---
title: "Half of Creative Decline Flags Are Country Mix"
description: "We split 2,890 fortnight comparisons on AppLovin creative sets into the part the country weights moved and the part the creative moved. Half the flags vanish."
canonical: "https://lemon-ai.com/resources/creative-decline-country-mix"
markdown_url: "https://lemon-ai.com/resources/creative-decline-country-mix.md"
language: "en"
image: "https://lemon-ai.com/og/resource-creative-decline-country-mix.png"
image_alt: "Creatives flagged for a rising cost per install, showing how many were still flagged once the country mix was held still, against a measured chance line."
date_published: "2026-08-30"
date_modified: "2026-08-30"
authors: ["Gregory Potemkin"]
schema_types: ["Article","BreadcrumbList","Organization","Person","WebApplication","WebPage","WebSite"]
---

# Half of Creative Decline Flags Are Country Mix

Creative decline diagnosis

We split 2,890 fortnight comparisons on AppLovin creative sets into the part the country weights moved and the part the creative moved. Half the flags vanish.

![Gregory Potemkin](https://lemon-ai.com/images/join-us/gregory.webp) 

By [**Gregory Potemkin**](https://lemon-ai.com/authors/gregory-potemkin)  
Founder & CEO  
Published August 30, 2026 

**A creative’s cost per install is a spend-weighted average across countries whose weights the network sets, not you. We split the change for 2,890 fortnight-over-fortnight comparisons on 1,349 AppLovin creative sets carrying $15.1 million of spend, over June 15 to August 9, 2026, using AppLovin’s own reported country dimension. Of the creatives whose blended cost per install rose 20% or more, only 51.0% still cleared that bar once the country mix was held still. The median flagged creative rose 33% blended and 21% inside its own countries. An independent Mintegral panel put the same survival rate at 50.9%.**

The flag is not wrong. It is reading a number that mixes two different events, and it cannot tell you which one it caught.

[ ![Share of AppLovin creative sets flagged for a rising blended cost per install that still cleared the same threshold once only the within-country part of the change was counted. At a 10 percent flag, 61.0 percent of 828 flagged comparisons survived against a measured chance line of 63.6 percent. At 20 percent, 51.0 percent of 469 survived against 60.0 percent. At 30 percent, 39.2 percent of 283 survived against 51.3 percent. At 50 percent, 22.0 percent of 118 survived against 46.2 percent.](https://lemon-ai.com/images/resources/creative-decline-country-mix.svg) ](https://lemon-ai.com/images/resources/creative-decline-country-mix.svg) 

The steeper the drop looked, the less of it the creative accounted for. The chance line barely moves.

## Every fatigue rule we found reads a number the network controls

The rules themselves are reasonable and specific. A current guide to measuring creative fatigue puts a refresh at a [“20 to 30% decline across two or more metrics”](https://segwise.ai/blog/how-to-measure-creative-fatigue), replacement at “CTR down 40% or more”, and treats a [“CPM increase above 18% over a two-week window”](https://segwise.ai/blog/how-to-measure-creative-fatigue) as meaningful fatigue, measured on a rolling seven days against a thirty-day baseline. It is careful about itself: “The numbers below are common industry norms, not platform guarantees, and you should tune them to your own data.” Published July 6, 2026, verified August 30, 2026.

Tuning does not help here, because the problem is not the threshold. It is the number the threshold reads.

Every one of those metrics is computed across whatever countries the creative happened to run in during the window. On a self-optimising network you did not choose that distribution and it does not hold still. AppLovin’s own scaling guidance is to [“Select all countries that you actively support”](https://support.applovin.com/en/growth/promoting-your-apps/analyze-and-optimize/best-practices-for-scaling-user-acquistion-campaigns) and to run “a single, global budget for optimal spend allocation towards the best outcomes across all countries”. Verified August 30, 2026\. Read that as a statement about your reporting. In the panel below, the median creative’s most expensive country costs 11.3 times what its cheapest one does, and you have been asked to put that spread behind one budget the network reallocates daily.

