Skip to content

Ad Platforms ยท AppLovin

Last Updated: August 11, 2026

Strategy

Campaign architecture, audience strategy, scaling, and the relaunch-over-recover doctrine for AppLovin. Read alongside the playbook for the platform model and vocabulary. Production specs and compliance live in creative; budget mechanics, pausing, timing, and spend cuts in media buying; KPIs and attribution in reporting and tracking; the API and automation in dev. Per-client specifics belong in that client's project, not here.

Audience strategy

Audience strategy is a campaign-level setting on AppLovin. You pick one of three per campaign, and it shapes who the model optimizes toward. Treat the three as different strategies suited to different situations, not a fixed ladder. The right choice depends on the account's stage, how much weekly conversion data it has, the vertical, and how long the customer's journey runs. Where the read isn't clear, default to the sequence in "Launching a new account" below.

  • Universal: optimize for purchases from all customers, new and returning.
  • Prospecting: optimize for customers who haven't bought before.
  • Discovery: reach users who've never visited the site. It filters out anyone who already carries an AppLovin cookie, so it's a clean surface for genuinely new visitors.

How Tierra uses each:

  • Prospecting is the primary scaling vehicle on most accounts. It's where net-new-customer spend lives once an account is established.
  • Discovery runs as a small sidecar, roughly 5 to 10% of the main campaign. It isn't a scaling lever; it's an audience-freshness insurance policy and a way to read incrementality. Skip it on quick-purchase funnels where the impression sequence carries little value, or on accounts that already run more than about 70% new-customer traffic.
  • Universal works as a broad seed at the very start. The failure case is running Universal permanently alongside other campaigns, where it drifts toward roughly 85% retargeting and cannibalizes the prospecting campaign. Reserve it for the seed, not for a standing parallel campaign.

Launching a new account

Start a new account on a single Universal seed campaign. Don't fragment into five campaigns and funnels on day one; that's the classic new-account mistake. With no site visitors yet, Universal effectively is discovery, so it's the right way to gather the first learnings. Starting a brand-new account directly on Discovery or Prospecting is not the move.

Run the one Universal campaign until it reaches a baseline of a couple thousand dollars a day. Then, rather than discarding what it learned, transition that same campaign in place to Discovery, which keeps its learnings, and stand up a Prospecting campaign alongside it as the primary vehicle for net-new-customer scale. Universal is right for the seed only; the standing-campaign failure case is covered above.

Hold this sequence as a strategy to fit to the account, not a hard rule. Weekly conversion volume, the vertical, and the length of the customer journey all shape which audience setup fits. Where the read isn't clear, default to this Universal-seed, then Discovery-plus-Prospecting sequence.

Multi-campaign architecture

Running multiple campaigns is an option for larger accounts, and it becomes more of the norm the bigger an account gets. Per AppLovin (CEO, 2026-03-25): the biggest accounts run multiple campaigns, and a campaign is best thought of as a "folder" for assets. Treat older "one campaign per account" rep guidance as superseded.

A split is only worth it when each campaign chases a genuinely different audience. Working dimensions:

  • Age or identity, for example a 30-year-old versus a 60-year-old buyer.
  • Use case, for example solution-aware versus problem-aware framing.
  • Post-click funnel, meaning a distinct landing page tied to that audience and its creative.

Same product, same funnel, same offer, just different creative talent is not a legitimate split. AppLovin won't invent different audiences for you.

Don't duplicate creative across campaigns. Sharing creative defeats the whole point: delivery follows the creative's learning across both campaigns and collapses the audiences back together. Each campaign's roster should reflect the audience it chases.

A typical established account, roughly $5k a day and up, runs one main Prospecting campaign plus a small Discovery campaign alongside it, at about 90% and 10% of budget. That's the baseline, not a ceiling. As the account grows, the main campaign is what splits into several audience-differentiated campaigns, with the small Discovery campaign still running alongside them.

When spend concentrates too far, for example more than half on a single product or creative set while everything else sits under 1% each, break it up with a restructure:

  • Split into separate campaigns, each isolating one product, audience, or hook.
  • Give them equal daily budgets and broad geo.

Isolating each factor forces delivery to find the audience for it on its own, instead of locking the whole account onto one.

