We all have our quirks, and one of mine is that I need to know how stuff works. Sometimes that helps me, because it takes me down rabbit holes that turn out to be useful. Most of the time it leads to me lying awake at night thinking about pointless crap. The other day I was lying awake thinking about pointless crap when I had a thought about the bot farm that stole my ad budget. What if iPhone users are worth more to advertisers simply because they’re harder to fake?
For a lot of the pointless crap I think about, there’s no way to prove it one way or the other. Fortunately, this time there is. I run a small puzzle app called Dayzle. The app needs advertising, and since I’m my own marketing department, I’m free to waste my budget any way I want. So what if I ran the exact same ad campaign on iPhone that got scammed on Android? Will I just waste more money, or can I prove that iPhone users are worth more because they’re harder to fake?
Some background, if you missed the first post. Earlier this month I spent $220 on a Google Ads campaign for our Android app, with the goal set to “installs”. 33 of the 56 installs I paid for opened the app once, spent zero seconds on any screen and never came back. It was a bot farm, and every fake install made the campaign look like it was working, so Google kept sending the farm more of my ads.
Here’s the thought that kept me up. Advertisers pay about 19% more to reach an iPhone user: $19.63 per thousand rewarded video views in the US, against $16.49 on Android. The usual explanation is that people with iPhones have more money to spend. But if a bot farm can eat 60% of an Android budget, maybe part of that premium is just advertisers paying for fewer fakes.
My first theory about why iPhones would be harder to fake was hardware, and it was a lovely theory right up until I checked it in the morning. An Android bot farm can be built from cheap phones plugged into any old PC, and I assumed an iPhone farm needed a Mac for every phone or two. It turns out you need a Mac once, to set up the software that controls the phones, and after that a Linux or Windows machine can run them. Real test setups run 5 to 8 iPhones per computer, about the same as Android. A Mac mini costs about US$599, which won’t stop anyone.
The reasons that survived the morning are about what a farm is allowed to fake:
- On Android, you don’t need a phone at all. Emulators are software that pretend to be phones, and one server can run hundreds of them. Apple’s version only runs on a Mac and can’t install apps from the App Store, so app install fraud on iPhone requires a real iPhone.
- Every iPhone install has to come from the App Store. The farm that found us on Android installed our app from a saved copy of the file, which is how 20 phones ended up running a version the Play Store had stopped offering. On iPhone, every download is locked to the Apple account that downloaded it, even for a free app, so a saved copy won’t run on anyone else’s phone.
- Every install needs its own Apple account, and every Apple account needs a phone number to set up.
I’m an economist by training, so here’s the version I find convincing. On Android, most of a farm’s costs are fixed: once the servers are running, one more fake install costs almost nothing. On iPhone, every fake install costs a real phone’s time, a real Apple account and a real download. And every account that gets flagged as fake is a real loss too: a phone number, the time it took to set up, and every install it can’t make anymore. The Android farm that found us didn’t even need an account, because it installed our app from a saved copy. When each fake install costs real money, you should get fewer of them.
I had a fourth reason, and it turned out to point the other way. iPhones ask you whether an app can track you, and I assumed that meant Google would see less of what happens on an iPhone, which would make the farm’s job harder. Then I remembered that the farm answers that question too, and it presumably says yes every time. Among our real iPhone players this month, about 1 in 3 who saw Apple’s question said yes. So the installs Google can see most clearly on iPhone might be the farm’s, which would be very on brand for this story. The good news is that it’s measurable. If the installs that behave like bots say yes almost every time and real people say yes a third of the time, that’s one more way to tell them apart.
So here is the experiment, and the dumb part is on purpose. The iPhone campaign is identical to the Android one: the same headlines, the same images and videos, CA$40 a day, and the goal set to “installs”, the exact setting that got farmed last time. If I change anything else, I won’t know what made the difference. I’m aware that paying twice for the same lesson is not a strategy you’d find in a marketing textbook.
My prediction is a bit of math. What I actually pay for is real users, so my true marketing cost is what I spend divided by the real users I get, and real users are just installs minus bots. On Android, 56 installs minus 33 bots left 23 real users for $220, or about $9.57 each. Now suppose the whole iPhone premium is paying for fewer fakes. Then a real user should cost the same on both phones, because otherwise everyone would buy the cheaper one. At 19% higher prices, the same $220 buys about 47 iPhone installs, and to get real users at $9.57 each, 23 of those 47 need to be real. That’s about 51% bots, against 59% on Android, and it’s the break-even line. If iPhone lands near 51%, the premium is exactly the bot discount and the market has priced it right. My guess is that it beats the line by a lot: under 30% bots, which would make a real iPhone user about 30% cheaper than a real Android one, even though each install costs more. Under 20% wouldn’t surprise me.
Once the install numbers are in, I’ll switch the goal to “won a puzzle”, the same change I made on Android after the farm showed up, and see what that does to the price.
So, will I just waste more money? Maybe. If I’m right, part of the reason iPhone users are worth more is that they’re harder to fake, and a known-broken ad setup should be noticeably less broken on iPhone. If I’m wrong, I’ll have paid for the same mistake twice, but at least it’ll be settled and I can stop thinking about it, so if anything I’ll sleep better. I’ll report back with the results so that I’m the only one that has to burn money on this.