Parallel-campaign reset

The force-spend tactic Tierra uses when the main campaign is locked onto a dominant creative and won't allocate spend to new assets. Clone the main campaign as a small secondary and force-feed it the assets the main campaign starves. The goal is retraining the algorithm's spend distribution, not chasing short-term ROAS. Read alongside the playbook and media buying.

This is the override to the default no-pause rule. When the question is "we have new creatives AppLovin won't spend on," the parallel-campaign reset is the answer before you pause anything in the main campaign.

A worked case

One account was running on AppLovin with the algorithm locked onto a "December creative set" that had been the dominant winner for 3+ months. Meta was generating fresh winners, but importing them produced "struggles to find an audience": the algorithm wouldn't allocate enough spend to the new creatives to learn them.

Decision and launch, from an internal meeting:

"A cloned secondary campaign was launched to force spend toward high-performing Meta creatives, since AppLovin continues favoring the December creative set. The goal is not immediate ROAS lift, but to retrain the main campaign."

The framing was explicit: short-term ROAS is not the goal. The cloned campaign exists to redistribute the algorithm's attention by giving new assets enough impressions to develop their own spend signal.

Quantified outcome, from a follow-up internal meeting:

"The side campaign successfully reduced concentration on the previously dominant ad set (top creative dropped from ~50% to ~25% of spend). D0 ROAS was reported as the second-highest this month."

So:

Top-creative spend share was cut in half (about 50% to about 25%).
D0 ROAS was the second-highest of the month, so the tactic produced real performance lift in the same window, not just diagnostic value.

The same doc flagged a risk: the account can revert to the December creative set. The learning isn't permanent. Without sustained pressure (continuing to feed the secondary campaign new assets, or eventually consolidating the learnings back to main), it can slip back.

When to deploy

Any of:

A single creative set is taking more than 50% of daily spend and the team has new creatives the main campaign won't spend on.
A Meta winner (or any externally validated winner) has been ported in and is starving for impressions in the main campaign.
Diagnosed stuck delivery: better creatives sit idle while a dud concept keeps spending (see algorithm mechanics).
Account-age effect: older accounts, and accounts that have reduced spend, fixate on low-performing mature customers instead of high-performing new customers (found in back-end analysis on one account). A parallel campaign with new-audience creative counters this.

When not to deploy

The main campaign is in the learning phase (under 15 conversions/day, or in the first 7 to 10 days of a new spend tier). Wait for it to stabilize before splitting attention.
The starved creatives haven't passed Tierra's own quality bar. A parallel campaign feeding bad creatives is just a worse version of bad creatives running. The tactic works when the constraint is platform allocation, not creative quality.
You'd be cloning a campaign whose learnings are themselves the problem. That's a full restart, not a parallel reset, and it's expensive; AppLovin's rough rule of thumb is on the order of $20k of spend to recover, though that scales with the account's daily spend. For a true restart, upgrade the main campaign in place instead (Universal to Prospecting on the same campaign ID is about 24 hours of volatility, not a full relearning cost).

How to size the secondary

The worked case used a small clone (the exact size wasn't in the notes; the typical pattern is about 10% of the main daily budget). The point is enough to learn the new assets, not enough to compete with main for the same audience pool.

Other anchors from Tierra's clients:

A launch-to-measure incrementality test on Discovery alongside the main ran at under $1k/day.
A split on one account: $9k to main, $1k to a new Discovery campaign, the same roughly 10% principle.
A landing-page-test variant: clone the top 5 creative sets, swap the landing page, run at $1k/day for 10 days. Same architecture, different goal (testing the page), which confirms the pattern.

If the account's daily envelope is too small to support a meaningful clone (under about $2k/day total), the parallel-campaign tactic probably isn't the right tool yet. Get the main to scale before splitting.

What it's not

Not pausing the dominant creative. The whole point is leaving main untouched while introducing parallel pressure. Pausing main means sequence death and worse outcomes (see the pause doctrine in the playbook).
Not a campaign-type switch. This creates a new campaign with cloned structure, it doesn't change the existing campaign's settings. To switch campaign type, upgrade in place.
Not a per-product restructure. One account's restructure split into 5 campaigns that each isolated a single product. That's a different intervention: it solves the "products other than the flagship can't break through" problem via campaign separation. The parallel-campaign reset works at the asset level; the restructure works at the product or concept level. Both force the system out through structure, but the unit of separation differs (see the worked restructure in strategy).

algorithm mechanics: why this works (when the model gets stuck, the force-via-structure lever).
strategy: campaign-structure choice (Discovery, Prospecting, per-product architecture).
dev: programmatic clone via the Campaign Management API (the status field is silently ignored; audience strategy is silently set to Universal).