A worked example: one multi-product account had scaled spend about 8x faster than its creative could keep up, and delivery had locked onto the flagship product. The top creative set held about 38% of spend, the top five about 65%, and 106 of 120 sets got under 1% each. On top of that, 92% of spend sat on the flagship, whose ad-rejection rate ran about 44% against roughly 5% for the other products, so the account's biggest exposure was also its hardest product to keep approved. Within-campaign budget shifts didn't move it. Splitting into five campaigns, one per product at equal daily budgets and broad geo, forced delivery to find each product's audience on its own instead of serving flagship buyers stale creative, and it isolated the compliance risk so a flagship rejection spike no longer starved the whole account.

Some clients go further and stand up separate sub-brand accounts, not just campaigns, when the audiences are genuinely distinct. This is the heaviest form of separation; reserve it for real sub-brands, not for splitting one brand's audiences.

Creative-set composition

Build each creative set so you can read what wins. Keep one element the same across every asset in the set (the hook, the concept, or the offer) and let the others vary, or keep the whole set to a single concept. Then when a set starts capturing spend, you know which element drove it. A set that mixes several concepts together gives you spend you can't attribute to anything.

AppLovin's reference sizing for a set:

  • Around 4 videos, as AppLovin's baseline; our rosters often run more in practice.
  • No more than 10 interactives, usually 5 to 8.
  • About 25 to 30 sets per campaign.

Those numbers keep the combinations learnable. Four videos by 8 interactives is 32 nodes; a 10-by-10 set is 100 nodes, too many for the model to learn from.

AppLovin claims no delivery difference between single-asset and multi-asset sets. Tierra's default at launch is single-asset sets, one video by one interactive exposing exactly one node, because attribution reads cleaner. Switch to multi-asset sets for iterations once the concept is validated.

Creative volume

Creative drives the large majority of performance on AppLovin, on the order of 90%, so it's where most of the effort should go, well ahead of tuning campaign settings.

Build a large roster of creative to test; a broad roster out-tests careful curation. How you release it depends on the assets. When every asset is a proven, distinct top performer from elsewhere, releasing a large batch at once works, and AppLovin now handles high volume well; a big launch of vetted assets into a fresh campaign can even improve from day one (see the mass-injection rescue in algorithm mechanics). When the assets are unproven, drip them in smaller daily batches, because a large single-day drop of untested creative (around 20 or more videos) causes ROAS dips while delivery rebalances. In short, batch what's proven and drip what's not, and when in doubt lean toward launching vetted assets together rather than trickling them.

A testing framework

Run tests as a loop: form a hypothesis, write a short brief, ship the variant, then read the result by factor rather than by ad. Break each result into its factors (hook, concept, spokesperson, offer) and keep a running list of losing factor-pairs so you stop repeating them. Don't hard-kill a factor on a single loss, though. A factor that underperforms across several accounts or audiences gets deprioritized, not banned, because the platform shifts and revives things. Keep the factor data somewhere you can pivot and slice, not baked into filenames, and keep set names short (a couple of tokens) rather than encoding everything into the name. Treat this as a recommended approach to adopt and adjust, not a fixed system.

Scaling

Scaling and creative volume go together. As you scale spend, scale creative output with it; if launches dry up while spend holds, performance drops off soon after.

Scaling too hard too fast is one of the most common ways to break an account. When in doubt, step up gradually and wait for a few days of settled data to confirm each increase before making the next. The specific cadence (how big each increase is, how many per day, and when to make them) lives in media buying.

Relaunch over recover

When a campaign degrades, the default move is to reset delivery rather than fight the algorithm in place. A relaunch means starting fresh delivery, through a parallel campaign or a rebuilt structure, so the model explores again from a clean state instead of staying stuck on the same narrow set of winners. Tuning inside the broken campaign rarely brings it back.

The parallel-campaign reset, for when the main campaign locks onto incumbent winners and refuses to test new creative:

  1. Clone the new creatives into a second campaign that runs with fewer learnings. The acknowledged downside is that out-of-gate performance suffers.
  2. Those new creatives now get room to spend and explore fresh segments.
  3. At the same time, clone the same sets back into the main campaign, so once they earn segment learnings in the side campaign the main can pick them up.
  4. End state: the new creatives are spending in the main campaign, and the side campaign is retired or repurposed.

