Why Machine Learning Sucks for Predicting Ad Success & What Big Media Layoffs Mean for Marketing

Marketing School - Digital Marketing and Online Marketing Tips

Eric Siu and Neil Patel

Business, Marketing, Careers

4.6 • 1.4K Ratings

🗓️ 12 January 2024

⏱️ 20 minutes

🔗️ Recording | iTunes | RSS

🧾️ Download transcript

Summary

In episode #2656, we discuss a paper that reveals the limitations of sophisticated ad engagement propensity scoring techniques. Emphasizing the importance of creativity and the need to add a unique spin to marketing strategies, the conversation then shifts to the increasing number of media layoffs in 2023, with major companies like BuzzFeed experiencing significant stock drops. We stress the importance of adapting marketing approaches to stay ahead in a rapidly changing landscape. Tune in for valuable insights and predictions. Don’t forget to help us grow by subscribing and liking on YouTube! Check out more of Eric’s content (Leveling UP YT) and Neil’s videos (Neil Patel YT) TIME-STAMPED SHOW NOTES: (00:00) Today’s topic: Why Machine Learning Sucks for Predicting Ad Success & What Big Media Layoffs Mean for Marketing (00:34) Discussion on the limitations of machine learning algorithms in predicting campaign success (01:37) Graph showing the discrepancy between machine learning models and control tests in predicting campaign success (02:58) Conclusion that creativity cannot be predicted and the importance of adding a unique spin to ads (06:11) Conversation on the increasing prevalence of advertisements in various aspects of life (08:57) Exploring the idea of embracing the advertising trend rather than complaining about it (09:33) Introduction to Peter Drucker's perspective on the two main things that matter in business: marketing and innovation (10:08) Emphasis on the need for businesses to adapt to the changing marketing landscape or risk failure (10:24) Media layoffs in 2023: Impact on marketing (11:12) Reasons for media layoffs: Lack of adaptation and cheaper content creation (11:51) BuzzFeed's stock plummeted 97% in five years (12:32) Overvalued media companies and lack of profitability (13:09) Major news outlets announcing layoffs (14:11) CNBC's viewership during economic downturns (14:38) That’s it for today! Don’t forget to rate, review, and subscribe! Go to https://www.marketingschool.io to learn more! Leave Some Feedback: What should we talk about next? Please let us know in the comments below Did you enjoy this episode? If so, please leave a short review. Connect with Us: Single Grain << Eric’s ad agency NP Digital << Neil’s ad agency X @neilpatel X @ericosiu See omnystudio.com/listener for privacy information.

Transcript

Click on a timestamp to play from that location

0:00.0	All right, so we're going to kick this episode off by talking about how and why machine
0:06.5	learning isn't what you might think it might be in terms of predicting ad campaign
0:12.0	success.
0:13.0	So let me explain what that means first and we're going to go through a paper here and
0:15.9	then Neil now we're going to jump through a couple of different ideas that we have for today's
0:19.1	topics but basically let's say your Facebook or meta okay let's say your Facebook or Meta, okay, let's say your Google, you have a huge corpus of data and you see a lot of people running billions of ads out there, right? So you haven't a sense for what tends to work and what doesn't.
0:34.0	Now you would assume that these machine learning algorithms would have, would do a good job of predicting the next campaign success, right?
0:41.0	And like I would assume the same thing too,
0:44.0	because you work off of data.
0:45.6	But here's an interesting thing.
0:47.0	This paper, and I read part of this paper,
0:49.8	I won't say I read the whole thing here,
0:51.3	but Neil, can you see my screen?
0:55.1	Okay so here's what it says over here for those of you that haven't
0:58.5	subscribed on YouTube go subscribe on YouTube because we're just gonna get better
1:01.1	and better there but this this tweet says, fascinating paper,
1:04.8	it finds that even with access to rich data sets of user level features,
1:09.1	sophisticated ad engagement, propensity scoring techniques fail failed to approximate the causal lift results of
1:15.3	RCTs. RCTs just call a test, right? So RCT just means like a test. That's all it means at the end of
1:21.1	day. Just think of it as an experiment. And in some cases,
1:24.0	dramatically overestimate them. So this graph over here that those, those of you that are watching right now,
1:30.9	this just shows like in the RCTs in red so that's the original test right
	...

Please login to see the full transcript.

Previous episode | Next episode

Disclaimer: The podcast and artwork embedded on this page are from Eric Siu and Neil Patel, and are the property of its owner and not affiliated with or endorsed by Tapesearch.

Generated transcripts are the property of Eric Siu and Neil Patel and are distributed freely under the Fair Use doctrine. Transcripts generated by Tapesearch are not guaranteed to be accurate.