When CPI is the right model
CPI fits products whose value event is frequent and close to the install: casual games, utilities, social and dating apps with fast funnels. If your monetization event happens weeks later — a deposit, a subscription renewal — pure CPI buys blind, and CPA or hybrid models serve better. The first thing we do with any brief is map the pricing model to your funnel math, because the model decides what the buying algorithm optimizes for.
How we keep CPI honest
An install is only worth buying if a real device behind it can plausibly become a customer. Every CPI source passes fraud filtering before your MMP — emulators, click injection, SDK spoofing and install farms are blocked at placement level — and then faces weekly scoring on post-install behavior: retention, session depth, payer share. Cheap installs that behave like bots are the most expensive line item a UA budget can carry; our scoring exists to find and cut them before they scale.
Scaling CPI without diluting quality
The classic CPI failure: a profitable test scales, quality dilutes, and the account bleeds margin for a month before anyone reads the cohort data. We scale in controlled steps — budget doubles only for sources holding target economics in the weekly scoring — and pair every budget increase with fresh creative, because scale doubles frequency and frequency kills creatives. The result is CPI volume that stays profitable at ten times the test budget, not just at the test budget.