I know a spell
That would make you well
Write about love, it could be in any tense, but it must make sense
Belle & Sebastian – Write About Love
Some companies believe that product management can be reduced to scientific experiments.Instead of using intuition and customers interaction one should run experiments and measure results.
While I have great faith in measurable product management , I think that the dream of product management without the human factor is wrong and dangerous.
Everyone seriously involved in pattern recognition and data mining knows that one can’t just throw tons of raw data into an algorithm and expect to gain (artificial) intelligence.
In most cases it is hard to build a large data set to train the algorithm. Once such data set is built , the raw size is too big for any algorithm to train on. As a result ,the raw data needs to be reduced through feature extraction. For example, if we want to build a face recognition algorithm in a video stream we can help the algorithm by removing the soundtrack. While in theory the soundtrack can add information to the algorithm, we guess it is not very helpful.
The process of feature selection and even dataset selection involved intuition and domain knowledge. This is similar to the generic scientific model.
In 2008 wired magazine claimed scientific method is obsolete in “The End of Theory: The Data Deluge Makes the Scientific Method Obsolete“. The article claims that models are not needed anymore, as data is stronger than models.
However, even the first example is wrong
Google’s founding philosophy is that we don’t know why this page is better than that one: If the statistics of incoming links say it is, that’s good enough
But Google’s early success was not just because of the algorithm. The clean UI,text only ads and great performance were crucial. I’m confident it was intuition\product management that led into these decisions. Moreover, the statistics for incoming links from fraud (link farms) are also very high. The algorithms needs “help” on the features that identify fraud.
Google Suggest brings another example. It is a great feature which interactively “guesses” the search term for the end-user.
For example, typing the word “Robert” suggests the following :
But looking for the word “Naked” brings no results at all:
But “Nak”still shows some alternatives:
Does anyone think the algorithm decided on this feature based on statistics ?
Obviously, someone decided that following the real statistics of the human mind would be too dangerous.
I’m very much in flavor of usability research and detailed numerical specs. But in most scenarios, the psychology, human interaction and models are crucial for a building a great product.
- Can Google’s algorithm do subjective quality calls? (zdnet.com)
- “Google Updates Social Search Algorithm” and related posts (beanstalk-inc.com)
- The scientific method is alive and well (blogs.discovermagazine.com)