Friction in a purchase flow does not raise conversion. It lowers it — that part is not in dispute, and any team that has run the test has watched the number fall. What friction can buy is information, and the trade is worth making only when the information is worth more than the volume. A step that costs every prospect the same thirty seconds tells you nothing about which of them was worth keeping. It simply costs you the ones who were closest to leaving anyway.

Michael Spence's 1973 job-market signaling model gives the condition that has to hold. A costly action carries information only when the cost falls unequally on people who differ in the trait you cannot observe — in his case, schooling that is genuinely less costly to complete for the more productive worker. Public economics runs the same instrument in the other direction and calls the result an ordeal mechanism: Nichols and Zeckhauser argued that a deliberately inconvenient application process can target a benefit better than means-testing, because the time cost of queueing weighs most on the people whose outside options are best. Both are the same device. Impose a cost the type you want pays easily and the type you don't cannot justify.

The operational test for any step you are considering is whether its cost correlates with the thing you cannot see. Three questions about intended use cost a serious buyer almost nothing and cost a casual browser their remaining interest, so they sort. A mandatory sales call sorts as well, but on company size and calendar slack rather than willingness to pay, which is why it fills a pipeline with the wrong meetings. An extra password confirmation sorts on nothing at all and is pure loss. Volume is the price of the screen, and a screen pointed at the wrong variable is a price paid for nothing.