What counts as a normal CPI?

There's no single number — it varies heavily by geo, vertical and tier, so the benchmark isn't someone else's figure from a chat, it's your own ROI model: a CPI is fine if the funnel's economics still work at that level.

What matters more for CPI — the creative or the witrina?

Both do, but a slow or irrelevant witrina wastes even a strong creative — the user made the click, then either didn't wait for the load or didn't see what was promised.

Does cloaking lower CPI directly?

Not directly — it doesn't change your auction bid, but it stops impressions being wasted on bots and off-target traffic, so effective CPI on real users comes out lower than the raw account numbers show.

Is a wheel-of-fortune prelander a one-time trick or does it hold up?

It holds up as long as the warm-up honestly matches the offer — if what the wheel promises doesn't match what comes next, conversion further down the funnel drops and eats the gain you got in installs.

How many variants should I test at once without spreading budget too thin?

Two or three is typical — each variant collects meaningful stats faster than it would with five or six running on the same budget.

How to Lower CPI in PWA Campaigns

CPI isn't just an auction bid — it's click price divided by click-to-install conversion. Here are the levers that actually move it: witrina speed, geo and tier choices, cloaking, and prelanding.

What CPI is actually made of

CPI is click price divided by the share of clicks that make it to an install. That gives you two independent levers: lowering the click price itself (targeting, bid, how the platform scores your creative) and raising click-to-install conversion (witrina speed and relevance). Both need work, because a cheap click with a low install rate often lands at the same CPI as an expensive click with a high one.

Witrina speed matters more than it looks

On mobile connections, especially in Tier-2/3 geos, every extra second of load time loses part of the audience before they've even seen the offer. A lightweight, fast-loading PWA witrina, the kind APEX builds by default without unnecessary weight up front, directly lifts the share of users who reach the install step. It's worth testing load time on a throttled connection that mimics the target geo, not your own fast office Wi-Fi.

Geo and tier are an ROI question, not just a CPI one

CPI in Tier-2/3 geos usually runs lower than Tier-1, but payout per action is lower there too — comparing geos on CPI alone, without the payout side, doesn't tell you much. Within a single geo, CPI also swings with the season and with how many other buyers are currently targeting the same audience; a sudden CPI spike isn't always a creative problem, sometimes it's just heavier competition in the auction.

Cloaking hits CPI indirectly, but noticeably

Filtering bots, datacenters and off-target geo doesn't lower the click price directly, but it stops budget from being spent on traffic that was never going to install anyway, so effective CPI on real users ends up lower than the raw impression numbers suggest. A campaign that doesn't get flagged to moderators early also simply runs longer, which means you're not constantly re-warming a fresh domain and account, and that has its own cost.

Creative and prelanding raise click-to-install conversion

The closer the creative matches what a user actually sees on the witrina, the fewer clicks go nowhere — people who clicked on a mismatched promise almost never install. A warm-up prelander, APEX's wheel-of-fortune mechanic for example, placed before the install step raises the share of users who make it there, because they arrive already engaged instead of arriving cold off a click, and that mathematically lowers CPI even at the same click price.

Split tests instead of guesswork

Instead of changing the witrina or creative on a hunch, run several variants in parallel with sticky-by-visitor splitting and compare them on the full click → install → deposit chain against real traffic. The winning variant takes the bulk of the budget, not the one that just feels stronger, so CPI comes down on numbers, not gut feel.

FAQ

What counts as a normal CPI?: There's no single number — it varies heavily by geo, vertical and tier, so the benchmark isn't someone else's figure from a chat, it's your own ROI model: a CPI is fine if the funnel's economics still work at that level.
What matters more for CPI — the creative or the witrina?: Both do, but a slow or irrelevant witrina wastes even a strong creative — the user made the click, then either didn't wait for the load or didn't see what was promised.
Does cloaking lower CPI directly?: Not directly — it doesn't change your auction bid, but it stops impressions being wasted on bots and off-target traffic, so effective CPI on real users comes out lower than the raw account numbers show.
Is a wheel-of-fortune prelander a one-time trick or does it hold up?: It holds up as long as the warm-up honestly matches the offer — if what the wheel promises doesn't match what comes next, conversion further down the funnel drops and eats the gain you got in installs.
How many variants should I test at once without spreading budget too thin?: Two or three is typical — each variant collects meaningful stats faster than it would with five or six running on the same budget.

Build your PWA in APEX in minutes

Cloaking, anti-bot, push, split tests and your own domains — in one service.

Get started free