First, decompose the rise
Break the cost per install into three terms for each campaign and platform: the cost of a thousand impressions (CPMCost per 1,000 impressions.), click-through and store conversion. The price of each impression, divided by the share of viewers who tap and then by the share of store visitors who install, gives the cost per install. Compare the terms over the period of the rise and a comparable earlier period. The CPA calculator walkthrough for app installs sets out these inputs and shows how much each lever is worth, so the arithmetic need not be rebuilt here. Take the store conversion from the store page visits the ads produced, not from the listing's overall rate across every visitor.
Then mark the campaign's optimisation goal for each week. If the rise starts the week the goal moved from installs to an in-app action, set it aside as a deliberate change and judge it on cost per activated user. Finally, check the reporting source and the platform. The CPA walkthrough warns that installs reported late, in aggregate or by model can make recent cost per install look worse than it is, and the app attribution guide explains that Apple's SKAdNetwork postbacks arrive after a timer and in aggregated form. Compare like periods from the same source, and keep iOS and Android apart.
The decomposition sorts the causes. Impression price rising across every campaign points at the auction. Click-through falling on the same audience points at creative. Store conversion falling points at the listing. All three steady while the cost rises with each budget step points at scale. Installs that cost more but do not activate are a separate fault, covered by the diagnosis of installs that never become active users.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Click-through falls on the same audiences while impression price holds | Click-through by creative and by week since launch, and what each asset shows in its opening moments | Creative worn out, or built to describe the app rather than show it | Rotate in fresh concepts that open on the app being used |
| Cost per install jumps the week the campaign goal changed | Optimisation goal history against cost per install and cost per activated user | Optimisation moved to a deeper in-app event, which raises each install's price by design | Keep the event if cost per activated user fell; report both costs together |
| Click-through holds but fewer store visitors from the ads install | Store conversion for ad traffic, and which listing each campaign lands on | A store page converting fewer of the visitors the ads send | Land each campaign theme on a matching custom listing and fix the page |
| Delivery stalls or swings after an SDK, partner or schema change | Event postback timing and counts from the measurement partner to each network | In-app events reaching networks late or not at all, so bidding learns from less | Restore near-real-time postbacks and a consistent event definition |
| Impression price rises across campaigns, creative and networks at once | Impression price against the same weeks last year and against the account's own history | Seasonal or competitive pressure raising the price of impressions | Plan budgets for the season and judge cost against like periods |
| Each budget step buys fewer installs than the last | Marginal installs per extra unit of spend at each step | Budget pushed past what the audience supports, so marginal installs cost more | Hold or step back to the level where the marginal cost meets the target |
Fix, verify and prevent each cause
Creative that has stopped earning the tap
The app install campaigns guide says creative is where most of the performance variance in app campaigns lives once the optimisation event is set, that feed creative fatigues within weeks, and that ads should open on the moment of value and show the interface honestly rather than describe it. The auction pricing guide adds the price effect: in feed auctions a bland or tired ad loses on predicted response and its impression price drifts up.
Fix: retire the assets whose click-through has decayed, and brief a new batch of concepts that show the app being used in the first second, with alternative hooks for each.
Verify: click-through recovers on the same audiences and, with it, the cost per install, while cost per activated user holds or improves.
Prevent: keep a regular production rhythm of new concepts, as the guide recommends, and track click-through by asset age so decay is spotted before the cost rises.
A deeper goal doing its job
Install optimisation finds people likely to install, cheaply; event optimisation finds people likely to act after installing, at a higher price per install and a far lower price per useful user. The install campaigns guide describes that trade and the usual sequence: start on installs, move to the activation event once volume allows, and only then consider a revenue event.
Fix: if the rise followed a deliberate move to an event, there is nothing to undo; check instead that the event occurs often enough for the network to learn, and step back to a shallower event if delivery stalls.
Verify: cost per activated user, or per registered user, falls against the period before the move.
Prevent: report every app campaign on CPIAverage ad cost of each app install. and cost per activated user side by side, and record each goal change with its date, so the next reviewer does not mistake design for decline.
A listing that converts fewer of the ad's visitors
Every paid install passes through a store page the advertiser does not fully control, so a listing that converts less raises the cost of every campaign that lands on it. The custom store listings tutorial builds variant pages whose first screenshots repeat each ad's promise, and notes that a page converting more of the same clicks makes each install cheaper.
Fix: point each campaign theme at a listing that matches its ad, and where every source's conversion fell, work through the diagnosis of a dropped app store conversion rate.
