Skip to content

Ad Platforms · AppLovin

Last Updated: August 11, 2026

Media buying

Managing budgets, scaling, pausing, recovery, and reading performance on AppLovin. Read alongside the playbook for the platform model and vocabulary. Campaign structure and audience strategy live in strategy; KPIs and attribution in reporting and tracking; the API and automation in dev.

How buying works on AppLovin

  • Budget is set at the campaign level, not on individual creative sets or ads.
  • The creative is the targeting. AppLovin has no audience or demographic targeting, and the one dial you set is geography (country, and US state). Beyond that, the algorithm decides who sees an ad based on who responds to it, so your creative mix is what steers the audience.
  • The algorithm spends by node (one video with one interactive at the set's URL), not by creative set. The same node can live in several sets, so pausing one set just sends that spend to the node in another set. Pausing a set usually doesn't do what it looks like it will.
  • You don't need to pause weak ads. The algorithm already starves them of spend on its own, so manual pausing usually adds nothing and can backfire (see Pausing). Spend that effort on new creative instead.
  • The campaign's goal (its ROAS or CPP target) mainly controls how freely the campaign spends, not how well it performs. Set it loose enough that the campaign spends; tightening it slows pacing but won't improve results. Per AppLovin, the goal target isn't a real performance lever.
  • Which goal type you optimize toward is worth testing per account. On at least one account, optimizing toward a ROAS goal beat a CPP goal (the Cost-Per-Purchase goal) sharply on the same account. Treat that as a per-account test, not a rule, and adjust to what the account shows. The goal-competition mechanics behind this live in algorithm mechanics.

Scaling

Scaling cadence is Tierra's guidance, not an AppLovin rule. AppLovin publishes no limit on how much or how often you change budget.

  • Early on, or while running well above target, big jumps are fine. Doubling the budget doesn't reset the algorithm's learning, so an aggressive early ramp is safe.
  • On a mature account near its target, move in steps of about 20% or less, no more than twice a day.
  • Scaling too hard too fast is one of the most common ways to break an account yourself. When in doubt, step up gradually and let a few days of matured data confirm the last jump before the next.

Ramp pattern: start around $1 to $2k a day for about a week to get a read and tune creative, then keep pushing toward a ceiling. Performance varies by day of week, so read a full week before judging a step rather than reacting to a single day.

An aggressive restructure buys a grace window. AppLovin will grant roughly two weeks where efficiency is expected to dip before it demands efficiency again. Set that expectation with the client before you start, so the early dip reads as planned rather than a problem.

Timing

  • Make budget changes at midnight UTC, the start of the account's day (7pm Eastern / 4pm Pacific in winter, 8pm Eastern / 5pm Pacific in summer).
  • A change later in the day crams the remaining budget into a few hours and can burn thousands in the last hour.
  • A mid-day increase also triggers an immediate charge.
  • The tracking day for URL parameters also rolls over at midnight UTC, so an ad launched right around that boundary can land on either side in tracking. Check that before chasing a mismatch between the dashboard and your tracking tool.

Billing

AppLovin is prepay (AppLovin's billing doc). At midnight UTC the card is charged to fund the next day's budget, minus any balance already on the account. A mid-day budget increase is charged immediately.

Keep a primary and a backup card on file. A failed charge can stall the campaign.

New accounts start with a daily card limit (often around $5k) that the rep can raise on request. Keep the total account budget at least $100 under that limit, so a full day's charge never hits the cap and locks the account. Check the limit before any budget jump that would cross it.

Separate from the card limit, the account has its own rep-gated cap on daily spend. It's a third ceiling, distinct from both the per-campaign budgets and the card's daily charge limit. AppLovin raises it on request when ROAS is strong, and will lift it materially after a stretch of strong ROAS. A billing or card cap set too low can auto-pause campaigns until the rep raises it, so check both the card limit and the account spend cap before a big step-up.

Pausing

Default: don't pause underperforming creatives. The algorithm already cuts spend to weak ads on its own, so pausing gains nothing. And removing an asset can kill the sequences it was part of, because the algorithm won't reliably replace it. Pausing low-spend losers is the move that actually hurts.

There's one case where pausing helps: a single creative set is spending more than roughly half of all account spend, is starving the other sets of budget, and is performing poorly. Pausing that one set forces the algorithm to spread spend back across healthier sets, and the account can recover fast. In one case, a set at about 77% of spend was paused and account ROAS went from 1.1 to 2.7 the same day.

Only use this for a runaway set. If you're trying to get new creatives to spend rather than break up a monopoly, use the parallel-campaign reset below.

More than half of spend on one set is a structural signal, not a sign the creative is great. The algorithm has settled into a rut or run out of new audience. Fix it structurally, not with more of the same creative.

Parallel-campaign reset

This is how you force the algorithm to test creatives the main campaign won't spend on, and how you split-test a landing page.

