Tag Archives: Algorithm

One of the challenges any of us have in business is to predict the future.

. (Photo credit: Wikipedia)

The hardest part of my job – and maybe yours – is seeing over the horizon to help my clients get prepared for what is to come.  That might be a change in a market or it might be a change in technology.  No matter what it is, any of us who look ahead do so by gathering data.  In many cases that data is some measure of past behavior – how people bought from your website for example.  In many cases, those data points are put into some sort of algorithm which predicts what is to come.  Increasingly, many marketers and others use these models to drive their own business behaviors as the amount of data available grows exponentially.  While I’m not a believer that “big data means big problems,” a blind reliance on these algorithmic predictions can mean just that.

Let’s take one simple form of algorithm.  You probably see it every day.  it’s known as collaborative filtering and if you’re on Amazon or Netflix or any other site with a recommendation engine you’ve used it.  You may also have seen things offered to you as content on YouTube.  The algorithms use measures of your past behavior as well as of others like you (“people who bought XYZ also bought…”).  But what if you were buying a gift and the purchasing is not reflective or your tastes or interests at all?  What if someone else used your browser to search and purchase?  Cookies are browser-based – they have no way to tell if the activity is from one person or six.

Another problem.  Algorithms are built by people and those people are..well…human.  They might have confirmational bias operating as they refine the formula to eliminate noise – data that’s not germane to the prediction at hand.  The problem is that you don’t really know if it’s noise until it proves to be not significant.  Maybe it’s a new trend that your model misses altogether.

The thing to keep in mind is that modeling can only go so far.  It’s not very good  at predicting the unexpected.  It tends to ignore outliers.  As with all things, you need to ask questions, search for facts, and draw your own conclusions.  Yes, it’s impossible to make sense of all this data without algorithmically based analysis.  Just remember that while machines don’t make computational errors it was a human that gathered the data (or installed the code that does) and wrote the formula.  People often don’t compute.  Make sense?