Can You Build a Better Paper Clip?

Doesn't this suggest a certain room for improvement? For example, choosing a different metal might enable the construction of less-expensive paperclips, or perhaps corrugating the wire (like a hairpin) might enable the clip to retain its strength while using less material in its construction.

Certainly, you can use different metals, and you can use strengthening techniques that allow less metal to make a paperclip of the same strength and fatigue life. But it has to be some kind of metal, it has to hold several sheets of paper together, and it has to have standard shape.

Just because unclippy agents sometimes try to make better paperclips doesn't mean that clippy agents shouldn't try to do so.

Right, but these new designs aren't "better paperclips"; they're not paperclips at all.

The mental back-of-envelope calculation went something like this:

11B paperclips per year sold in the US, most made in the US; Officemate are one of the two dominant US paperclip vendors; let's say 3B/year. They sell for "about a penny" each, so let's say $30M in revenue; I don't know what margins are like in the paperclip business, but let's say $3M/year in gross profit. Usage and sales of paperclips probably aren't all that unstable from year to year; let's use a horizon of 5 years and say that there's $15M of profit at stake here. A 1% improvement in sales or profit margin is therefore worth about $150k, and anything that offers such an improvement at less than that cost is worth while.

Now, how do you find out what's happening to all those paperclips? Well, I'd have thought you try to identify the really big users (which you might do by talking to your largest customers) and to get a handle on what typical users do (which you might do by surveying random people).

I don't have a very good sense of how expensive this sort of survey is, so here's some further back-of-envelope stuff. Suppose that to find 100 people who have used paperclips in the last week and are willing to talk to you for a few minutes you have to trawl through 1000 people and take about 2 minutes for each; and suppose that your 100 qualified people take you 10 minutes each. Then that's 3000 minutes, or about 50 hours, for 100 paperclip-users. Perhaps you need to pay $20/hour to the people operating the survey (for the people themselves, or the use of whatever facilities they need, or whatever), making $1000 for those 100 paperclip-users. And perhaps you need to give them some incentive -- say $10 each -- to participate in your survey. That's $2000 for 100 paperclip users. So for $50k you can survey 2500 people. (Only $50k because I'm assuming that some of the $150k goes to investigating the biggest users, that some of it goes to pay the clever-and-expensive people who help you make sense of the results, and that some of it is left as margin to reduce the probability that you end up making a loss on the exercise if the information you get really isn't much use in optimizing the design, marketing, etc., of your paperclips.)

2500 people seems like a lot more than you'd really need to get a decent picture of how your products are being used. (Of course you wouldn't, in any case, survey that many people. You'd start with a smaller number and use what you find to guide your next round of investigation.)

For the avoidance of doubt, my actual mental back-of-envelope calculation was less detailed.

Anyone whoever doubted that is an idiot.