I’m continuing to finish writing the Apple Ads guide – it’s already grown to 40 pages of raw text.

I’m honestly trying to create something genuinely useful, to explain my experience in as much detail as possible.

There’s still a lot of work ahead, but for now, here’s an excerpt from the guide.

First, a few axioms.

Axiom #1:
The only correct way to decide whether a source, campaign, creative, or keyword is successful is based on financial metrics.

Axiom #2:
The only correct way to make a decision is to rely on statistically significant data.

Axiom #3:
Good metrics for one business may be unacceptable for another business due to many factors:

I’m not saying that other companies might have completely different goals: capturing market share, “investing” in the founders’ salaries, or upselling users to other products.

From these three axioms, two conclusions follow:

Let’s add two more theorems to these 3 axioms and 2 conclusions.

Theorem 1:
You need “empty pockets” to evaluate each keyword based on final ROAS. Given that ad auctions can change, it’s important to make decisions quickly. And for this, you have to make decisions based on incomplete data.

Theorem 2:
Any ad campaign (like a product, business, or happy relationship) is partly the result of luck.

Great. Now let’s dive deeper.

Why should we rely only on financial metrics?

The world is nonlinear. Not every impression leads to a click, not every click leads to a product page view, not every product page view leads to an install, and each install…

well, you get it.

For example, in Meta Ads, you can optimize for reach and get hundreds of millions of very cheap impressions, which result in exactly $0 revenue.

Or you can optimize for installs, which will also result in very little revenue.

Meanwhile, if you optimize for purchases in Meta Ads, your CPM and CPI will rise several times, but your ROAS will also be much higher.

Apple Ads is far from Meta Ads, but similar principles still apply:

Do you know what guarantees high ROAS? High ROAS guarantees high revenue and low spend.

In other words, only high ROAS can guarantee high ROAS.

Different keywords can attract different audiences.
Different GEOs can attract different audiences.
Different screenshots can influence which audience comes into your app.

This is why intermediate metrics are not always relevant.

In my Apple Ads campaigns for OpenChat, there are at least 2 keywords that show unmatched intermediate metrics and, most importantly, some of the lowest cost per trial.

I haven’t seen numbers like this in this niche for a long time.

And this happens repeatedly in different countries.

But users from these keywords, with rare exceptions, convert terribly to payments. Just some anomaly.

I also have a couple of GEOs with terrible cost per (trial + 1-year subs). By my benchmarks, such campaigns shouldn’t pay off. However, in the end, ROAS turns out above average.

This is because annual subscriptions prevail for these keywords in these countries, not weekly trials.

There’s also the reverse case. I mainly rely on 7-day ROAS as my primary metric for deciding whether to stop a campaign.

However, some countries have very low 7-day ROAS, yet their cumulative ROAS is within the normal range.

All because users in these countries aren’t ready to take annual subscriptions (which greatly affects near-term ROAS), but still

So I can’t rely 100% even on metrics that are closest to ROAS in the funnel. What to say about CPT/CPI/TTP/CR, etc.?

To calculate everything perfectly, you’d need an analyst to build predictive ROAS and account for all such nuances. But we’re a small company, and we can’t afford that yet.

Despite this, I still can’t wait for all available information that would let me 100% decide whether to keep a keyword running.

I have to rely on any available intermediate metrics, whether there’s enough statistically significant data.

Even if CPA doesn’t always correlate with final ROAS, I will still turn off campaigns with high CPA.

The smaller the budget, the more you need to rely on early metrics.

Any successful campaign (and company) is partly luck.

When I first launched Apple Ads for my project, I did everything not by the book:

I launched many countries within a single campaign using broadly matched keywords.

This allowed me to quickly find working GEOs and turn the source into a profitable one (I’ll discuss this approach later).

Only then did I arrive at my current structure.

However, it’s crucial to get some positive feedback from the start. Otherwise, it’s impossible to scale any source.

Your task is to find one successful GEO and a few successful keywords so that

And for this, you’ll have to turn off keywords and campaigns even when the data is incomplete.

You can return to these keywords and retest them later. At first, your goal is to find at least one working combination.

You should define the concepts of “too low” or “too high” based on your historical data.

The smaller your budget, the more you’ll have to rely on random chance. Unfortunately, that’s the way it is.