The mechanism is not new. Ad tech has [warned about Simpson’s paradox in programmatic buying](https://www.adexchanger.com/data-driven-thinking/marketers-cant-overlook-simpsons-paradox-in-programmatic-buying/) since at least 2018\. What we could not find anywhere was a measurement of how often it decides a creative’s fate.

## The split takes four lines of arithmetic

Write `w` for a country’s share of the creative’s spend in a window and `r` for the creative’s rate in that country. A blended rate over countries is then a weighted average, and its change between two windows separates exactly, with nothing left over:

```
R      = sum of w * r
mix    = sum (w2-w1)(r2+r1)/2
within = sum (w2+w1)(r2-r1)/2
change = mix + within
```

The identity holds term by term because `a2*b2 - a1*b1` is always `(a2-a1)(b2+b1)/2 + (a2+a1)(b2-b1)/2`. There is nothing to fit and nothing to tune. Across our 2,890 comparisons the largest residual was smaller than one part in a trillion, which is floating point and not method.

Split the right rate. Installs per dollar is a spend-weighted average of country rates, so it separates cleanly. Cost per install is the reciprocal of that average rather than an average of country CPIs, so decomposing it directly does not work. Run the split on installs per dollar or on ROAS, then read the answer back in whichever direction your team talks.

### One creative, two fortnights

|               | Fortnight 1                       | Fortnight 2                       |
| ------------- | --------------------------------- | --------------------------------- |
| United States | $2,000, 250 installs, $8.00 CPI   | $6,000, 800 installs, $7.50 CPI   |
| Brazil        | $6,000, 3,750 installs, $1.60 CPI | $2,000, 1,600 installs, $1.25 CPI |
| Blended       | $8,000, 4,000 installs, $2.00 CPI | $8,000, 2,400 installs, $3.33 CPI |

Blended cost per install rose 67%. The creative got cheaper in both countries it ran in. Nothing about the asset changed. The network moved three quarters of its spend from a $1.60 market into a $7.50 market.

In installs per dollar the blended rate fell from 0.500 to 0.300\. The mix term is -0.292 and the within term is +0.092\. Mix accounts for 76% of the movement, and it points the opposite way from the creative. This example is constructed to be clean. The measurement is not.

## What we measured

The panel is AppLovin’s advertiser report as AppLovin returns it: day by campaign by creative set by country by platform, for six advertiser accounts, 245 app and platform combinations, 236 countries, June 15 to August 9, 2026\. That is 21,140 creative sets, $17.4 million of spend and 32.3 million installs. Country is a dimension AppLovin reports at this grain. Nothing in the panel is allocated, modeled or predicted.

Each creative set was compared against itself across adjacent fourteen-day windows, inside the same account, app and platform. A comparison had to carry at least $500 of spend and 50 installs in both windows and touch at least three countries. Countries below 1% of a comparison’s spend are pooled into one bucket, which is deliberately conservative: movement inside that bucket stops counting as mix. That leaves **2,890 comparisons on 1,349 creative sets and $15.1 million of spend**, with a median of 70 countries behind each comparison and 15 of them clearing the 1% pooling floor.

The control repeats the identical procedure on alternating days inside a single twenty-eight-day window, so each half carries fourteen days of delivery, the same as a real window, and no real change separates them. Whatever the procedure reports there is its own noise.

## Half of the flags do not survive holding the mix still

Take the common rule: flag a creative when its blended cost per install rises 20% or more between fortnights.

|                                             | Real comparisons | Interleaved control |
| ------------------------------------------- | ---------------- | ------------------- |
| Comparisons                                 | 2,890            | 2,494               |
| Flagged                                     | 469 (16.2%)      | 100 (4.0%)          |
| Still flagged with the mix held still       | 51.0%            | 60.0%               |
| Median share of the movement carried by mix | 35.8%            | 22.2%               |
| Mix term larger than the within term        | 36.9%            | 22.0%               |
| Creative improved inside its own countries  | 12.6%            | 7.0%                |

The flag is doing real work. Flags fire four times as often on real fortnights as on the control, so most of them are catching something. What they are catching is often not the creative. **49.0% of flagged creatives did not clear the same 20% bar once only the within-country part of the change was counted**, against 40.0% of the control’s noise flags. On 12.6% of flagged comparisons the creative got cheaper inside the countries it kept running in while its blended cost per install rose.

