remind, restate a specific value moment, remove one blocker Frequency risk Wasted reach on uninterested cold users Fast fatigue and annoyance inside a finite, repeatedly-hit list Measurement Store analytics plus MMP plus platform reporting (Apple App Analytics) First-party cohort return rate, plus platform reporting, plus a holdout Incrementality risk Moderate — brand and organic-search overlap High — many targeted users would have returned anyway Scale ceiling Large; bounded by market and payback Hard-capped by dormant cohort size and match rate Disqualified when Retention after install is near zero, or store conversion is broken No behaviour-defined cohort, no reason to return, or deep links fail
Read the last two rows first. Re-engagement has the smaller ceiling and the higher incrementality risk. That combination is why it so often looks efficient in a dashboard and invisible in the business.
Is your app ready for paid re-engagement? A six-gate test
Run these gates before moving a single dollar. Failing any one of them means the money is better spent on acquisition, on the product, or not at all this month.
Gate 1 — A behaviour-defined cohort. "Everyone who installed" is not a cohort. You need a state defined by behaviour and time: no session in 30 days, started checkout without purchasing, finished onboarding but never reached the core action, trial ended without conversion. Google requires you to build audience segments of app users for engagement campaigns and documents how to create them (Google Ads Help).
Gate 2 — A credible reason to return. The message needs a specific value moment: new content the user asked for, a restored session, a resolved blocker, a genuinely relevant offer. "We miss you" is not a reason; it is a reminder that the app exists, which push notifications already deliver for free.
Gate 3 — Working deep links and post-login continuation. The tap must open the exact screen, and it must survive authentication. Google publishes both the technical requirement and an effective deep-link strategy for engagement campaigns (Google Ads Help); Apple and Google document the platform mechanics through associated domains and App Links (Apple, Android). A link that dumps the user on a home feed, or into a login wall that forgets the destination, converts a paid tap into a bounce.
Gate 4 — Reliable events and a populated audience. The return action must be logged consistently by the SDK and supported by the platform as an optimisation event (TikTok). The audience must also actually populate — Google publishes troubleshooting guidance for segments that fail to reach usable size (Google Ads Help).
Gate 5 — First-party cohort economics and lawful use. You need to know what a returned user is worth over a defined window, at variable margin, from your own data. You also need to confirm that using customer or behavioural data for ad targeting fits your consent basis, privacy policy and platform terms.
Gate 6 — A credible incremental test. A holdout of matched dormant users who are deliberately not exposed, or another design that isolates the media effect. Without this, gate 5's numbers will be inflated by users who were coming back regardless.
How should the next dollar be allocated?
Compare expected incremental contribution per dollar, not attributed conversions. State every input openly:
Eligible audience (E): users who can actually be matched and served in the chosen cohort.
Reachable exposure (R): the share of E your budget can reach at a sane frequency.
Incremental rate (I): exposed-group action rate minus holdout action rate — for re-engagement, the return-and-act rate; for acquisition, the install-to-value-action rate against the best available counterfactual.
Downstream value (V): paid conversion or revenue per incremental user inside the window.
Variable margin (m): the share of V that survives payment fees, delivery, support and platform commission.
Media cost (C) and observation window (W): identical for both programmes, or the comparison is meaningless.
Expected incremental contribution per dollar = (E × R × I × V × m) ÷ C, measured over W. Acquisition usually has a large E and a low I·V early. Re-engagement usually has a small E and a higher near-term I·V. Which wins depends entirely on your own numbers — there is no universal answer, and any source claiming one is guessing about your app.
A labelled hypothetical
This is an illustrative arithmetic example with invented figures. It is not a benchmark, a prediction, or a Sharply Labs client result.
