[rwissmann]

They likely already record and store all the necessary data (and more) in order to build out people's listener profiles.

Pools of money is not really a relevant metric. They collect all the subscription premiums and pay a portion of them out according to their contractual agreements.

Pay(artist_i) = 0.7 * sum_j [ premium(viever_j) * ratio_ji ]

is not that much harder to do.

I agree that auditing would be more complicated, but usually one does selective audits anyway. I doubt this would be a show stopper.

The confusion on the receivers' side would increase massively though. As someone else has pointed out, there would be a lot less correlation between #plays and payout.

(To be clear, I have no real opinion on whether this would be a better schema.)

[exelius]

Actually, it is computationally much more difficult. You're masking another step in that the "ratio_ji" has to be calculated and stored for every artist-subscriber combination, which is significantly more complicated than calculating a percentage overall. Since they are almost certainly using map-reduce to calculate these metrics, you're basically running a nested map-reduce. I mean, it's not an unsolvable problem, but it takes a non-trivial amount compute resources that aren't free to solve a problem that isn't really a problem.

But yeah, there are many reasons its not calculated this way. Revenue predictability is another reason - companies like to be able to forecast revenue and that's hard if revenue has only a tenuous link to the other KPIs you use to run your business (which is the point you made).

Honestly, the current payout scheme makes more sense. Is it totally fair to the labels? Maybe not; but the labels it's most unfair to aren't making enough under either scenario for it to matter much. Would it really matter to an indie label if their monthly payout was $20 instead of $5? Once you get bigger than a few hundred listeners, the law of large numbers kicks in.