I’m standing in line this morning waiting for a cup of burnt, overpriced coffee from a Dunkin Donuts pushcart in Penn Station in New York City. I can’t help but to hear the gentleman behind the cart say to every still drowsy New Yorker… “yes sir.. large or extra large?”

I did consider for just a moment that they only had the two sizes… this IS New York City… but when he asked me I said medium. He flashed me a big smile and handed me a medium coffee. Indeed there are still four sizes at “Dunkin.”

I also thought for a moment about a meeting I had later that day at a large brand retailer we do business with. We’ve been discussing implementing a collaborative filtering technology. I asked the vendor to come in a few times, each time he drowns the poor client in the mathematical models they implement to make recommendations to the consumer of additional products they are likely to be interested in.

In many collaborative filtering engines there is a tool where the marketer can seed the application with “hints” that tweak the algorithm to increase the liklihood of making recommendations that are more important or profitable for the marketer.

These tweaks proved a substantial increase in sales online over the control. Perhaps we had learned something from the coffee guy in Penn Station that day.