An app has 60,000 dormant users with no session in 30 days. Match and eligibility leave E = 30,000. A $6,000 budget reaches R = 50% at acceptable frequency, so 15,000 exposed. The exposed cohort returns and completes the core action at 8%; the matched holdout returns at 5.5%. Incremental rate I = 2.5%, giving 375 incremental returns, at $16 each. Downstream 60-day value per incremental returned user is $9 at 70% variable margin, or $6.30. Incremental contribution = $2,363 against $6,000 spent — a loss on that window, despite a $16 "cost per reactivation" that would look excellent in any dashboard.
The same $6,000 in acquisition buys 3,000 installs at $2. Store and onboarding losses leave 900 users reaching the core action; suppose 800 of those are judged incremental against organic baselines. Sixty-day value per incremental activated user is $11 at the same margin, or $7.70, giving $6,160 — roughly break-even on the window and improving if 90-day retention holds.
Note what changed the answer: not the headline cost per action, but E, I and the observation window. Change dormancy to 14 days and the holdout return rate rises, shrinking I further. Extend W to 180 days and acquisition improves if retention does. The arithmetic is only as honest as the counterfactual behind I.
A controlled workflow for testing both
Define lifecycle states in one place. Active, at-risk, dormant, lapsed payer, incomplete-onboarding. One definition, shared by paid media, CRM and analytics.
Suppress overlap. Exclude installed users from acquisition targeting and exclude recently-active users from re-engagement, or the two programmes will bid against each other and both will report the same returns.
Choose one meaningful event per programme. Not "open" — an action that correlates with retained value in your own cohort data.
Validate deep links before spending. Test every route, on both platforms, logged in and logged out, cold start and warm start.
Match observation windows. Same W for both, chosen from your payback reality rather than from reporting convenience.
Hold out where feasible. A randomised untreated slice of the eligible cohort is the cheapest honesty you can buy.
Reconcile three data sources. Platform reporting, MMP, and first-party analytics will disagree; decide in advance which one governs the decision, and use the store-side view for acquisition funnel questions (Apple App Analytics).
Judge on post-return retention or payback, not on the return itself. A user who reopens once and lapses again the same week has cost you money.
Why a cheap reactivation can be worthless — and a cheap install can be too
A low cost per reactivation mostly measures how easy the users were to reach, not how much behaviour you changed. Dormant users on the verge of returning are the cheapest to convert and the least incremental; a well-optimised campaign will find them first, because the algorithm is rewarded for cheap conversions, not for changed minds. This is the core reason a re-engagement programme can post a falling CPA quarter after quarter while first-party monthly actives stay flat.
The mirror-image error occurs in acquisition. A falling CPI often signals cheaper, lower-intent inventory, broader placements, or a shift toward users who install and never activate. If you optimise toward install and your post-install event quality is poor, the platform will efficiently deliver exactly the outcome you asked for. Meta's own training pushes teams toward app events for targeting, optimisation and measurement precisely because the install alone is a weak instruction (Meta Blueprint). Our guide to Meta post-install events covers that setup in depth, and the Google app campaign goal guide covers the equivalent decision on Google.
Diagnostic matrix
Symptom Most likely cause First move --- --- --- Re-engagement spends but returned users don't stick The reason to return is generic; the user gets a reminder, not a value moment Rebuild the offer around one specific in-app moment; measure 14-day post-return retention, not returns The audience won't populate Cohort too narrow, match or eligibility limits, event not firing Widen the dormancy window, verify event delivery, and work through the platform's audience troubleshooting (Google Ads Help) Clicks open the wrong screen Deep-link routing or post-login continuation is broken Re-test associated domains / App Links and the authenticated route (Apple, Android) Platform reports wins, first-party lift is absent Attribution is claiming organic returns Run a holdout; treat platform numbers as a delivery report, not a verdict Acquisition scale stalls at target cost Ceiling on audience, creative, or store conversion — not on budget Diagnose which, using creative testing and store-page conversion as separate levers Both look efficient, blended economics worsen Overlap, double-counting, or unfunded downstream costs Suppress overlap, reconcile sources, and re-check margin (attribution architecture)
When to fund acquisition first
Fund acquisition when the installed base is small relative to your addressable market, when your dormant cohort cannot fill a campaign at a sane frequency, when retention among recently-acquired users is acceptable, and when the store funnel converts. Early-stage apps are almost always in this state: there is simply not enough dormancy to buy. Choosing among acquisition channels is a separate decision, covered in the paid channel stack guide.
