Two cohorts of advertisers, same market, same auctions. They differ by 2.4 percentage points in transaction cost, and by 19.4 percentage points in media productivity loss.

That is the ANA’s Q1 2026 Programmatic Transparency Benchmark, and the ratio is the whole story: the money lost inside the inventory is roughly eight times the money lost in the fee stack. The end result is that the weaker cohort paid 19.04 dollars per thousand qualified impressions against 7.46 for the stronger one. Same unit, 2.55 times the price.

So the honest answer to “how do I reduce CPI” is not “bid better”. It is that your cost per real install was mostly decided before the bid was calculated, by which impressions you were willing to buy at all.

The mobile-specific version of that number

Pixalate analysed over 80 billion programmatic impressions in Q2 2026 and put invalid traffic at 39 percent in United States mobile apps, and 41 percent globally for mobile app traffic, against 20 percent for web and 26 percent for CTV.

That figure is measured on buy-side open auction traffic before filtration, which is exactly why it sits so far above the post-filtration numbers in the ANA benchmark. The two are not in conflict. They measure different points on the same path, and the distance between them is the value of the filter.

For a UA team the practical reading is blunt. In open-auction app inventory, the single largest determinant of cost per install is what share of your impressions were addressed to a human.

The supply path is the second lever, and it is not about fees

Jounce Media’s Q1 2026 benchmarking finds that just 68 percent of available programmatic supply is maximally direct, and that 22 percent of all programmatic investment still flows to wasteful auctions. Duplicate paths to the same impression do not make that impression cheaper. They add hops, each with a take, and they let the same inventory bid against itself.

Note that this is one firm’s methodology, published alongside a client case study, so use the industry figures and ignore the marketing.

The third lever is creative throughput, and the market moved

AppsFlyer reports top gaming advertisers producing between 2,400 and 2,600 creative variations per quarter, up 25 to 30 percent year over year, while total gaming UA spend grew 3.8 percent to 25 billion dollars.

Two numbers from the same release explain why that matters for cost. Paid install share rose 10 percent year over year while ad impressions rose 20 percent. The market is buying roughly twice the impressions per install it needed a year ago. Creative variance is the main lever against that, and it is now a baseline rather than an optimisation.

The checklist

1. Recompute CPI as a quality-adjusted number and rank sources on that. Divide spend by installs traceable to fraud-free, measurable, viewable, non-MFA impressions rather than by raw installs. A source that looks cheap on reported CPI and expensive on the adjusted version is the one draining the budget, and the 7.46 against 19.04 dollar spread is what that looks like at cohort scale.
2. Pull sellers.json and the SupplyChain object for your top 20 spending app bundles and cut every path that is not maximally direct. This is one afternoon against 30 days of log-level data, and it removes duplicate routes to the same impression rather than negotiating the price of any one of them.
3. Turn on pre-bid invalid traffic filtering for mobile app inventory, and re-baseline your CPI target upward on the same day. Announce the new baseline before you flip the switch, for the reason in the limitation below.
4. Put measurability rate in the weekly report as a standing line, and keep a hard MFA exclusion list. Lower-performing advertisers lost 22.7 percent of spend to non-measurable inventory in Q2 2026 against 10.4 percent for higher performers, and MFA exposure roughly doubled to 1.1 percent in Q1 2026 after sitting between 0.4 and 0.6 percent throughout 2025.
5. Set a creative throughput floor per quarter and hold to it. If the top of the market ships 2,400 variations a quarter and your team ships 40, no bidding strategy closes that gap.

Why the contract shape decides how hard this is

Every item on that checklist is an attempt to stop paying for something that was never a person. On a CPM or CPI contract, that work is entirely the advertiser’s, because the supplier is paid on delivery either way and has no financial exposure to the quality of what was delivered.

For example, at SpinX the model is CPA with MMP-verified attribution, on owned and operated supply plus direct publisher relationships. Invalid traffic hits our margin before it reaches a client’s ROAS, and the conversion that triggers payment is confirmed by the advertiser’s own measurement partner. Filtering is not a service line we add to a plan. It is the only way the model stays solvent.

If your CPI looks competitive and your cost per real user does not, that gap is worth measuring on pay-per-result inventory at spinx.io.