How a walled garden works
Google, Meta, Amazon, TikTok and LinkedIn each run a closed system. You buy inventory inside it, the platform matches your ads to its own logged-in users, and it reports back through its own interface. What you get out is aggregated: totals, breakdowns and audience segments. What you cannot get out is the user-level record that would let you join a platform’s clicks to another platform’s clicks, or to your own customer database, on equal terms.
That closure is deliberate. The logged-in identity graph is the reason these platforms can target and measure so well, and letting it leave would hand a rival the advantage. The practical consequence for an advertiser is that each garden marks its own homework, and none of them can see the others.
Why walled gardens matter
Because their reporting is not neutral. A platform decides which clicks and views count, how long the credit window runs, and how missing data is estimated. Every one of those choices is defensible on its own and every one of them tends to flatter the platform making it. When two gardens both see the same buyer, both will claim the sale, which is the main reason platform and analytics totals disagree.
It matters commercially too. The more of your measurement, audience data and creative production lives inside one garden, the harder it is to move budget elsewhere, and the less able you are to answer a simple question: what would have happened if we had not advertised there at all?
Where it goes wrong
The common error is reading garden reports as a scoreboard and ranking channels by the ROAS each one prints. Those figures were produced under different rules and cannot be lined up. Adding them together is worse, because it counts shared customers more than once.
A second error is letting first-party data drift. Businesses build audiences inside a platform, never export the equivalent list of customers to their own system, and discover during a policy change or an account restriction that their audience was never really theirs. A third is assuming a platform’s own lift study is independent evidence — it is useful, but it is still the garden grading itself.
What to do about it
Treat each platform as a source of tactical detail — which creative, which audience, which placement — and take profitability decisions outside the garden, from your own order or enquiry records. Keep a customer list and consented contact data in a system you control, so the relationship survives any single platform.
Where the spend justifies the effort, verify a channel with an incrementality test rather than a platform report, and design your tracking and reporting so the same conversion definition is used everywhere. For smaller advertisers, including most businesses in Nepal, full experiments are often out of reach; a simple thank-you page question about how the customer found you costs nothing and gives an outside view the gardens cannot provide.