The median flagged creative is the clearest way to hold this. Its blended cost per install rose 33%. With the country mix held still it rose 21%. About a third of the typical flagged rise was the map, not the ad.

## The steeper the drop, the less of it is the creative

This is the part we did not expect. Raising the flag threshold does not concentrate the real cases. It does the opposite.

| Flag threshold                    | Flagged | Still flagged, mix held still | Control |
| --------------------------------- | ------- | ----------------------------- | ------- |
| Blended cost per install rose 10% | 828     | 61.0%                         | 63.6%   |
| Blended cost per install rose 20% | 469     | 51.0%                         | 60.0%   |
| Blended cost per install rose 30% | 283     | 39.2%                         | 51.3%   |
| Blended cost per install rose 50% | 118     | 22.0%                         | 46.2%   |

At a 50% jump in blended cost per install, only 22.0% of creatives had actually got 50% worse in the countries they ran in. The control stays near half the whole way, so this is not the procedure sagging under a harder threshold. Large blended moves are large precisely because a large share of spend relocated, and relocating spend is what the network does, not what the creative does.

The practical consequence inverts a common habit. Teams triage the worst-looking declines first because they look most urgent. Those are the ones most likely to be about where the budget went.

## The mix hides real declines as well as inventing them

The error runs both ways. Among comparisons that never tripped the 20% flag, 5.6% had a within-country decline past that same threshold, hidden by a favorable move in the mix. That is 136 creative sets whose real deterioration the blended number concealed, against 2.8% in the control.

So the blended rate is not simply pessimistic. It reorders the queue in both directions, and a review that starts from it will spend time on creatives that are fine while leaving ones that are not in rotation.

## A second network with a different report says the same thing

AppLovin is one engine with one allocation policy, so a result that only holds there is a fact about AppLovin. We ran the identical procedure on Mintegral, whose [Advanced Performance Reporting](https://helpcenter.mintegral.com/en/docs/advanced-ad-delivery-report) returns Location and Creative as dimensions of one report, so the country split is reported there too and comes through an entirely separate integration. Verified August 30, 2026.

That panel is 48,646 creatives, 332 app and platform combinations, five accounts and $2.0 million of spend over the same dates. It is thinner per creative, so the floors drop to $200 and 25 installs, leaving 737 comparisons on 372 creatives.

|                                | AppLovin | Mintegral |
| ------------------------------ | -------- | --------- |
| Comparisons                    | 2,890    | 737       |
| Flagged at 20%                 | 16.2%    | 21.8%     |
| Still flagged, mix held still  | 51.0%    | 50.9%     |
| Median share carried by mix    | 35.8%    | 35.6%     |
| Median flagged, blended        | +33%     | +31%      |
| Median flagged, mix held still | +21%     | +20%      |

Two networks, two reporting paths, two different sets of advertisers, and the survival rate lands within a tenth of a point. We take that as evidence the effect is structural rather than a property of one engine.

## On day-7 ROAS the mix is not the story

The same decomposition on revenue gives the opposite answer, and it is worth saying plainly.

AppLovin’s advertiser report also runs in cohort mode, returning day-7 purchasers and day-7 revenue at the same creative set and country grain. Restricting to June 15 to July 26 so every cohort had its full seven days before the last restatement, that panel is 6,358 creative sets across 28 app and platform combinations and $4.6 million of spend, leaving 214 comparisons on 137 sets.

Flag a 20% fall in day-7 ROAS and 22.0% of comparisons trip it. The median flagged creative carries only 9.0% of its movement in the mix term, and 85.1% of flags survive holding the mix still. Country mix is not what moves day-7 ROAS at this grain.

That is not good news. The control trips 16.1% of the time. A day-7 ROAS flag on a fortnight is barely separable from noise, which matches what we found when we [re-ranked the same creatives twice inside one fortnight](https://lemon-ai.com/resources/creative-test-spend-threshold) and again when [pooling assets into creative sets failed to stabilise a revenue ranking](https://lemon-ai.com/resources/concept-level-creative-analysis). The country decomposition is a cost-per-install instrument. On revenue at this horizon the answer is that you do not have enough signal to decompose.

