创业与商业thecorrespondent.com编辑精选
新的互联网泡沫：在线广告
The new dot com bubble is online advertising (2019)
2020 — 2021，它一次次回到首页
黑点是上首页（≥ 50 分）的那次投稿，空心点是没冲上去的投稿，点越大分数越高。
导读
这篇 2019 年的调查报道指出，在线广告可能是一个巨大的泡沫：eBay、宝洁等大公司多次实验发现，停投或削减在线广告后销量几乎没有变化，说明大量广告费的效果被严重高估。文章质疑整个数字广告行业的价值可能建立在难以验证的归因之上。
为什么经久不衰：它对数字广告有效性的根本质疑在多年后依然成立，且随着隐私监管和归因技术变化，今天读来更具现实意义。
HN 评论精选
大品牌难衡量广告效果，但小创业公司能清晰测出广告确实有效
我记得 2019 年原文刚发表时就读过，不明白为什么现在又翻出来，它当年就是错的。像可口可乐这样的大品牌确实很难衡量在线广告回报——停投一周大概什么都不会变，这和电视广告是同一个问题。但任何做过低知名度消费类小创业公司的人都能很容易地衡量广告回报：你能看到是哪条广告把用户带来、他们花了多少钱。大多数创业公司迟早会因为某种原因停投一周广告，然后立刻看到收入下滑。我参与过一家消费硬件创业公司，COO 每周在全员会上公布 Google、Facebook、电视等各渠道的细粒度 ROAS 数据，反复 A/B 测试并直接按收入优化，广告有效是毫无疑问的。
avalys查看英文原文
I remember reading this article back in 2019 when it was originally written. Unclear why it is being dusted off now when it is still just as wrong as it was then. Yes, it is probably hard for a large, well-known brand like eBay or Procter & Gamble or Coca-Cola to measure their return from online advertising. If Coca-Cola stopped all advertising for 1 week, would anything really change? Probably not. And it’s not as if anyone is clicking Google search ads for Coca-Cola and ordering a 6-pack right there. This is the same problem that these companies have with TV advertising. But anyone who has ever run a small, consumer-focused startup with low brand recognition can very easily measure their return on ad spend, and will spend a lot of time doing this. You can easily tell which specific ad referred someone to your website, and how much money they spent once they got there. If they’re not convinced by this data, at some point most startups will find the opportunity to simply turn off all advertising for a week for one reason or another. And can usually see the drop in revenue immediately. I was involved in a consumer hardware startup where our COO shared granular ROAS numbers in our all-hands every week for Google, Facebook, TV advertising across multiple networks, etc. They regularly A/B tested different advertisements and messages across different media and directly optimized for revenue. It was clear beyond a doubt that this advertising worked. The company would not have been viable without it. The fact that this proof is easily and readily available from small, lesser-known companies is part of why large companies continue to spend money on advertising despite the benefits being much harder to measure.
有效广告靠独立测试和真金白银衡量，别信平台和代理的指标
一本《Scientific Advertising》加一些思考就足以做好广告。有效广告有四条规则：1) 无视在线平台（尤其是卖广告的）的所有指标；2) 无视广告代理公司的指标；3) 做有意义的测试；4) 用美元衡量。难的是说服创始人停止乱动以便测出结果。经验法则：如果没人对测试结果感到不安，说明两组差异太小，应该做更大的测试。
jlbnjmn查看英文原文
A copy of Scientific Advertising and some thinking are about all that's needed to do the job well. There are essentially 4 rules for effective advertising: 1) ignore all metrics from online platforms (especially the ones that sell ads) 2) ignore all metrics from ad agencies 3) run tests that make sense* 4) measure in dollars The hard part is convincing the founder/owner to stop messing with stuff long enough that you can actually measure the results. Also, companies tend to way overspend on ads and data scientists while underspending on skilled advertisers. (They can be forgiven for the latter, there are too few skilled advertisers, and lousy advertisers usually sell themselves as experts.) *Example: your job is to advertise a business with 20 locations in 2 cities. You want to know if radio ads are better than search ads. So you do the obvious thing: buy radio ads in city A and search ads in city B, then compare sales. If the difference in sales is big enough that managers get defensive and start pointing out why the test wasn't valid, you may have a winner. Run it in both cities and see if it still works. (Rule of thumb: if nobody is upset with the outcome, the variations were too similar. Run bigger tests.)
