The first thing anyone notices about the App Store is the thirty percent. It is the wrong number to worry about. Thirty percent is a price, it is written down, and you can decide in advance whether the remaining seventy is a business. What is not written down anywhere is the rest of the arrangement: that the shelf, the search results, the payment relationship and the right to be installed at all belong to somebody else, and are lent to you on terms that can be rewritten without your agreement.
That is not a complaint. The alternative, building your own distribution from nothing, is so much more expensive that almost nobody starting out should choose it. It is simply worth being honest that what looks like owning a product is often renting a channel, and that the rent is not really charged as a percentage. It is charged in the probability that the terms change.
What dependence actually costs
Suppose in any given year there is some chance that the channel you depend on does something that materially hurts: a ranking change, a policy that outlaws a feature you built on, a review process that suddenly reads your category differently, a competing first-party app shipped in the operating system. Call that probability \(q\). The chance of getting through \(t\) years untouched is:
A small annual risk is not a small risk once it is compounded over a decade.
The numbers are unkind. At a ten percent annual chance, a decade of dependence carries roughly a thirty-five percent chance of arriving intact. At twenty percent, which is not unrealistic for anything sitting close to a platform's own ambitions, it is about eleven. Nothing has to go badly wrong in a particular year for the tail to be thin. It just has to be possible every year.
Two things follow. The first is that channel risk is a property of duration, so it changes the kind of product worth building, not merely the marketing. A product that has to run for ten years to be worth building is a different bet than the same product with a payback measured in months. The second is that \(q\) is not fixed. It rises the closer you sit to whatever the platform intends to do itself, and falls the more boring and self-contained your product is. Being uninteresting to your landlord is a real form of insurance.
Rent paid in permission
The part that costs most is rarely the money. It is that permission has to be asked for repeatedly, and the asking has a variance you do not control. A release can sit in review while a launch date passes. A feature can be approved for two years and then not. The correct response is not indignation, it is inventory: knowing which parts of your product exist at somebody else's discretion, and not letting that set grow quietly.
So we try to keep a small number of things ours regardless of what any platform decides. Whoever the customer is, and a way to reach them that does not route through a store. The domain, so that an address we have published for years keeps working. The data, in a form that can leave. None of that produces growth on its own. All of it means a bad month is a bad month rather than the end.
The version of this I believe
Borrowed distribution is worth borrowing. It is the reason one person can put something in front of millions of people without a sales team, and dismissing it as a trap is the kind of purity that produces no users at all. The mistake is not using the channel. The mistake is building as though the channel were an asset you hold rather than one you rent, and then being surprised when the terms change.
Practically, that means a few unexciting rules. Assume the channel changes at least once in any decade, and check that the product still means something if it does. Do not build a core mechanic on a permission that is granted by policy rather than by physics. Keep the direct relationship, even when it is smaller and slower than the borrowed one, because it is the part that survives.
The way to think about it is that distribution you borrow buys you time, and time is exactly the thing you should be spending on building something you own. Companies that forget the second half of that sentence do very well for a while, and then discover that everything they built was standing on ground rented by the year.