4 ms·

I have reasons to believe that Facebook earnings correspond to unstable practice: namely, both brands and games have bad metrics (i.e. uninformative) of their a

by bertil 13y ago

I have reasons to believe that Facebook earnings correspond to unstable practice: namely, both brands and games have bad metrics (i.e. uninformative) of their activity, and both are re-considering how their price fan- and user-acquisition. There is a stunning lack of understanding of social network dynamics among Facebook client companies. I have been through a lot of interviews lately, and no one to whom I talked seemed to know basic things like the Friend paradox, avalanche thresholds, or that the first fans are naturally more engaged -- Facebook ad purchase interface doesn’t really help either. Nothing about game session duration, rhythm? Nothing about the thousands of odd Pakistani accounts liking the page of random family restaurants in the North of Ireland? Some people at Facebook know those very well, but as far as I can tell, no one has connected that issue with a Wil-E-Coyote moment: strong sales out of momentum, filling an inflated inventory but widespread skepticism. There is indeed far more ads in the Facebook mobile thread, but almost exclusively for miss-targeted offers (I recently moved to a different country, twice, and I keep seeing things for local apps in a city 2,400 miles away) and for mobile games with Zynga-like gameplay: pay or… well, you can pay too. I’m not sure Facebook is aware of that; I am sure however that Zynga wasn’t, or at least that the people in charge of the financial stability of the company dealt with that issue in an unethical manner. Seeing both a trend and someone to chosen to reinforce it scares me.

KaoruAoiShiho 13y ago
Is this what you're talking about? http://arxiv.org/abs/1401.4270 http://arxiv.org/abs/1401.4270
- rajacombinator 13y ago
  yea I'm curious about "avalanche thresholds" and how it is a basic thing people should know about. (seems like gobbledegook from my outsider perspective.)
  - bertil 13y ago
    Well, that's how avalanche work in the mountain: you don’t really get any less gobbledegook than pack of snow crushing you or not. In the case of social networks, the idea is to make sense of how people start adopting a behaviour. For that, you have many models, from ‘susceptibility’: a user adopts if it pleases her, as soon as any of her neighbour adopt it; to avalanches: one can’t tell the benefit without using it, so, as a proxy, they use a number of friends using it as a threshold. There are variants, including where users can stop using it; a more obvious one is using the weight relations; a more significant is considering the structure of the network: can one potential consider that their friends using it might be part of the same group (a ‘facet’ to re-use Field’s vocabulary) and does it encourages me to adopt or no, on the contrary. When you talk to people about adopting trends, they usually explain that “every body is doing it”, but if you dig using several examples, “everybody” ends up being a form or another of those model. It might be ‘influencers’ that may of may not be a stable or identifiable group, too. If you are concerned in targeting ads to the right people, understanding which case work for you is key. This is why Facebook tried (controversial) ads indicating that some of your friends were using the service already. But even without that creepy factor of expressly indicating that, you might want to measure whether you have better conversation rate for people with no friends, one friend, two or three, four to nine, ten to twenty, or more than that already using your service. Cases include situation where those friends might know each other, or be part of clusters or cliques (respectively, groups where everyone knows almost- and everyone else). That would allow you to target your advertising to the specific behaviour of your service. I’ve been studying that for the past ten years (so not exactly when Facebook started, but initially on other networks); since, I’ve made considerable efforts to make a complex and counter-intuitive discipline clear enough for any marketing intern to grasp it; I’m not sure “gobbledegook” is exactly how I want it to be described to me.
prostoalex 13y ago
59% of their advertising revenue is off mobile, where their main products are app installs and click-through ads. They don't break it down, but an ad product boosting likes to a business page that you're describing as highly susceptible to fraud is probably not a best-seller, and isn't even offered on mobile.
- bertil 13y ago
  Those are two distinct manifestation of potentially different issues. Mobile is over-invested because there is a presumption of high user value that isn’t adjusted with history. Organised click-farms use non-commercial links to avoid detection. There is a lot to be said about Facebook detection technology there.
  - prostoalex 13y ago
    What I'm saying is that desktop like ads are non-essential for Facebook's revenue figures, because hardly anybody buys those ads. YouTube had a similar issue when they started paying out content producers based on views, and all of a sudden "view farms" appeared out of nowhere, generating required thousands of views. It's a big deal if you buy ads on YouTube and get charged per views, but in the grand scheme of things it's a drop in the bucket for Google Inc.

by bertil 13y ago

KaoruAoiShiho 13y ago

Is this what you're talking about? http://arxiv.org/abs/1401.4270 http://arxiv.org/abs/1401.4270

rajacombinator 13y ago
yea I'm curious about "avalanche thresholds" and how it is a basic thing people should know about. (seems like gobbledegook from my outsider perspective.)
- bertil 13y ago
  Well, that's how avalanche work in the mountain: you don’t really get any less gobbledegook than pack of snow crushing you or not. In the case of social networks, the idea is to make sense of how people start adopting a behaviour. For that, you have many models, from ‘susceptibility’: a user adopts if it pleases her, as soon as any of her neighbour adopt it; to avalanches: one can’t tell the benefit without using it, so, as a proxy, they use a number of friends using it as a threshold. There are variants, including where users can stop using it; a more obvious one is using the weight relations; a more significant is considering the structure of the network: can one potential consider that their friends using it might be part of the same group (a ‘facet’ to re-use Field’s vocabulary) and does it encourages me to adopt or no, on the contrary. When you talk to people about adopting trends, they usually explain that “every body is doing it”, but if you dig using several examples, “everybody” ends up being a form or another of those model. It might be ‘influencers’ that may of may not be a stable or identifiable group, too. If you are concerned in targeting ads to the right people, understanding which case work for you is key. This is why Facebook tried (controversial) ads indicating that some of your friends were using the service already. But even without that creepy factor of expressly indicating that, you might want to measure whether you have better conversation rate for people with no friends, one friend, two or three, four to nine, ten to twenty, or more than that already using your service. Cases include situation where those friends might know each other, or be part of clusters or cliques (respectively, groups where everyone knows almost- and everyone else). That would allow you to target your advertising to the specific behaviour of your service. I’ve been studying that for the past ten years (so not exactly when Facebook started, but initially on other networks); since, I’ve made considerable efforts to make a complex and counter-intuitive discipline clear enough for any marketing intern to grasp it; I’m not sure “gobbledegook” is exactly how I want it to be described to me.

prostoalex 13y ago

59% of their advertising revenue is off mobile, where their main products are app installs and click-through ads. They don't break it down, but an ad product boosting likes to a business page that you're describing as highly susceptible to fraud is probably not a best-seller, and isn't even offered on mobile.

bertil 13y ago
Those are two distinct manifestation of potentially different issues. Mobile is over-invested because there is a presumption of high user value that isn’t adjusted with history. Organised click-farms use non-commercial links to avoid detection. There is a lot to be said about Facebook detection technology there.
- prostoalex 13y ago
  What I'm saying is that desktop like ads are non-essential for Facebook's revenue figures, because hardly anybody buys those ads. YouTube had a similar issue when they started paying out content producers based on views, and all of a sudden "view farms" appeared out of nowhere, generating required thousands of views. It's a big deal if you buy ads on YouTube and get charged per views, but in the grand scheme of things it's a drop in the bucket for Google Inc.