socialdna, predict the present, twitter bombs, statistics, sampling

Predicting the present

We use the oft-repeated quote

Prediction is very hard, especially about the future

in lecture and on its face it seems like a ridiculous assertion---what else is there to predict besides the future?

However, there are lots of situations where the future, in terms of available data, is simply what has just happened. I forget frequently, but most things are not instantaneous. It takes time to compile box office receipts from all of the different theater chains across the country. It takes time to poll voters on the telephone (most survey agencies even only started calling cell phones!). It takes time to survey and identify how many people have a job (heck even defining what is a job is something that must be evaluated).

The process of collecting this information creates a lag. Even if it's the January jobs report, then that means the data is really December, November, or October's jobs report. This lag exists in most places, it just comes down to a question of how big is the lag. If we're talking about the stock market high-frequency algorithms compete at microseconds and fraud reporting at credit agencies is on the order of a second, these processes and others at these time scales will be dominated by (predictive) algorithms. Other areas have a much longer lag, especially those that require surveying the population (politics, epidemiology, etc.).

But what new tool lets us accerelate the data collection process? The Internet

The same tools that allow us to influence each other more quickly than ever before, making it even harder to predict what will be a run away success, are also the same tools that let us monitor the system in real-time. However, the internet is not the only way. You can of course use the data that your own business collects so long as you devise a way to make that data collection valid.

This is why Duncan Watts' article in Nature is so pertinent to this lecture. The internet has truly brought social science into the 21st Century and made it possible to quantify effects like social influence in a way that was impossible 20 years ago. In his book, Everything is Obvious Once You Know the Answer, he even goes so far as to say that the internet is to social science as the telescope was to astronomy.

The Measure and React strategy

Fashion, an industry marked by fickle preferences (made even more apparent if you've ever found yourself shouting at the judges on Project Runway), is about predicting what designs will resonate with your clientele. Even at the low end, you need to decide if a Family Guy t-shirt with the evil monkey or a Peter quote is more likely to fly off the shelves. Making a wrong prediction results in large stock on shelves that must be sold at a heavy discount and lost revenue during the time period while you increase production of the most popular items. Even worse, fashion is a highly competitive industry so it may not even be possible to switch production and get more stock of an item out while it's still in season.

It's also an extremely interesting business sector right now because of the intense amount of new competition and different business models being employed. At one end of the spectrum we have Japan's Uniqlo which specializes in bucking trends, selling basics (t-shirts, jeans, socks) which has a relatively small number of styles in inventory (~1000), which allows it to achieve large scale and competitive pricing in manufacturing. At the other end, we have Sweden's H&M and Spain's Zara who both specialize in 'fast fashion'-capitalizing on selling new trendy pieces at an affordable price.

However, the two companies have gone about it in two different ways. H&M has focused on partnering with known designers to elevate its brand and recognition. On the other hand, Zara has focused on solving the problem of predicting trends through its business practice.

Zara brought manufacturing in-house, which is both more expensive and at odds with the practices of most of its competetitors. What this additional cost allows for though, is rapid production of new items something that is infeasible when dealing with contract manufacturing partners. This ability to rapidly produce items lets Zara practice what is called the Measure and React strategy.

This strategy is relatively simple, Zara makes a relatively large line of pieces in small volumes that it thinks will be trendy in the coming season. Once the new line is stocked in the stores Zara monitors which pieces are moving off the shelves and puts higher volume orders for those pieces directly into production (possible since its in-house). Not all of the pieces in the line are popular, but since all of the initial orders were low volume even unpopular pieces don't saddle the retailer with significant unsold stock of undesirable items.

So does this change really pay off? It absolutely does for Zara, even with the increased cost of manufacturing. There is a great piece over at Edited about the difference in sales between Zara and H&M. We see that a key difference emerges when we examine how much of a discount is typically applied to move stock off the shelves finally. H&M currently has ~24% of their stock discounted on-line, Zara only has ~3%. Using a measure and react strategy Zara is able to decrease the amount of items they must discount by 87.5%.