Indonesia produced 870 million mobile game downloads in a single quarter, about 45 percent of everything Southeast Asia installed.

Thailand, with less than a quarter of Indonesia’s population, produced the region’s highest in-app purchase revenue in that same quarter, 162 million dollars.

Both figures are Sensor Tower’s, from the same report. Read together they explain why so many regional campaigns look busy and land nowhere. Southeast Asia ranks second in the world on downloads and seventh on in-app purchase revenue. The volume and the money are not in the same country.

That is not a targeting problem. It is a campaign structure problem, and the auction solves it against you.

What a single regional campaign actually optimises for

Give one campaign six countries, one budget and one bid, and it will buy the cheapest qualifying installs it can find. Mechanically, it has to. So budget flows to the market with the deepest supply of low-cost inventory, and most of your volume lands where monetisation is thinnest.

You did not choose that country weighting. The auction chose it, and it chose on price.

The spread inside the region is wider than the spread between regions

Adjust’s H1 2026 data on shopping apps makes the point with one metric, one vertical and one half-year. Singapore was the fastest growing e-commerce install market in the world at 67 percent year over year. Vietnam grew 42 percent. Indonesia grew 36 percent. That is a 31 point spread between three markets people routinely buy as one line item.

Then the ranking inverts. On session growth over the same period, Indonesia led at 62 percent and Singapore followed at 58 percent. The market with the slowest install growth had the fastest engagement growth.

Run those three countries together, judge them on installs, and you will conclude Indonesia is the weakest. Look at sessions and it is the strongest. Same data, same half-year.

Device mix makes the point harder to argue with

Statcounter’s August 2026 figures for mobile operating system share:

1. Philippines, 85.55 percent Android.
2. Indonesia, 77.07 percent Android.
3. Thailand, 64 percent Android.
4. Malaysia, 54.09 percent Android.
5. Singapore, 51.06 percent iOS.

There is no regional default that is correct here. A 60/40 Android assumption, the sort that quietly lands in a media plan, is wrong in all five markets. And your country weighting is what picks your platform mix, which then picks your measurement stack, your creative specs and your attribution reality.

One more number from the same source, because it changes what you ship rather than what you bid: 23.3 percent of Indonesian handsets sit in Statcounter’s “Unknown” vendor bucket. Roughly a quarter of that device base is long tail hardware your creative was probably never tested on.

The organic cushion is thinning, so the mistake costs more than it used to

The global paid to organic install ratio reached 0.72 in H1 2026, up 26 percent on 2025 and 47 percent on 2024, in Adjust’s data. Emerging markets used to forgive sloppy geo allocation, because organic installs quietly topped up the numbers. That subsidy is shrinking, which means blended CPA is converging on real CPA and structural waste surfaces faster.

Payments add the last layer. Google, Temasek and Bain’s e-Conomy SEA 2025 report puts cash at 39 percent of regional transaction value, down from 48 percent in 2023, with 10 countries running national QR systems and 8 offering cross-border QR interoperability. A checkout flow tuned for card entry converts differently in Jakarta than in Singapore, and that difference lands in your CPA rather than in your media plan.

Five things worth doing this month

1. Split the region into one line item per country before you split by anything else, and cap Indonesia’s share of regional spend. It is roughly 40 percent of regional gaming downloads, so left unconstrained it absorbs the budget by default.
2. Set the platform split per country from observed device share, not from a regional average. iOS-first measurement for Singapore and Malaysia, Android-first and low-spec creative variants for the Philippines and Indonesia.
3. Pick one post-install KPI and let the two markets you most often blend run separately against it for a full quarter. Install rank and engagement rank move in opposite directions here, so install-level comparison will mislead you.
4. Localise the checkout and the store listing before localising the creative. Wallet and QR penetration differ by market and they move the event you are optimising toward.
5. Rebuild CPA targets per country from your own data rather than from a published benchmark, for the reason in the next section.

Limitation

There is no credible public CPI benchmark by Southeast Asian country for 2026. Adjust’s H1 2026 release publishes a global median of 1.08 dollars, North America at 3.12 and India at 0.10, and nothing per SEA market. Every “CPI by country 2026” table ranking in search today is an aggregator page with no named dataset, no sample size and no methodology. Using one is worse than using nothing, because it hands a wrong number the authority of a benchmark.

The second limitation is operational, and it is the real reason teams avoid this. Six countries means six creative sets, six languages, six payment flows and six learning curves. Splitting a campaign six ways splits the learning signal six ways too. A test that reaches significance in three weeks as one regional line item can take three months per country, by which point the creative is stale.

Buy regionally to find the market, buy hyperlocally to scale it, and know which of the two you are doing. Singapore and Malaysia can reasonably share a setup, 5 points apart on iOS share. The Philippines sits 36 points from Singapore and cannot.

Where the supply relationship decides the outcome

Country-level control is only real if you can see and choose the inventory behind each market. On a resold chain, “Indonesia, Android, gaming” is a label applied somewhere upstream, and the gap between a Jakarta puzzle player and whatever cleared cheapest under that label stays invisible until it reaches your D7 numbers.

For example, at SpinX a large share of the inventory is owned and operated, with the rest from direct publisher relationships, so app, placement and market are known values rather than inferred ones. Combined with a CPA model and MMP-verified attribution, a per-country test is priced on outcomes in that country rather than on impressions labelled with it.

If your Southeast Asia numbers look fine in aggregate and disappointing per market, that is worth testing on pay-per-result inventory before the next budget cycle, at spinx.io.