A flaw of the freemium model

One of the many things I have been debating is whether or not I want to do a freemium model for Comp Versions. I am not a big fan of giving away too much for free, mainly because I am a one-man shop and have extremely limited resources. One of the major benefits of freemium is the idea that free users will convert to paying over time. This works for some services, where the cost/pain of switching is too high/great. Evernote, where you store all your thoughts and jottings – has been famously successful with converting users – because by the time you exceed the limits on the free account, you can’t imagine leaving so you upgrade. The same thing applies to Dropbox – for file sharing and syncing across multiple machines.

However, the folks over at 37Signals don’t seem to agree. They have found that regardless of how long users have been members, they still don’t convert.

All of this stuff is well known.

What just occurred to me the other day though, is that not only is the free service expensive to offer from a Cost of Goods Sold perspective ( in that users use bandwidth, storage space, etc. ) – but it is extremely important that in order for the users to convert, they need to receive good service. So that adds another cost, which is even more costly because it involves human contact. I realized this after a bad experience with Heroku. This isn’t a rant post, but it occurred to me that I am just a free user, so technically they shouldn’t really deploy too much resources to support my requests. They need to tend to paying customers. But this is the inherent trap/flaw. If they don’t allocate enough resources to make sure I am pleased with my experience (especially my first experience), there is a lower probability that I will become a paying customer.
Who would hand over their money to a service that treats them like crap? Granted, in my case, I complained about the quality of the service and a wonderful Support Manager – Christopher Stolt to be exact – came to my aid and apologized for the service failings. If I were not building my own app and thinking about quality of service, etc., I would have not complained but just taken my business to Engine Yard or rolled my own VPS – without mentioning anything. But I did and got better service.
So this begs the question, if your future new paid users are your current non-paid users, how much resources do you allocate to them? I don’t know yet, but what is obvious is that supporting free users like this drains on the precious resources and gross profit margins of the app/company. Especially given that a significant portion of the free users will never convert. Evernote, the wild success of converting users from free to paid over the long term, is only able to convert 20% of their free customers to paid in the first 2 years. That leaves them having to deploy human support resources on, say 50% – 60% of free users that will never convert.
Talk about throwing good money after bad!