Desktop software CPI benchmarks and rates

The published rates

Per cleared install — attributed by a single-use signed token, countersigned by the receiving application, and still in use 48 hours after the claim.

Category	Per cleared install	macOS	Windows	Examples
Utilities and consumer	EUR 12–18	Same	Same	Menu-bar tools, cleaners, converters, window managers
Creator tools	EUR 20–30	Same	Same	Screen recorders, editors, design tools, capture apps
Prosumer and B2B	EUR 35–50	Same	Same	Developer tools, database clients, API clients, team software

Network fee: 30%. Serving app earns: 70% of the price that cleared, or 84% of it in value if taken as credits. Currency: EUR, integer cents. Auction: second price, above the floor only, 30% of inventory reserved at floor.

These are facts about GHAN's prices, fixed in packages/shared/ratecard.ts and returned by the public, unauthenticated get_rate_card tool. They are not estimates of a market rate.

How cost per install works for desktop software

An advertiser pays a price each time their application is installed as a result of a placement. Two questions decide whether any given CPI rate is good, and neither is the number itself.

Question one: what counts as an install? A network that bills on the install event charges you for software landing on a disk. A network that bills on retention charges you for software still being used. Those are different products at the same nominal price.

Question two: who sets the price, and is there a floor? In an auction with no floor the clearing price is the price, so a funded competitor arriving in your category prices you out entirely. GHAN publishes a floor that never rises and that anyone can always buy at, and runs a second-price market only for priority above it, with 30 percent of each category reserved at floor. Price discovery is not the problem; price discovery with no floor is. See bidding.

A worked calculation

A creator tool advertising at EUR 24 per cleared install, over one campaign week:

Metric	Count	Billed
Impressions	40,000	EUR 0 — never billed
Clicks	1,200	EUR 0 — never billed
Install claims received	260	EUR 0 at this stage; 260 × EUR 24 moves into escrow
Rejected by a fraud rule	18	EUR 0 — holds refunded
Did not survive to 48 hours	61	EUR 0 — holds refunded
Cleared installs	181	EUR 4,344
Effective cost per install claim	—	EUR 16.71
Effective cost per retained user	—	EUR 24.00

The two effective figures at the bottom are the ones worth carrying between networks. A network billing on the install event would have charged for all 260 and reported a EUR 24 CPI — while your actual cost per retained user was EUR 34.47, 44 percent higher than the headline.

These counts are an illustration of the arithmetic, not observed GHAN data. GHAN has not traded yet.

Definitions

Term	Definition
Impression	A templated card rendered by the SDK in a member application, signed by that app's key. Never billed.
Click	A person activated the card; a single-use token was issued with a 15-minute expiry. Never billed.
Install	The application arrived on a machine. On its own this is not a billable event on GHAN.
Install claim	The receiving SDK found the token on first run and signed a claim with its own key. Triggers an escrow hold, not a charge.
Cleared install	An install claim that survived to 48 hours of real use. The only billable unit.
Billed install	Synonymous with cleared install on GHAN. Distinct elsewhere.
Excluded fraudulent install	A claim rejected by one of the fourteen rejection conditions. Refunded automatically, exposed with a reason code.

Most disagreements about CPI are disagreements about which of these rows the price attaches to.

Comparing a CPI rate to search spend

1. Take the search spend for a period.
2. Count the installs from that spend that were still active 48 hours later — not total installs.
3. Divide. That is your real cost per retained user.
4. Compare it to the rate card above, which is already a retained price.

Step 2 is the one most teams have never run. It usually moves the number by a factor of two to four. Longer treatment: GHAN vs Google Ads.

Methodology for the observed report

Once trading starts, this page gains a dated observed section generated from the ledger rather than written by hand:

• Sample period — explicit start and end dates, never a rolling unlabelled window.
• Split — by category band, by platform (macOS, Windows), by country, and by campaign volume decile.
• Reported per band — median and interquartile range of effective cost per cleared install, the clearing rate, and the fraud exclusion rate.
• Exclusions — every chain rejected by a fraud rule is excluded from the CPI figures and reported separately, so the two numbers cannot be traded off against each other.
• Minimum sample — no band is published below 200 cleared installs in the period. A band under the threshold is shown as insufficient sample, not omitted.
• Reproducibility — every figure recomputable by a partner from get_ledger output for their own share of it.

First publication: the month the first chains clear. Progress: status.

Until then, the only real numbers on this page are GHAN's own published prices, and they are labelled as such.