Insurance is a very bad example because it is predicated on not knowing who will be the outlier. If you know who is the outlier you can just charge them more m

Insurance is a very bad example because it is predicated on not knowing who will be the outlier. If you know who is the outlier you can just charge them more money directly. The numbers matter here but when I say below cost, I mean that the average price paid by the outlier is less than if they were paying their weighted share. Example: Let's assume I'm selling something such that by selling one unit at $1 I make a reasonable profit. If I have 100 subscribers and 95% of those subscribers use an average of 5 units per/month but 5% use an average of 50 units per month, I have to provision for 725 units/month (955 + 550). If I charge everyone the same monthly rate, I would charge $7.25/month per person to get my $725 of revenue. But 95% of the people are only using 5 units. They are paying almost 50% more than actual usage would justify. This is because they are subsidizing the 5% who are each getting 42.75 units subsidized for 'free'. Their monthly rate is 'below cost'. I doubt the real-world data is bimodal in exactly this way but I think this illustrates what I was trying to communicate. I did find this interesting quote from http://www.techhive.com/article/2055502/why-your-internet-service-provider-wants-to-nix-all-you-can-eat-data-buffets.html http://www.techhive.com/article/2055502/why-your-internet-se... > According to Comcast, the current median monthly data usage for its customers is roughly 16GB. AT&T claims that the top 2 percent of its users (in terms of data consumption) eat up 20 percent of its network’s total available bandwidth. Telecom analyst Teresa Mastrangelo, principal at Broadbandtrends, suggests that the 80/20 rule applies, in that 80 percent of the data is going to 20 percent of users. “It may even be closer to 90/10,” Mastrangelo says.