## Your account is not the average

Across the six AppLovin accounts, the share of flags surviving the mix hold ran from 11% to 78%. The largest account by spend sat at 53%. On one account the median flagged creative’s within-country rate had actually improved while its blended cost per install rose by half.

Do not carry 51% into your own review. Carry the procedure. The number is a property of how concentrated your delivery is, how far apart your markets price, and how aggressively your goals let the engine move budget, and those differ per account and per campaign structure.

## What the split cannot tell you

It cannot tell you a mix-driven decline was a mistake. If the network moved spend into cheaper, lower-value markets because your goal told it to buy volume, it did what you asked, and your blended economics really did get worse. The decomposition does not dispute that. It tells you which lever the problem is attached to, and the creative is not that lever.

It is also not a causal claim about the creative. Holding the mix still removes one confounder. Placement, audience, auction pressure, seasonality, competing creatives entering the same ad group and attribution drift are all still in the within-country term. The [fatigue measurement protocol](https://lemon-ai.com/resources/creative-fatigue-measurement) lists the rest of that queue, and the country check belongs at the front of it because it is the cheapest one to run and the one that most often changes the verdict. For the wider diagnosis, including the audience and placement effects this split does not touch, start from the [creative fatigue analysis guide](https://lemon-ai.com/resources/creative-fatigue-analysis).

Three more limits worth stating. The panel is eight weeks, so slow decay over a quarter is outside it. Fourteen-day windows were chosen to match the fatigue rules in circulation; at twenty-eight days the survival rate rises to 54.1%, so the answer depends on the window you review on. And a creative that enters or leaves a country entirely has no rate in one window, so its whole contribution is counted as mix. Restricting to countries present in both windows moves survival from 51.0% to 52.2%, which is small enough that the convention is not driving the result.

## What to do with each answer

| What the split says                                      | What is actually wrong  | What to do                                                                       |
| -------------------------------------------------------- | ----------------------- | -------------------------------------------------------------------------------- |
| The within-country term carries the drop                 | The creative            | Refresh or replace it, and write the next test from what the earlier version did |
| The mix term carries it, country rates flat or improving | Where the budget went   | Leave the creative alone. Look at country goals, targets and campaign structure  |
| Both terms negative                                      | Both, and they compound | Fix the allocation first, then re-read the creative on the corrected mix         |
| One country collapsed and it carried the spend           | That market             | Auction pressure, store page, localization or seasonality, not the asset         |

When the answer is the creative, the finding still has to survive being turned into an instruction. [Turning a creative finding into a testable brief](https://lemon-ai.com/resources/creative-test-brief) covers the handoff, and [creative attribute analysis](https://lemon-ai.com/resources/creative-attribute-analysis) covers whether the attribute you blame is actually the one that moved.

## Why AppLovin’s asset report cannot run this check

The whole procedure needs one thing: the creative’s numbers broken out by country. On AppLovin that dimension exists at creative set grain and does not exist at asset grain.

The [Asset Reporting API](https://support.applovin.com/en/growth/promoting-your-apps/api/asset-reporting-api) returns exactly these columns: `asset_id`, `asset_name`, `asset_url`, `campaign`, `campaign_id`, `campaign_package_name`, `clicks`, `cost`, `creative_set`, `creative_set_id`, `ctr`, `impressions`. No country. No platform. The [advertiser Reporting API](https://support.applovin.com/en/growth/promoting-your-apps/api/reporting-api) does carry country, documented as a “Two-letter country code”, next to `creative_set_id`, conversions, sales and revenue, and returns them cohort-dated when you set `day_column` to `day`. Both verified August 30, 2026\. That is why this measurement is run at the creative set and why [getting creative-level outcomes out of AppLovin takes more than one report](https://lemon-ai.com/resources/applovin-creative-reporting).

Any asset-level country split, including the one inside Lemon’s own product, is the creative set’s reported country distribution allocated across the assets in it. That is a reasonable way to read a dashboard and a circular way to run this test, so we did not use it here.

## Where Lemon does this for you

This measurement needed nothing from Lemon. It is AppLovin’s advertiser report, pulled with country as a dimension, and four lines of arithmetic. If your team can export that report on a schedule, you can run it.

