This is the English version of the article.
In The Lean Startup, Eric Ries advocates for small batch sizes with a simple story.
Imagine that your task is to stuff newsletters into envelopes for an advertisement. Each envelope has to be addressed, stamped, filled with a letter, and sealed. Instinctively, we think that executing each task for all envelopes would be the most efficient approach.
But research shows that that is not the case. However unintuitive, completing each envelope one at a time has been shown to be consistently faster. Apparently this has even been proven experimentally.
Why is this the case? According to Ries, we fail to account for the time it takes to sort, stack, and move around large piles of half-complete envelopes. But more importantly, there are massive process-based advantages to the single-piece flow.
For example, what happens if the letters don’t fit in the envelopes? With the large-batch approach, we wouldn’t find out until the end. With a small batch, we would discover the problem almost immediately. What if the envelopes are defective and don’t seal? We’d have to perform extra work removing the letter from each envelope, whereas with a small batch, there is no wasted work.
Software architecture benefits from this approach, too. Pragmatic Programmer has a parallel concept called “tracer bullets”. The approach involves building a vertical slice of a system in order to ferret out any technical risks without investing significant resources. Abstractly, this is simply small batch sizing applied directly to programming.
As an engineer, the notion of tracer bullets appears to me like solid common sense. And yet, I was quite baffled that completing the newsletter envelopes one at a time is faster. Somewhere along my journey through the real world, I must have decided that “business process” and “product development” are too dirty for logical thinking to be applicable.
And yet, it does apply. Experiencing this disconnect while reading The Lean Startup was disconcerting, but ultimately enlightening.