[saltythrowaway]

Throwaway for obvious reasons.

I joined to work on meta ads semi-recently. My thinking was that they were under invested in techniques I knew were successful from my time at Google.

What I found was a lot of smart people who knew they were under invested and we're capable of closing the gap, but the infrastructure they sat upon was too hard to use to do anything different, and management that rewarded the appearance of great work more than great work. The result was a lot of config changes making noise in metrics that could be read positively, without implementing the state of the art systems that took a year before you could write a fancy success workplace post about it. That would risk poor performance reviews and the shuffling of reasources away from your team as a result.

And now, it's becoming apparent that the very very smart people started leaving after the stock dropped and the replacements are both more expensive and less able to get used to Facebook's infra. The result on the ground looks like a very serious engineering death spiral that management has not signaled they understand or have a grip on.

[pevey]

Your behind the scenes perspective is very interesting.

From a small business customer perspective: I have a small local business that I run on the side. It's a franchise type of business like what you would see in your local strip mall. We used to get our best leads from Facebook ads. More than 90% of the clicks were mobile. A high percentage of the clicks would set a tour and a decent percentage of those sign up their kid(s) for the program we offer. We went from about 20 good leads per month to 0-1 for the same ad spend. Yes, it really was that extreme. I don't know who facebook is showing our ads to, but they do not want our product.

I just got around to canceling with the agency that handles our ad spend today. The N of this story is 1, but if other businesses are experiencing anything close to the same, fb's revenue decline is just beginning.

[hintymad]

> management that rewarded the appearance of great work more than great work. T

Two reasons: 1. The management can't really differentiate appearance from substance. 2. The management is not incentivized to reward great work. Either way, it shows the failure of Meta's culture.

> but the infrastructure they sat upon was too hard to use to do anything different

Ironically, so many Meta engineers had such superficial understanding of their platforms and infrastructure that they couldn't even pass the most basic interview discussions on systems design. Seems another sign that Meta grossly over hired or had a culture that focused only on some so-called impact.

[dom96]

> The result was a lot of config changes making noise in metrics that could be read positively, without implementing the state of the art systems that took a year before you could write a fancy success workplace post about it. That would risk poor performance reviews and the shuffling of reasources away from your team as a result.

Meta has recently switched to a performance review cycle of 1 year (from 6 months). It sounds like one of the reasons this was done was to help make such work more possible.

[saltythrowaway]

I think the actual issue is the underlying culture. Google worked fine with multiple review cycles and you could get a good rating working on something hard that only shipped when it was ready.

However it is an optimistic sign in that leadership must get that part of the problem (long term vision is penalized) and is taking steps to address it. Maybe I judged them too harshly above; I don't know what I would do in their position to turn the ship around.

[eachro]

if you're allowed to say, what sort of techniques from google are they underinvested in?

[saltythrowaway]

Embedding-similarity based retrieval for one. Probably shouldn't get into the details here but Google has made incredible advances in the retrieval step of your typical retrieval / ranking pipeline and for a variety of reasons meta has not implemented them in production, despite having open sourced tools in this area. There are a few other advances like this Google has been using for long enough to publish on, meta engineers have working prototypes, but nothing is in production and everything breaks all the time.

[tomaskafka]

Oh, I have a similar story from a local country-leading job portal. I tried to persuade them to invest in embedding-based search instead of the category (which is often wrong and ambiguous by both advertisers and searchers) + keywords, but '2 people for 1 quarter to demonstrate a prototype able to be tested in production' seemed too risky for the management, despite having really cool results from the offline prototype :/.

The risk of missed opportunity is harder to visualize and thus rarely considered.

[candiddevmike]

Channel your saltiness into uncovering illegal/fraudulent stuff happening within ads and become a whistleblower. Based on anecdotes (some within these comments!) stuff is clearly sideways with how ads are being delivered and attributed.