Verify: store conversion for ad traffic recovers in the console's source reports, and the decomposition shows the cost falling through that term alone.
Prevent: add the landing listing to every campaign's launch check and keep the register of which campaigns point at which page.
Bidding that learns from a thinner signal
Networks optimise towards the events they receive. The attribution guide explains that events sent late or batched arrive after the network's optimisation window has moved on, that the event a campaign optimises to must be defined identically in the app, the measurement partner and the network, and that on iOS the conversion value schema is the only post-install signal for most installs. A broken SDK update, a changed postback setting or a reworked conversion value schema can starve the system without any visible error.
Fix: confirm postbacks fire close to real time for each network, restore one definition of the optimisation event everywhere, and change the iOS schema rarely so results stay comparable.
Verify: event counts at each network match the partner's for the same period, and delivery steadies within the learning period the network's help centre describes; the Google Ads Help centre documents its own App campaign settings.
Prevent: include a postback check in the release checklist and alert the paid team whenever event counts at a network drop away from the partner's.
A market that got dearer
A buyer controls the inputs, not the market. The auction pricing guide is explicit that how many competitors want the same people, and what those people are worth to them, sits outside the advertiser's control, and the install campaigns guide notes that seasonal competition for the same audiences, especially before major shopping periods, raises the cost of every install whatever the bid strategy says.
Fix: compare against the same weeks of the previous year, shift flexible budget to quieter periods, and lean on channels where the audience is less contested.
Verify: the impression-price term explains the rise in the decomposition, click-through and store conversion are steady, and cost falls back when the season passes.
Prevent: build seasonal impression prices into the budget plan, and read every cost figure against the account's own history for the same period rather than against last month.
Spend beyond the responsive audience
Every scaling path eventually meets Diminishing ReturnsEach extra unit of spend or effort buys less extra result than the one before it.: the most responsive people were reached first, so each further increment of budget buys fewer or dearer results. The guide to testing and scaling performance campaigns explains that the average cost climbs slowly while the marginal cost climbs fast, and that scaling decisions belong on the marginal figure.
Fix: calculate marginal cost per install, and per activated user, for each recent budget step; hold or step back to the level where it meets the target, and find the next scaling path in new creative, audiences or markets.
Verify: at the held budget, marginal cost settles near the target and the average stops rising.
Prevent: agree before each step the marginal cost at which scaling stops, and scale in steps rather than doubling, as the install campaigns guide warns.
A prevention routine for install costs
Make the decomposition the standing report. Each week, for each campaign and platform, show the impression price, click-through, store conversion from ad traffic, cost per install and cost per activated user, against the same weeks of the previous year as well as the period before. Annotate every goal change, creative batch, listing change, budget step and release, so the next rise can be matched to a dated event before anyone guesses.
Keep three habits running underneath. Brief new creative on a rhythm rather than in response to a bad week, and retire assets by their click-through decay. Check the measurement chain after every release, because the cheapest way to raise install costs is to stop telling the networks who the good users were. And set each campaign's length, spend and scaling limits before launch, as the install campaigns guide advises, judging the result at the end rather than reacting to daily movements in a figure that the auction and late postbacks both distort. The app install campaign launch checklist covers the pre-launch half of this routine, and other app symptoms are listed in the app marketing troubleshooting section.
Twirlbadger Run Club reads a quarter of rising costs
Twirlbadger Run Club, an invented running app, watches its Android cost per install climb from £1.60 to £2.70 over a quarter on its main feed network. The decomposition shows three movements. Its best video, running for eleven weeks, fell from 1.4% click-through to 1.2%, which explains the rise to £1.90 in the first two months. In the sixth week the team moved optimisation from installs to "first run logged", which raised cost per install from £1.90 to £2.20 while cost per activated runner fell from £9.50 to £6.10. In the final month impression prices rose by about a fifth, matching the same weeks the previous year, as fitness apps competed for new-year audiences. Store conversion from ad traffic held at 31% throughout. Twirlbadger keeps the event goal, ships four new videos that open on a run being tracked, and plans next year's budget with the seasonal price built in. With the new creative, cost per install settles at £2.35 and cost per activated runner at £6.40, despite the season.
Common mistakes
Reading cost per install alone, and cutting the event-optimised campaigns that buy the best users. Blaming the auction before checking click-through and store conversion. Comparing this month with last month across a seasonal peak. Changing goals and budgets every few days, which restarts learning. Treating late iOS postbacks as a cost rise.
