Relaunch playbook

How to plan a monthly relaunch cycle, force spend onto assets the main campaign is starving, and bring dormant winners back. Read alongside the playbook, creative, and media buying.

The monthly relaunch

Once an account has scaled, rebuild a batch of fresh creative sets from recent winners about once a month, twice at most, on a steady day you hold to. Pull the trailing 7 days from AppLovin's asset-level reporting and take the top 3 to 5 spending videos and the top 4 to 8 interactives by CTR. Pair them across a grid of video-and-interactive combinations rather than matching each interactive to "its" video. Videos and interactives often push different pain points, and in practice the mismatched pairings tend to beat the matched ones.

Relaunching does two things: it keeps proven winners in rotation before they go stale, and it spreads spend back across assets, which fights the concentration that builds up when the model gets stuck on a narrow set of winners.

A freshly relaunched set usually starts softer than the incumbents and improves over the following weeks, so don't judge it in the first few days.

Weekly iteration cadence

The monthly relaunch keeps proven winners in rotation; a weekly iteration rhythm layered on top keeps the creative itself moving. Adopt it as a recommended cadence and tune the fixed day and the batch size to the account.

Each week, on a fixed day you hold to, pull the trailing 7 days and pick the top-spending assets, counting an interactive (end card) as half its spend for this selection so the videos don't get crowded out. Build variations off those winners across the week, changing one variable at a time so you can read afterward which change moved the result. Launch the next week's batch on the same fixed day, then repeat. The fixed day matters as much as the picks: a steady cadence is what makes the week-over-week reads comparable.

Naming the relaunch sets

Name each set in the batch so you can read how it performed afterward. Encode what went into it: which videos, which interactives, and how many times each has been used before (its "use-count").

One account's convention, which ran about 10% above that account's average, pairs videos and interactives across a use-count grid. It starts with high-use videos against low-use interactives, then steps down the video use-count and up the interactive use-count row by row, and finishes with a set built from brand-new videos. Its set names looked like this:

<PREFIX>_R_top_spend_7_use_vids_top_spend_1_use_end_cards
<PREFIX>_R_top_spend_6_use_vids_top_spend_2_use_end_cards
<PREFIX>_R_top_spend_5_use_vids_top_spend_3_use_end_cards
<PREFIX>_R_top_spend_4_use_vids_top_spend_4_use_end_cards
<PREFIX>_R_top_spend_3_use_vids_top_spend_5_use_end_cards
<PREFIX>_R_top_spend_2_use_vids_top_spend_6_use_end_cards
<PREFIX>_R_top_spend_1_use_vids_top_spend_7_use_end_cards
<PREFIX>_R_NEW_vids_top_spend_4_use_end_cards

Read the names literally. top_spend_7_use_vids means the top-spending videos that have been used seven times, paired with top_spend_1_use_end_cards, the top-spending interactives used once. <PREFIX> is a client or product code, _R_ marks it as a relaunch, and end_cards is a legacy token that refers to the interactive (kept for naming continuity).

This exact grid is one account's scheme, not a universal rule. Adapt the naming to the account. The point that carries across accounts is that every set's composition is legible from its name, so you can tell afterward which combinations worked.

Bringing back dormant assets

An asset that hasn't spent in the account for two or more months can be relaunched as if it were new (Tierra's stricter rule of thumb is three months). Per AppLovin, anything two or three months old, or unspent that long, is effectively a new asset to the algorithm.

The bulk version: pull every top-spend creative that's gone dormant past that window and relaunch them all at once. It's a fresh creative drop without producing anything new, and it's a fast way to break up concentration (one account used it when a single product had taken 92% of spend). When an account is genuinely broken or stuck rather than just concentrated, a heavier version of relaunching applies: launch a large batch of fresh ads, on the order of 100, into a single new campaign, which on AppLovin tends to perform from day one (see the mass-injection rescue in algorithm mechanics).

Interactives stay fresh longer than videos. On one account, two hero-format interactives were still top-3 by spend across about three months, so treat interactives as the longer-lived asset class when planning relaunches.

You don't always have to wait out the dormancy window. To make a proven winner read as fresh inventory sooner, apply a sub-perceptible change (for example a microsecond of added duration) and give it a refresh filename. It re-enters as new without the two-to-three-month wait.

