Platform ROAS will not tell you. A pragmatic way to allocate spend across Meta, Google and TikTok.平台 ROAS 不会告诉你。一套在 Meta、Google、TikTok 之间分配预算的务实方法。
Allocating budget across channels is where most marketing plans quietly go wrong. Each platform reports a flattering ROAS, none of them account for what the others contributed, and so the "data-driven" split is really just the loudest dashboard winning the argument.跨渠道分配预算，是大多数营销计划悄悄出错的地方。每个平台都报一个好看的 ROAS，没有一个会算上其他平台的贡献，于是所谓"数据驱动"的分配，其实只是"嗓门最大的看板"赢得了争论。
Start by separating roles, not just channels. Search and shopping mostly harvest existing demand; paid social and TikTok mostly create it. If you cut the demand-creation channels because their last-click ROAS looks weak, your "efficient" search campaigns slowly starve. The channels are a system, not a leaderboard.先区分"角色"，而不只是"渠道"。搜索与购物多半在收割已有需求；付费社媒与 TikTok 多半在创造需求。如果你因为"末次点击 ROAS 看着弱"就砍掉创造需求的渠道，你那些"高效"的搜索计划会慢慢被饿死。渠道是一个系统，不是一张排行榜。
For the actual numbers, triangulate. Use a lightweight marketing-mix model for the top-down split, validate the surprising parts with geo holdout tests, and let platform data handle only in-flight optimization within each channel. You will not get a perfect answer — you will get a defensible one, which is far better than the false precision of summed platform ROAS.至于具体数字，用三角校验。用一个轻量的营销组合模型做自上而下的分配，用地理对照实验去验证其中反常的部分，平台数据只负责各渠道内部的实时优化。你不会得到一个完美答案——但你会得到一个站得住的答案，这远胜过"把各平台 ROAS 加总"那种虚假的精确。
Then make it a habit, not a quarterly fire drill. Re-allocate on a fixed cadence against actuals, move budget in measured steps rather than lurches, and always keep a small slice for testing a new channel before you need it. The goal is not a perfect split today — it is a process that keeps finding a better one.然后把它变成习惯，而不是每季度一次的救火演练。按固定节奏依实际数据再分配，预算挪动要分档、而非猛拉，并始终留一小块预算，在你需要某个新渠道之前就先去测它。目标不是今天分得完美——而是一套能持续找到更优解的流程。