Force-spend tactics, for when the main campaign won't allocate spend to assets you need to test:

  • A parallel small-budget campaign at $350 to $500 a day per product. Same cloned structure, new creative, same audience. The goal is to force delivery, not to scale.
  • A warm-up campaign, 10 to 20% of total budget reserved for underperforming or new assets so each one earns a spend baseline before you judge it. Useful during scale-up, when winners otherwise absorb everything.
  • A no-spend high-click bundle: group assets that got minimal spend but strong CTR into one set. The set often takes off; if it becomes a top spender within ~3 days, the account needed a creative refresh.

Cloning has a cost. A cloned campaign resets its learning and, on a typical account, needs on the order of $20k in spend before it performs like the original. Treat that as a rough benchmark, not a fixed figure, and scale it to the account. Budget for it.

Upgrade before you spin up new. Changing creative or targeting inside the existing campaign keeps its learning and costs about 24 hours of volatility. Spinning up new (a clone or a fresh build) resets learning and pays that recovery cost. Reserve it for when you genuinely need an isolated experiment, or the existing campaign is too locked in to fix in place.

Assets that haven't spent in the account for 2 to 3 months can be treated as new for a relaunch. Relaunching a batch of dormant winners at once is another way to force the algorithm to test them again.

This next part is about relaunching proven assets, separate from the daily drip of net-new creative that keeps running the whole time. Once an account has scaled, rebuild relaunch sets about once a month (twice at most): take the top 3 to 5 spending videos and pair them with the top 4 to 8 interactives by CTR over the trailing 7 days, in fresh sets. See relaunch playbook for the naming grid and the full tactic decision tree.

Recovery

Recover by adding, not cutting. When performance dips, the first move is an influx of new creative plus landing-page conversion work, not pulling spend back. Cutting daily spend is a last resort, covered in media buying; pausing is covered there too. Cross-reference those rather than second-guessing them here.

Diversification gates

When an account is concentrated and you're testing diversification, set account-health guardrails so an unhealthy trade fails fast. Standard pattern, scaled to account size:

  • D0 new-customer CPA within 130% of baseline over 5 consecutive days. 130%, not tighter, because adding new-campaign spend inflates account CPA mechanically before the new campaign converts.
  • D0 ROAS at 80% or more of baseline over 5 consecutive days.
  • If either trips, pause the new campaign, restore the main budget, and diagnose.

Scale the percentages to the account. On a $1k-a-day account 130% leaves no room; on a $50k-a-day account it's too loose. The five-consecutive-days rule travels better than the exact numbers.

When to relaunch instead of recover

Symptoms that warrant a relaunch:

  1. The top creative set is more than half of account spend, and CPA is climbing or performance is otherwise falling off meaningfully. Either the set has run out of fresh audience to reach, or delivery has locked onto it and stopped exploring alternatives. This is a structural signal, not a verdict on the creative's quality, so don't pause it; clone alternatives next to it and let delivery find replacements.
  2. Ten or more days into a creative drought, where launches dropped well below the spend-matched rate.
  3. CPM doubling with no performance improvement, meaning delivery is over-retargeting into stale audiences.
  4. New launches going unspent for 7 or more days despite healthy launch volume, meaning delivery is refusing to test fresh inventory.
  5. Account-wide D0 new-customer CPA or D0 ROAS degraded past the diversification-gate thresholds for 5 or more consecutive days.

Coming from Meta

AppLovin shares patterns with Meta but rewards different instincts. If you're coming from Meta, watch these:

  • Pausing: on Meta you cut low-ROAS creative fast. On AppLovin you don't pause; the algorithm starves weak ads on its own, and warm-up creatives still build the purchase sequences your closers convert against, so leave them running.
  • Concentration: on Meta, letting the algorithm consolidate spend onto a few winners is normal. On AppLovin, more than 50% of spend on one set is a structural problem, not a sign of a great creative.
  • Targeting: on Meta you still feed the system audiences and signals. On AppLovin geography is the only lever you set, and the creative itself is the targeting.

Both platforms reward testing plenty of creative, so keep the volume up here too. On importing Meta winners: some port directly, like UGC and mashup formats, and others struggle, so treat a Meta win as a leading indicator, not a guarantee. The budget-level, geo-only, and no-pausing mechanics live in the playbook and media buying.