Clone the main campaign into a small second campaign at about 10% of the budget and feed it the new creatives. You're not chasing ROAS in the small campaign; you're giving the new assets enough impressions to build their own track record. Then clone those same sets back into the main campaign so it can pick them up once they've proven out. You're done when the new creatives are spending in the main campaign and the side campaign can be retired.

Reach for it when:

  • One set is over half of spend and you have new creatives the main campaign starves.
  • An imported Meta winner can't get impressions.
  • The account is stuck on a weak concept while better creative sits idle.

A small clone like this is cheap. Cloning the whole campaign to start over resets its learning and costs far more: a rough rule of thumb from AppLovin is on the order of $20k of spend to get back to where you were, though that scales with the account's daily spend rather than being a fixed number.

Landing-page tests use the same setup. Redirects, rotators, and same-URL split tests break the algorithm's sequencing and corrupt the pixel, so never split-test on a single URL, especially while an account is still learning. Instead, run a side campaign at about 10 to 15% of budget with your top 6 to 10 creative sets pointed at the test page for 7 to 10 days.

Full case study in parallel campaign tactic.

Seasonal and promotional creative

Run seasonal or promotional creative in its own campaign, separate from the evergreen one. A time-boxed push has a different goal and a short life, and leaving it in the always-on campaign distorts that campaign's learning. Keep the evergreen campaign clean so its optimization reflects steady-state demand, not the spike, and retire the promo campaign when the push ends.

How the algorithm behaves

Full detail in algorithm mechanics. The parts that change buying decisions:

  • The algorithm can get stuck spending on a mediocre concept while stronger creative sits idle. Adding budget or more copies of that concept won't fix it; the fix is structural (a parallel-campaign reset or a restructure).
  • An asset that hasn't run in two to three months can be treated as brand new again.
  • Creative sets get better over months. A set rarely peaks on day one; its ceiling rises as it ages, and the algorithm will move a large share of spend onto newer sets as they prove out. This is why patience beats early pausing.
  • Changing a setting in place (say, the audience strategy on the same campaign) costs about a day of volatility. Cloning to a brand-new campaign costs significantly more, because it resets the campaign's learning. Change in place unless the campaign's own history is the problem.
  • Audience saturation is real, and the rep can pull a frequency breakdown on request. If CPM jumps and CPA slips on a steady set, check frequency before blaming creative fatigue.

Learning phase and thresholds

  • Exit the learning phase at roughly 15 to 20 conversions a day, counted across the whole campaign, not per ad. Leave two to three days between big changes.
  • $500 a day is the launch floor for a CAMPAIGN, not for a creative set or an ad. Budget is only ever set at the campaign level (see "How buying works" above), so this floor attaches to the campaign — never write it as a per-creative-set or per-ad budget. It applies unless you're taking over an account that's already scaled. Any less and the campaign has too little to learn from.
  • Give a warm-up ad about $500 of spend before you judge it. Below that there isn't enough data.

How many campaigns the order volume supports

AppLovin needs roughly 5 new-customer orders a day per campaign to optimize. So the practical ceiling on campaign count is your total daily new-customer orders divided by about 5. Splitting a fixed budget across more per-product or per-audience campaigns than that starves every one of them below the floor, and none of them optimize.

Size the number of campaigns to the order volume first, then decide how to divide them. This is the sizing constraint on any multi-campaign structure, including the parallel-campaign reset and the diversification test below.

Diversification guardrails

When testing diversification on a concentrated account, watch these over five straight days:

  • D0 new-customer CPA within about 130% of the pre-test baseline. New spend inflates account CPA before the test converts, so 130% leaves room.
  • D0 ROAS at 80% or more of baseline.

If either breaks, pause the new campaign, restore the main budget, and diagnose. Scale the percentages to the account's size; the five-day rule holds better than the exact numbers.

Getting a stalled asset to spend

Gather assets that got little spend but strong CTR into a single creative set. Grouping them this way often gets the algorithm to start spending on them. This works inside the main campaign, unlike the parallel reset.

Reading performance metrics

Never read one metric on its own. Read CPM, CPC, and add-to-cart rate together.

  • CPM up and CTR up does not mean the creative is winning. A higher CPM means the platform is buying more premium inventory, where people click more readily, so the CTR lift may be about placement, not the creative connecting. Check CPC and add-to-cart for the truer read.
  • CPM down, CTR up, and no purchase attempts is the sign of a low-value-traffic pocket. It's rare but real; take the data to the rep. Per AppLovin: "in rare cases the model gets into pockets of low value traffic, but that shows up much differently in the platform: low CPMs, high CTRs, no purchase attempts."
  • Rank assets by the spend they capture, not their CTR. Where the algorithm puts spend is the real signal.

Recovery: new creative first, spend cuts last

When an account slips, the fix that actually works is more new creative plus landing-page conversion work (improving the page so more visitors buy).

Cutting daily spend is a last resort, not a first response. Pull back only after new creative and conversion work have failed, because:

  • An account that pulls back is often very hard to climb back up.
  • Cutting spend slows your testing.
  • It slows the new creative that would bring an elevated CPM back down.

Spend cuts fight the very things that fix an account.