4

Here's a thought experiment that I can't come up with a good answer for.

Imagine you have 5 YouTube videos that you want to reach 50,000 views (10,000 views per video) by the 10th day after publishing the videos on YouTube. Of course, you have no idea how these videos are going to perform so you've set aside some money to buy views on Google AdWords to ensure that by day 10, you have at least 50,000 views total (organic is preferred, but you're willing to pay for views).

The main requirement/constraint is that you want the purchased views to look "organic." (Imagine making a bet with a friend that your 5 videos will organically hit 50,000 views. Huge spikes at the end of the video from buying views will look suspicious). So, while you can buy views at the very last hour, the spike at the end would look odd and your friend would not believe you had done it organically.

Any view that you buy above 50,000 is wasted money because you've already hit your goal. You can buy views from Google AdWords for .01 cents- .15 cents; since buying views is auction based, the final cost per view will depend on your targeting criteria and other factors. For simplicity, assume a fixed cost per view of .05 cents.

What statistical concept would you use to buy just enough views per day to ensure that by day 10, you are at or over 50,000 views while minimizing the number of views you have to buy (you don't want to buy views if you don't have to!). How would you spread out the paid views by day to ensure that at the end of day 10, both paid and organic views add up to at least 50,000 between the 5 videos? What kind of statistical model would help answer this question?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
MetroidPrime77
  • 41
  • 3
  • Nothing in this model until the last sentence gave any reason to do anything other than wait until the last hour and buy enough views to hit 50k. I think, then, that you need to clarify what "want it to be spread over the 10 days" means a little more to be able to answer the question reasonably. – Danica Mar 30 '15 at 08:02
  • 1
    Hi Dougal, I've added detail to the constraint. – MetroidPrime77 Mar 30 '15 at 16:30
  • 2
    This still doesn't read like a statistical question. It is framed as an optimization problem--which is good--but it's hard to see yet where "statistical concepts" would play a role in *answering* the question until some statistical concepts are first used in *framing* the question (which indeed could be done and might be fruitful). Furthermore, it's still incomplete in the absence of a *quantitative* characterization of all the elements--especially of how much of a buying spike would be "odd". – whuber Mar 30 '15 at 17:00

0 Answers0