Increasing sales - or usage of a product - doesn't change the probability of a failure or the the occurrence rate. It changes the NUMBER of failures.

IF harm is not a certainty of a failure but is a proportion of occurrences then the same math applies. increasing sales/usage will increase the NUMBER of events that result in harm. not the probability.

This can of course increase the cost of dealing with the failures as that is tied to the number and not the proportion of sales/uses. It can I suppose increase the 'probability' that the Customer base or the media or facebook hears about the failures and gets mad at you.

I think the numbers matter, over and above simply media, facebook, or company reputation/cost considerations. We are, afterall, dealing with real-world harm to individuals...

I take your point

@yodon that you have no control over distribution, but assuming a random distribution, if a failure mode that could result in death has a probability of 1/100000, then if only 10 units are sold, the probability of this manifesting is remote, whereas if millions of units are sold then it's an almost certainty that someone is likely to die. In this sense we are talking about the probability that

*someone* dies, no?

Could it not be the case that the former case is deemed acceptable, but the latter is not?