What Lemon does is keep the report in a shape where the check is one click rather than a project. [Creative Analytics](https://lemon-ai.com/creative-analytics) holds country, platform and campaign as breakdowns on every creative row across the connected networks, marks network-reported values apart from reconstructed ones, and keeps the definitions stable between months so a comparison means the same thing twice.

Run the split on your next fatigue review before you pause anything, and keep the windows, the floors and the pooling rule fixed so the number means the same thing next month. If your account turns out to sit at 78% rather than 11%, that is worth knowing too. It means your flags are mostly honest and your review time is already going to the right creatives.

## Primary sources

- [AppLovin: Asset Reporting API](https://support.applovin.com/en/growth/promoting-your-apps/api/asset-reporting-api)
- [AppLovin: Reporting API](https://support.applovin.com/en/growth/promoting-your-apps/api/reporting-api)
- [AppLovin: Best practices for scaling user acquisition campaigns](https://support.applovin.com/en/growth/promoting-your-apps/analyze-and-optimize/best-practices-for-scaling-user-acquistion-campaigns)
- [Mintegral: Advanced Performance Reporting](https://helpcenter.mintegral.com/en/docs/advanced-ad-delivery-report)
- [Segwise: How to measure creative fatigue](https://segwise.ai/blog/how-to-measure-creative-fatigue)
- [AdExchanger: Marketers can’t overlook Simpson’s paradox in programmatic buying](https://www.adexchanger.com/data-driven-thinking/marketers-cant-overlook-simpsons-paradox-in-programmatic-buying/)

On this page

- [Every fatigue rule we found reads a number the network controls](https://lemon-ai.com/resources/creative-decline-country-mix#every-fatigue-rule-we-found-reads-a-number-the-network-controls)
- [The split takes four lines of arithmetic](https://lemon-ai.com/resources/creative-decline-country-mix#the-split-takes-four-lines-of-arithmetic)
- [What we measured](https://lemon-ai.com/resources/creative-decline-country-mix#what-we-measured)
- [Half of the flags do not survive holding the mix still](https://lemon-ai.com/resources/creative-decline-country-mix#half-of-the-flags-do-not-survive-holding-the-mix-still)
- [The steeper the drop, the less of it is the creative](https://lemon-ai.com/resources/creative-decline-country-mix#the-steeper-the-drop-the-less-of-it-is-the-creative)
- [The mix hides real declines as well as inventing them](https://lemon-ai.com/resources/creative-decline-country-mix#the-mix-hides-real-declines-as-well-as-inventing-them)
- [A second network with a different report says the same thing](https://lemon-ai.com/resources/creative-decline-country-mix#a-second-network-with-a-different-report-says-the-same-thing)
- [On day-7 ROAS the mix is not the story](https://lemon-ai.com/resources/creative-decline-country-mix#on-day-7-roas-the-mix-is-not-the-story)
- [Your account is not the average](https://lemon-ai.com/resources/creative-decline-country-mix#your-account-is-not-the-average)
- [What the split cannot tell you](https://lemon-ai.com/resources/creative-decline-country-mix#what-the-split-cannot-tell-you)
- [What to do with each answer](https://lemon-ai.com/resources/creative-decline-country-mix#what-to-do-with-each-answer)
- [Why AppLovin’s asset report cannot run this check](https://lemon-ai.com/resources/creative-decline-country-mix#why-applovins-asset-report-cannot-run-this-check)
- [Where Lemon does this for you](https://lemon-ai.com/resources/creative-decline-country-mix#where-lemon-does-this-for-you)

---

Related product

- [Creative Analytics](https://lemon-ai.com/creative-analytics)
- [Attribute Analysis](https://lemon-ai.com/attribute-analysis)