When to fund re-engagement first
Fund re-engagement when a large, behaviour-defined cohort exists, when there is a concrete new reason to return (restored feature, released content, resolved blocker, expired trial with a real offer), when deep links are verified, and when a holdout is possible. It is also the right first call when acquisition is already at or beyond its efficient ceiling and adding budget only raises cost per activated user.
When sequential testing is the responsible answer
If you cannot fund both at a readable size, run them sequentially rather than splitting a budget too thin to interpret. Two underpowered tests produce two unusable results. Pick the programme whose gates you pass most confidently, run it for a full payback window with a holdout, then use its result to size the second test.
When neither should be scaled
Neither campaign deserves more budget when day-30 retention is near zero, when the core action is rarely reached, when the store page cannot convert qualified traffic, or when you have no first-party view of what a user is worth. Paid media accelerates whatever the product does to users. If the product loses them, both programmes are buying churn at different prices, and the honest move is to fix activation first.
Questions app teams actually ask
Is re-engagement cheaper than acquisition? Cost per action is usually lower, because the audience already knows you. Cost per incremental action often is not, because a meaningful share of the cohort would have returned unprompted. Compare the two on incremental contribution or you will systematically over-fund re-engagement.
How large does the installed base need to be? Large enough that a behaviour-defined dormant cohort survives matching and eligibility and still fills a campaign at reasonable frequency and a readable test size. That is a function of your dormancy definition and match rate, not a universal user count.
Should a small-budget app run both? Usually not simultaneously. Run acquisition, build the base, and revisit re-engagement when the dormant cohort is real and the deep links are proven.
What is the right reactivation event? The earliest in-app action that predicts retained value in your own cohorts, and that the platform supports as an event (TikTok in-app events). Optimising to "app open" is optimising to the cheapest possible tap.
Do push and email make paid re-engagement unnecessary? They make it narrower. Owned channels reach users who still receive your messages; paid re-engagement is for users who have opted out, disabled notifications, or stopped opening them. Fund the owned channels first — they are cheaper — then test paid against the residual.
How should app re-engagement be measured? Exposed-versus-holdout return and retention rates in your own data, reconciled against platform and MMP reporting, over a window that matches payback.
Limits and tradeoffs you should state out loud
Identity matching is imperfect, so your eligible audience is always smaller than your installed base. Platform eligibility rules for retargeting products change and differ by market (TikTok App Retargeting). Consent and privacy constraints may exclude some users from targeting entirely. Selection bias is the default failure mode: the users easiest to reach are the ones least in need of persuasion. Small cohorts produce noisy tests that tempt teams into reading randomness as signal. Auction dynamics differ between install and engagement inventory, so costs are not directly comparable. And attribution is not incrementality — a platform reporting a conversion is telling you it showed an ad before an action, not that the ad caused it.
None of this makes re-engagement a bad investment. It makes it an investment that has to be earned with cohort definition, deep-link discipline and a holdout, rather than assumed from a low reported CPA.
Get the allocation decision audited
For app teams deciding how to split paid budget between net-new acquisition and re-engagement, Sharply Labs will examine your lifecycle definitions, audience eligibility and match reality, deep-link paths, event quality, holdout design, and first-party cohort economics. You receive a prioritised test and measurement plan showing which job should get the next dollar, how to test it, and what evidence would change the answer. No CPI, CPA, ROAS, lift, or growth outcome is promised — the deliverable is a plan and a measurement design, not a performance guarantee.
See how we work with app teams on mobile app growth and across growth services.