50 ms·

Twitter still has higher ad conversion rates. Machine learning can only do so much. Human editing will always win over machine algorithms, and Twitter users c

Twitter still has higher ad conversion rates. Machine learning can only do so much. Human editing will always win over machine algorithms, and Twitter users curate their follows. Also, machine learning only improves conversion rates by 2x. Meanwhile, I've seen a 10x conversion rate improvement when hand editing ads on a site. Machine learning ads are why you have out-of-context ads that destroy a sites user experience.

mbesto 11y ago
> Twitter still has higher ad conversion rates. Source? Genuinely curious, about to kick off a campaign on Twitter in an esoteric area (triathlon) and can't imagine it will be a higher ROI than FB. > Also, machine learning only improves conversion rates by 2x. Meanwhile, I've seen a 10x conversion rate improvement when hand editing ads on a site. I'm not sure you're following my line of reasoning here (or maybe I'm not following yours). The machine learning becomes useful when I try to target people's behaviors. For example, I say "Facebook I want to run an ad against all people who actively run and get them to buy running shoes". Facebook is going to optimize to serve up ads to people who (1) run (determined by messages they send to people "hey great run, I did 5km last week" or integrations with RunKeeper, Strava, etc, or maybe I have pictures of me running a race) (2) click ads (easy enough) and (3) will purchase online (determined by conversion pixels of previous campaigns). This is a gross oversimplification of course.
- argonaut 11y ago
  If we're on the subject of citing sources, what are your sources for saying that: > the thesis is 100% incorrect and guessing that: > Facebook is crawling every piece of data possible (pictures, messages, etc) and then draws conclusions of my interests there. In addition to this, all FB user info is being hashed and compared against consumer spending data That's a pretty sweeping statement. You'd be surprised at how naive (and distinctly not-amazing) the machine learning algorithms used in-production at places like Facebook can be (in-production, as opposed to cool research papers that aren't actually used in-production). Going off of your simplified example, there are tons of contexts in which someone might say "run" without meaning anything related to the athletic activity. That's the definition of shitty targeting (false positive).

mbesto 11y ago

> Twitter still has higher ad conversion rates. Source? Genuinely curious, about to kick off a campaign on Twitter in an esoteric area (triathlon) and can't imagine it will be a higher ROI than FB. > Also, machine learning only improves conversion rates by 2x. Meanwhile, I've seen a 10x conversion rate improvement when hand editing ads on a site. I'm not sure you're following my line of reasoning here (or maybe I'm not following yours). The machine learning becomes useful when I try to target people's behaviors. For example, I say "Facebook I want to run an ad against all people who actively run and get them to buy running shoes". Facebook is going to optimize to serve up ads to people who (1) run (determined by messages they send to people "hey great run, I did 5km last week" or integrations with RunKeeper, Strava, etc, or maybe I have pictures of me running a race) (2) click ads (easy enough) and (3) will purchase online (determined by conversion pixels of previous campaigns). This is a gross oversimplification of course.

argonaut 11y ago
If we're on the subject of citing sources, what are your sources for saying that: > the thesis is 100% incorrect and guessing that: > Facebook is crawling every piece of data possible (pictures, messages, etc) and then draws conclusions of my interests there. In addition to this, all FB user info is being hashed and compared against consumer spending data That's a pretty sweeping statement. You'd be surprised at how naive (and distinctly not-amazing) the machine learning algorithms used in-production at places like Facebook can be (in-production, as opposed to cool research papers that aren't actually used in-production). Going off of your simplified example, there are tons of contexts in which someone might say "run" without meaning anything related to the athletic activity. That's the definition of shitty targeting (false positive).