## Structured data

```json
{"@context":"https://schema.org","@graph":[{"@id":"https://lemon-ai.com/#organization","@type":"Organization","name":"Lemon AI","url":"https://lemon-ai.com/","logo":{"@type":"ImageObject","@id":"https://lemon-ai.com/#logo","url":"https://lemon-ai.com/brand/lemon-ai-mark-512.png","contentUrl":"https://lemon-ai.com/brand/lemon-ai-mark-512.png","width":512,"height":512,"caption":"Lemon AI"},"email":"hi@lemon-ai.com","sameAs":["https://cy.linkedin.com/company/lemon-ai","https://x.com/lemon_ai_adtech","https://www.g2.com/products/lemon-ai/reviews"]},{"@id":"https://lemon-ai.com/#website","@type":"WebSite","name":"Lemon AI","url":"https://lemon-ai.com/","publisher":{"@id":"https://lemon-ai.com/#organization"},"inLanguage":"en"},{"@id":"https://lemon-ai.com/resources/creative-decline-country-mix#breadcrumb","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Lemon AI","item":"https://lemon-ai.com/"},{"@type":"ListItem","position":2,"name":"Resources","item":"https://lemon-ai.com/resources"},{"@type":"ListItem","position":3,"name":"Half of Creative Decline Flags Are Country Mix","item":"https://lemon-ai.com/resources/creative-decline-country-mix"}]},{"@id":"https://lemon-ai.com/resources/creative-decline-country-mix#webpage","@type":"WebPage","url":"https://lemon-ai.com/resources/creative-decline-country-mix","name":"Half of Creative Decline Flags Are Country Mix","description":"We split 2,890 fortnight comparisons on AppLovin creative sets into the part the country weights moved and the part the creative moved. Half the flags vanish.","isPartOf":{"@id":"https://lemon-ai.com/#website"},"publisher":{"@id":"https://lemon-ai.com/#organization"},"inLanguage":"en","datePublished":"2026-08-30","dateModified":"2026-08-30","about":{"@id":"https://lemon-ai.com/#software"},"breadcrumb":{"@id":"https://lemon-ai.com/resources/creative-decline-country-mix#breadcrumb"}},{"@id":"https://lemon-ai.com/#software","@type":"WebApplication","name":"Lemon AI","url":"https://lemon-ai.com/","description":"Creative intelligence and cohort-revenue forecasting for mobile app and game user-acquisition teams.","applicationCategory":"BusinessApplication","operatingSystem":"Web","browserRequirements":"Requires a modern web browser","provider":{"@id":"https://lemon-ai.com/#organization"},"featureList":["Creative-level spend, installs, purchases, revenue, and ROAS","Creative attribute analysis","Evidence-led creative generation","Cohort revenue and ROAS forecasting through D365","Read-only reporting API and MCP access"],"offers":[{"@type":"Offer","name":"Creative Intelligence annual billing","price":"209","priceCurrency":"USD","url":"https://lemon-ai.com/#pricing","description":"Effective monthly price per app when billed annually.","availability":"https://schema.org/OnlineOnly"},{"@type":"Offer","name":"Cohort Prediction monthly subscription","price":"679","priceCurrency":"USD","url":"https://lemon-ai.com/#pricing","description":"Monthly price per app after a $1,399 one-time account setup.","availability":"https://schema.org/OnlineOnly"}]},{"@id":"https://lemon-ai.com/authors/gregory-potemkin#person","@type":"Person","name":"Gregory Potemkin","jobTitle":"Founder & CEO","url":"https://lemon-ai.com/authors/gregory-potemkin","image":"https://lemon-ai.com/images/join-us/gregory.webp","sameAs":["https://www.linkedin.com/in/gregory-potemkin/"],"worksFor":{"@id":"https://lemon-ai.com/#organization"}},{"@id":"https://lemon-ai.com/resources/creative-decline-country-mix#article","@type":"Article","headline":"Half of Creative Decline Flags Are Country Mix","description":"We split 2,890 fortnight comparisons on AppLovin creative sets into the part the country weights moved and the part the creative moved. Half the flags vanish.","image":"https://lemon-ai.com/og/resource-creative-decline-country-mix.png","datePublished":"2026-08-30","dateModified":"2026-08-30","articleSection":"Creative decline diagnosis","mainEntityOfPage":{"@id":"https://lemon-ai.com/resources/creative-decline-country-mix#webpage"},"author":{"@id":"https://lemon-ai.com/authors/gregory-potemkin#person"},"publisher":{"@id":"https://lemon-ai.com/#organization"},"inLanguage":"en"}]}
```