CPI (Cost Per Install)

CPI is the cost per counted install of a mobile app, typically measured via SKAdNetwork on iOS or the platform SDK on Android. This is an economics metric, which means it is only useful next to the other numbers in its chain. On its own it can be gamed: a great number on a tiny denominator tells you nothing, and a poor number can be the correct trade for volume. Read it alongside spend, order volume, contribution margin and the time window the platform used to attribute the result. Read it next to CPA (Cost Per Acquisition), CTR (Click-Through Rate), Conversion Rate (CVR).

The calculation itself is simple — CPI = Ad Spend ÷ Installs — and the judgement is entirely in the inputs and the window you choose. Treat the metric as a decision rule, not a scoreboard. Write down the threshold at which you would scale, hold, or cut before you look at the report — then let the number answer that question. Segment by campaign objective, audience temperature and creative concept, because a blended figure hides the two or three line items actually moving it. Worked through: $8,000 spend, 4,000 installs → CPI = $2.

Pull the number from one source of truth and keep the window fixed. Platform reporting, your analytics suite and your order system will disagree, usually because of attribution windows and refunds. Pick the system your P&L trusts, note the window, and compare like-for-like week over week rather than chasing daily noise. Why this matters commercially: CPI is the primary top-line KPI for mobile UA teams. But it's dangerous alone — installs that don't convert to a downstream event are worthless.

Xeli reads this metric back to the creative and the SKU that produced it, so the next production run is weighted toward what actually paid. Instead of a spreadsheet reconciling creative names to results, each rendered asset carries its concept, offer, ratio and product ID — which turns the metric into a brief for the next batch. In the context of metrics & economics, that means the concept stops being something a person re-applies by hand every campaign and becomes a rule the system enforces on every asset it produces.

The recurring problems are predictable: optimising to install instead of install + event; comparing ios and android cpi without noting the skad delay; ignoring creative fatigue on high-frequency ua. Each of these is a process gap rather than a knowledge gap — which is why the fix is usually a checklist, a template or an automated rule instead of more training.