>They generated about $1B for the company by creating a portal to (essentially) sell access to the company's existing data.
You are completely discounting the cost of creating and maintaining the underlying data. You can't just write that off as zero. It also doesn't matter if the data generates other sources of revenue, that's great but you still don't get to write the cost off against one revenue stream and not another, the only fair way to account for it is to split the cost across all revenue streams.
The marginal cost of selling the data is effectively 0, since the difference between selling "zero data" and "one data" is 0, and the difference between "n data" and "n+1 data" is also 0.
The sale of data was essentially "free" to the business as a revenue stream, it cost them almost nothing extra to sell the data (again, beyond building a portal and integrations). You CAN discount the cost of data maintenance, from the sale business since the core business (also profitable) is already factoring in the cost. You can't double-count the costs, and accounting DOES let you pick-and-choose where to apply those costs.
For average cost analysis across a portfolio, yes. For marginal cost analysis, the question is “what’s the difference in cost between n and n+1 units?”
Whether other streams succeed, fail, or don’t exist doesn’t change the cost to make the n+1 unit.
That's technically accurate, but the justification for the existence of this team and product (and their real cost base) cannot be predicated solely on the n+1 cost calculation. Suppose the other revenue streams do fail, now without this specific team and product the data set generation and maintenance would be completely uneconomic. In that situation the N+1 calculation will tell you to maintain the team and product no matter how expensive the underlying data set is.
OP was implying that this product was insanely profitable because you can ignore the cost of the underlying data set. I'm saying that makes no economic sense because marginal costs aren't the only costs.
To see how absurd it is, consider that the marginal cost of an n+1 visitor to Disneyworld is probably significantly negative, because they will buy things at concession stands and eat at the restaurant thus generating revenue even if the ticket is free. What are the implications of that for the ticket price you should charge for that visitor? Pretty much nothing because the fixed costs utterly dominate. Yet Disneyworld will run promotions and discounts and bundle deals with hotel stays and even flights. None of that will make any sense whatsoever if all you fixate on is the marginal cost of printing the ticket.
A critical part of any non-trivial business situation is understanding what question to ask for what reason.
When deciding “should we make the incremental effort to expose an already existing dataset as a new product?” you are quite reasonably going to look at different data than if you want to know “overall, how profitable were our data products in aggregate last year?”
That’s the main point of the article we’re discussing: how things get weird when marginal costs are very close to zero.
You are completely discounting the cost of creating and maintaining the underlying data. You can't just write that off as zero. It also doesn't matter if the data generates other sources of revenue, that's great but you still don't get to write the cost off against one revenue stream and not another, the only fair way to account for it is to split the cost across all revenue streams.