Zuckerberg just declared war on the entire advertising industry

[rcjordan]

Mark Zuckerberg says ads will soon be handled entirely by AI | The Verge

https://www.theverge.com/meta/659506/mark-zuckerberg-ai-facebook-ads

[Brad]

My bet is, after a time, there will be a demand for a pre-AI analog ad network like we have now.  Not for the big corps but for the small to medium advertisers.

[buckworks]

you don't need any targeting demographic

Um, savvy targeting makes a <huge> difference to the profitability of ad campaigns.

I don't want to be paying for ads while the AI fumbles around to discover (I refuse to say "learn") what works.

I'm turning into a grumpy old girl curmudgeon!

[Travoli]

>paying for ads while the AI fumbles around

My understanding is Google's approach has been to slowly remove manual Adwords tools and steer everyone into their automated ad optimization system. And then they optimize for profit.

Probably a good time to buy some META.

[buckworks]

>> optimize for profit

I had a bit of experience with some of Google's early automatic systems and it was meh. Even with two advisors from Google working on the account, their ads never outperformed mine. The campaigns I had built and tweaked by hand got a better CTR, lower CPC and a better conversion rate. I took care to craft different ads for different queries, and to use tightly relevant landing pages, whereas Google just used the same thing over and over and varied the bids.

I think my ideal would be hand-crafted campaigns to optimize my own way for the most important searches, then have separate campaigns to let AI spread its net to catch possibilities I might have missed.

[ergophobe]

Even with two advisors from Google working on the account, their ads never outperformed mine.

This has been my wife's experience as well. I'm not sure whether this is a basic weakness of current automated systems generally, or a function of account size, with small accounts not generating enough data for an algorithm to optimize effectively, but humans having a lot more context and understanding of the underlying business and the context in which that business takes place.

In the long run, I believe automated systems will outperform humans on all but the smallest campaigns, and even those with proper prompting. And by
"long run" I'm thinking at most 5 years, not 10.

[buckworks]

>> humans having a lot more context and understanding of the underlying business and the context in which that business takes place

This!

[ergophobe]

> This!

I do think that human advantage only exists for small campaigns though where you are mostly making guesses in the absence of a lot of data.

I would sometimes get requests to A/B test something and I would have to explain that the page they wanted to A/B test for, say, a spring sale, would take 400 days to get a big enough sample to have enough statistical power to find a 20% lift and many years to detect a 2% lift, by which time the spring sale is obviously over. In those cases, you are drawing on intuition and the intuition of a person who has watched a lot of similar campaigns is just going to have a feel that a self-learning system won't.

However, once the sample size gets big enough that the human is poring over data to make decisions, the human is no longer necessary in that system.

[buckworks]

>> human is no longer necessary

I'm not convinced of that. The campaign I mentioned above got 100k+ clicks in the test period and that wasn't enough for Google's tools to improve on my own results.

I could envision AI getting sort-of-profitable results, but for truly maxed-out ad performance I think a savvy human is irreplaceable. The larger the campaign the bigger the payoff for savvy human fine-tuning. The difference between, say, getting to 4% CTR instead of 2% sounds small, but it could double the profit.

>> watched a lot of similar campaigns

It's not just watching campaigns that matters, it's understanding the target customer. Emotions are a major factor in marketing and an AI just looking for patterns within datasets would miss a lot.

[ergophobe]

>>  100k+ clicks

I was thinking of campaigns with hundreds or thousands of ad variations and millions and millions of clicks as the situation where an AI will outperform a human because the number of variables becomes too great for a human mind to see patterns.

In other words, when you're in a "big data" situation (not just ads), self-learning systems shine and humans get overwhelmed.