增量性对照实验证明大部分数字广告确实有效
每隔几个月就有人来质疑数字广告。我始终回到增量性分析这个真正的证据上：把人群分成两组，一组看广告、一组不看，观察业务增长并衡量两组之间的提升。懂得大规模投放的公司一直在做这种测试，Facebook、Google 等平台也支持这类实验。最终大多数广告被证明是有效的，当然也有例外。
kposehn查看英文原文
As this is a repost (no offense to OP), I'm going to repost my original comment (thread here: https://news.ycombinator.com/item?id=21468505) about this which is still applicable: Reliably someone comes along every few months to question digital ads. I always come back to analyses of incrementality as the real proof. Take an audience of X people. Divide them in two. Show ads to your test group, don't show to control. Watch your business grow and gauge the lift between the two audiences. The companies that know how to advertise at scale do this constantly and can gauge the real effect of their ad dollars. Facebook, Google and others make these tests possible in their platforms, while other software suites such as Impact Altitude and VisualIQ allow you to do this kind of analysis and testing as well. In the end, most of it proves out to be incremental. There are notable exceptions of course, but when are there not?
个人化广告投放效果可疑，用户反而像占了便宜
我一直觉得有什么不对：我能在 YouTube 上听几乎任何歌，代价只是 5 秒后跳过广告。很多广告根本不是我感兴趣的产品，要么是快速致富骗局，要么是 Grammarly 这类软件的洗脑式重复投放。如果 Google 通过侵犯隐私建立了我的复杂画像，那他们显然没用它向我推销我感兴趣的东西。我倒觉得自己在这笔交易里占了便宜。
heurisko查看英文原文
I have had the thought that there is something wrong, that I can access nearly any song of my choice on YouTube, with the small penalty of skipping an advert after 5 seconds. Many of the adverts aren't even for products I'm interested in. They're either get-rich-quick schemes, or repeat-ad-nauseum ads for software like Grammerly, which I will never use. If Google has some sort of complex picture about me, by invading my privacy, then they're certainly not using it to sell me anything I am interested in. I guess if Google are funalling cash from advertisers to the record companies, then it will continue. But I have to wonder what would happen if advertisers actually look at how effective their advertisements are. I feel I'm getting a good deal out of this arrangement.
历次讨论
同类还值得读
微软已死（2007）
Microsoft is Dead (2007)
Paul Graham 2007 年的一篇短文，判断微软在互联网与创业竞争中已不再令人畏惧、失去影响力——所谓“死”并非字面倒闭。
创业与商业上首页 8 次 · 2007–2025paulgraham.comH&R Block 与 Intuit 游说反对简化报税（2017）
H&R Block and Intuit Lobbying Against Simpler Tax Filing (2017)
2017年ProPublica调查报道梳理了H&R Block和Intuit如何通过游说反对更简单的美国报税方式，尤其是政府预填税表和免费申报等改革。
创业与商业上首页 5 次 · 2017–2022propublica.org你本不该有老板
You Weren't Meant to Have a Boss (2008)
Paul Graham 2008 年的经典文章，论证人类在演化中习惯于小团体协作，而大公司的层级结构和老板制度是人为产物，对程序员尤其有害——因为编程的本质是创造新事物，而大公司…
创业与商业上首页 5 次 · 2008–2020paulgraham.com为什么要在经济糟糕时创业
Why to Start a Startup in a Bad Economy
Paul Graham 的经典短文，论证经济衰退并不是创业的障碍：真正决定成败的是创始人和产品，而非宏观环境；衰退期反而竞争更少、人才更易招、成本更低。
创业与商业上首页 5 次 · 2008–2022paulgraham.com做那些不可规模化的事
Do Things that Don't Scale
Paul Graham 2013 年的经典创业文章《Do Things that Don't Scale》。
创业与商业上首页 4 次 · 2013–2025paulgraham.com编辑精选