Last-click is dead and platform ROAS lies. A practical stack for knowing what your spend actually drives.末次点击已死，平台 ROAS 会骗人。一套切实可用的组合，帮你看清预算到底带来了什么。
Every platform reports its own ROAS, and if you add them up you will "earn" more revenue than your company actually made. That is not fraud — it is double counting. Each platform claims the conversions it touched, and many conversions are touched by several.每个平台都报自己的 ROAS，如果你把它们加起来，会"赚到"比公司实际收入还多的钱。这不是造假——是重复计算。每个平台都把它接触过的转化算到自己头上，而很多转化被好几个平台同时接触过。
The fix is not a single source of truth — it is triangulation. We run three measurements in parallel: platform reporting for in-flight optimization, marketing-mix modeling for top-down budget allocation, and incrementality tests for the questions that actually matter.解法不是某个"唯一真相来源"，而是三角校验。我们并行跑三套度量：平台报表用于实时优化，营销组合建模（MMM）用于自上而下的预算分配，增量实验用于回答真正重要的问题。
Incrementality is the one most teams skip and the one that pays off most. A simple geo holdout — turn a channel off in matched markets — answers the only question that matters: would this revenue have happened anyway? We have seen "winning" branded-search campaigns turn out to be 60% free.增量实验是多数团队跳过、却回报最高的一环。一个简单的地理对照（在匹配市场里关掉某渠道）就能回答那个唯一重要的问题：这些收入本来是不是也会发生？我们见过"表现优异"的品牌词战役，最后被证明 60% 本就是免费的。
Server-side tracking underpins all of it. With browser signal eroding, conversion APIs that send events from your server keep optimization fed and reporting honest. It is unglamorous plumbing — and it is the highest-ROI work we do for most new clients.服务端埋点是这一切的底座。在浏览器信号不断流失的当下，从你服务器发送事件的转化 API，既喂饱了优化、又让报表诚实。这是不起眼的"管道工程"——却是我们为多数新客户做的、ROI 最高的工作。