Not a nitpick, beacuse the cost to enter the data by the user is also not zero (thus supply is not infite either). These are subtle but not trivial things to keep in mind, when dealling with massive scale (A billion users, ect.)
For something like a list of a million user names and email addresses? You put it on pastebin and set up a script to email out links to it when you get a Paypal payment confirmed email. The only cost is to acquire the data, once that is done, there is zero cost.
If you want to talk in totally abstract terms, digital goods in general tend to have marginal costs associated with them. In the context of this discussion, there is no supply and demand factor, there are no marginal costs, and there is no market force called scarcity.
That's not what marginal cost means to a supplier. The question isn't whether it costs more to acquire 2000 email addresses than it does to acquire 1000 email addresses, the question is whether it costs more to distribute to twenty buyers than it does to distribute to ten.
Thus, the cost of hosting is a marginal cost (probably zero in this world of pastebins and digital lockers). The fee taken by the payment processor is a marginal cost. The cost of finding twice as many emails is not.
No, a "supplier" has to pay for all of his raw materials costs. That includes inventory costs as well as distribution. Of course you can always restrict your timeframe and assume away this cost (inventory as already incurred), but this is not true in the general sense. In particular, if this is true, by assumption, the there is a limited supply by deduction. If you increased your supply [of information bits, not duplicate bits], you would have to pay to incur inventory at that margin precisely. So you never have together zero marginal cost and unlimited supply, this makes no sense.
notatoad 1 day ago | link
I don't think that rule applies for digital goods where the cost of reproduction is zero. The supply is infinite.
To sum, "the cost of reproduction" is <not> the cost of "supply", unless the supply is assumed fixed. Thus the second sentence does not follow per-se.
I don't think you are understanding. Of course there are big costs in acquiring more product to sell. The question is: Do you have to pay those costs for each customer, or can you pay them once and amortize the cost over many sales?
For example, Adobe Photoshop probably costs a lot to design. It has really high fixed costs, because you need to hire good developers and implement a bunch of advanced operations. However, once Adobe pays the fixed costs, the marginal cost of Photoshop is pretty minimal: packaging, printing a DVD, maybe some marketing. It still costs a lot because the fixed costs are so high, and there's not much competition.
Conversely, a plumber has relatively low fixed costs: a truck, some tools, and some training. But plumbers also cost a lot, and this is because they have really high marginal costs: they have to spend an hour at the house of each and every customer.
So I agree with you, there may be high costs in acquiring email addresses to sell. My point is that they are in no way marginal costs.
The costs are marginal at the point of periodicity.
example: reseller> pays adobe every month/quarter
example: adobe> pays versioning costs every 24 months
Provided you shrink the window of analysis, you can say "already paid for inventory, just amortizing it". But in that case, you don't have unlimited supply, you just have whatever you paid for.
In the case of adobe, despite having "unlimited copies" of CS5, they would (eventually) run out of supply of salable product if they did not version into CS6. So while its trivially true they could make unlimited copies of CS5, its not a great idea to perceive this as unlimited supply. The supply that matters is the part people are willing to pay for--this is the marginal information content-- not the marginal bit content of what is delivered.
In some ways I don't think we're disagreeing, just focusing on different elements of the analysis. My larger point was exactly that -- keep in mind the broader elements that are considered as relevant by CxO.
THe CEO of adobe makes decisions, for examople, about how often to incur the marginal cost of versioning the next Creative Suite, how rapidly and how much to budget, etc. COO of facebook looks at the marginal cost of data centers for the next 200 million users, etc, in part because s/he is looking at timeframes and scales which are not the same at the level of a project team, etc.
Care to explain a bit more? On the face of it, I don't believe this can make sense. All costs scale, all costs are at some stage marginal. What are you proposing is the trigger or cause of incurrence?
I think the case they are describing is where the marginal cost is highly nonlinear and the price delta between two reasonable values is so negligible that the marginal cost isn't meaningful.
A 2MB file does not cost twice as much to email send to someone as a 1MB file; you aren't going switch to a different internet connection or email provider because of your file is twice as big. The first 1 byte is very expensive and every subsequent byte has no observable marginal cost until 20 orders of magnitude later.
Appreciate your comment. I think I was unclear earlier in my post. Was not trying to talk about the supply of > undifferentiated bits per-se. Those are trivial to scale, in minor orders. The micro cost is in the acquisition of <unique> bit sets. In your example, its not the cost to send or replicate the e-mail with a 2 mb attachment. Its the cost to acquire, verify, etc the contetents of the 2mb file have any value. [1]
You can't create more <valuable> data per-se by making X copies of the same data (in the sense of it having value for marketing/analytics). that is just monetizing existing data. ie, The marginal cost to relicate a set, provided it was given to you for free...just assumes away a non-trivial part of the equation....getting the data.
At scales of 100m to a 1Billion...is not trivial or costless. Lastly, if you only have one set of data (say 5m users of data), that is a finite supply. You could have 2 sets (10m users). Thats not the same as having 2 copies of 1 set (of 5 million). A customer might pay per user for a lead, but wont pay twice for two copies of the same info. now, if somebody shows up with 100s million, it might impact supply/demand (depending on comparabilit/uniquenss). But those differences cannot be assumed away at zero cost, imho. Hope this make more sense, was not trying to argue just for the sake of it.
________________
[1] That scales with data entry, etc (if nothing else) at the origin (ie, this is a FB user cost == per user). Even if its non-cash its ~$0.85c per 15 minutes of time for a western eurpoean ABC1, back of the envelope. And that scales linearly.
Thanks for the reply; I honestly am not sure about this situation in particular, I was mostly trying to explain how the statement about marginal costs can be true in any scenario.
I didn't actually intend to say that the cost here is mainly bandwidth or X for any X. My point is more than it can be true that the marginal cost of anything can be so many orders of magnitude less than the non-marginal cost for certain ranges that it isn't worth considering. Data sets based on facebook profiles almost certainly come from people approving shady apps or someone set up a crawler in a way that is able to get a lot of information before being detected as a crawler. In either of those scenarios, the person who set it up effectively paid a flat upfront cost and ends up with X number of users, and there are no linear costs (no manual verification or paid data entry at any point). They cannot spend half as much time and get X/2 users or even maybe X/100 amount of information.
In practice they could spend more time getting more users to give access to their random app, but the marginal cost function is just insanely nonlinear; its effectively 0 at some places and probably tends towards infinity an order of magnitude higher than that.