Force-spend tactics

When the main campaign won't put spend on assets you need to test, force it. This comes up when new launches sit unspent for a week or more, when one creative set is over 50% of spend and CPA is climbing, or when the account is locked onto older incumbent winners and won't explore.

The parallel small-budget campaign is the default (AppLovin endorses it). The rest are situational, used when the context below fits:

Tactic	What it does	Use it when	Trade-offs
Parallel small-budget campaign (default)	A cloned campaign at $350 to $500 a day per product, 2 to 3 sets each, on a cleaner surface with fewer learnings. Once the new assets spend and prove out, clone the same sets back into the main campaign so the learnings flow back.	New creative, or an imported Meta winner, can't get spend in the main campaign.	Only feed it assets that pass your quality bar; a parallel campaign of bad creative is just bad creative running elsewhere. Retire it or repurpose it as a Discovery sidecar once the assets prove out.
Warm-up campaign	Reserves 10 to 20% of daily budget to push spend onto new or underperforming ads until each clears about $500, the threshold before you can judge an ad.	You're scaling and the winners are absorbing all the budget.	It's a standing exploration surface, not a scaling lever, so it runs indefinitely.
No-spend high-click bundle	Groups assets that got little spend but strong CTR into one set, which often takes off as a bundle.	You have high-CTR assets that never got enough spend to prove out.	If the bundle becomes a top spender within ~3 days, the account needs new creative production, not just relaunches.
URL-swap parallel	Points top creatives at a different URL (a sale page, alternate offer, or presell) in a small parallel campaign, holding the creative constant so the algorithm learns against fresh page-side signal.	Landing-page-level diversification is the lever you want to test.	Validated on one account so far.

Avatar-based relaunch and split

Worth testing as a relaunch pattern: build avatar-pure sets, each set carrying creative for a single avatar instead of mixing avatars in one set. Launch them together, then read a ROAS split by avatar around day 5 to 7. Any avatar running at or above the campaign-average ROAS graduates into its own campaign, where it gets dedicated budget and room to scale without competing against weaker avatars for spend. Avatars below the average stay in the shared set or get cut. The split is what makes the strong avatar's economics legible, since a mixed set hides which avatar is actually carrying the ROAS.

Landing-page A/B testing

To A/B a landing page, don't route traffic through rotator or redirect links. Per AppLovin, those disrupt algorithm learning, because the model can't attribute results to a stable destination. Run the test instead as several fresh creative sets built from the same assets, each set pointed at a different creative-set URL, so each page earns its own clean learning. Rollback is pause-only: there's no delete endpoint on AppLovin, so you pause the sets pointed at the losing page rather than removing them. This is a head-to-head page test and is distinct from the URL-swap parallel force-spend tactic above, which holds creative constant in a small parallel campaign to diversify page-side signal rather than to pick a winning page.

The cost of cloning

Cloning a campaign starts it from zero learning, so getting back to where you were is expensive. AppLovin's rule of thumb is roughly $20k of spend before a clone performs like the original, but treat that as a rough figure that scales with the account's daily spend, not a fixed cost. A small account reads out on far less; a high-spend account burns $20k in a day.

So prefer upgrading in place when you can: modifying creative or targeting on the existing campaign keeps its learning and costs about 24 hours of volatility. Clone only when you genuinely need isolation (an incrementality test, an audience experiment) or the existing campaign is too locked in to repair, and budget for the recovery period when you do.

Anti-patterns

Relaunching low-CTR, low-spend assets together. They never had signal, and bundling them doesn't create any.
Pairing a relaunch with a budget jump, or with a creative pause. Either one is compound disruption; do one at a time (see the scaling cadence in media buying).
Re-uploading a byte-identical file of a live asset. AppLovin has no similarity clustering, so only a truly byte-identical upload gets deduped back to the original; that's the one case that won't behave as a fresh relaunch. Any change at all, down to a single frame, reads as a brand-new asset with no association to the original, which is exactly why the sub-perceptible refresh trick above works. So don't re-upload the exact same file expecting a fresh start; change something first (see creative).
Running the same top-spend-by-top-CTR list every month. It goes stale, so mix in at least one big-swing iteration each cycle.

strategy: account-level diversification gates (the 130%/80% guardrails).
creative: iteration rules for reading as a new asset vs a duplicate.
media buying: daily budget mechanics and scaling cadence.
algorithm mechanics: the mass-injection rescue for a stuck or broken account.