What Does a Predictive Analytics Project Look Like?
Building a predictive analytics solution is an attractive goal when companies decide to use their data better. But it is important to first understand what a predictive model will give you that a descriptive model won't. From there, you might wonder how to choose the right problem to start with at all.
While descriptive analytics is about understanding what your data is, predictive analytics focuses on identifying facts in your data that can help predict unknowns. With the right data and problem to solve, predictive analytics can do wonders for your company. These are the tools used in finance to decide creditworthiness, in retail to target advertising, and in public health to identify disease outbreaks. While these fields also have the opportunity to use even more sophisticated approaches like prescriptive analytics, it all starts with identifying patterns in data that relate to important concerns for your company.
Typically a good candidate for a predictive analytics project is identified by people who have noticed a pattern in their data which they believe helps them predict future events. The first phase of the project is to discern if those beliefs are accurate and if there may be other patterns which help predict related future events. For example, the entire field of technical analysis for stock trading is focused on identifying patterns in price data which people believe will allow them to predict the value of stocks in the future. However, there are plenty of people who believed they had identified patterns, which in time lost them money. One of the breakthroughs for the public in the world of investing was the development of index funds which beat many managed funds simply by returning investors the average movement in the market.
Once you have successfully demonstrated that the data does behave in a predictable way, the next phase of a predictive analytics project is, unsurprisingly, to make some predictions. In most cases, those predictions do not actually need to be better than the experts who have been using the data up to this point. In fact, it is often easier to get buy-in from users if the predictions merely support their understanding of patterns rather than replacing them. That does not mean the model should simply automate the patterns that were originally identified (though that is sometimes a good first step), but your first predictive model will usually need to be understandable to the people who suggested solving this problem in the first place. This is also your opportunity to reassure everyone that more analytics will not replace workers, but instead will allow them to do their jobs better.
Some time after the first predictive model has been implemented, it becomes feasible to develop better models starting from that baseline. There are several reasons for implementing predictive analytics in phases including: the quality of the data and metrics, the engagement of users, and simply a desire to receive an initial return on investment for the project before spending more resources to make the best model possible.
At Mashey we prioritize the right process for each individual project and organization. Contact us if you would like to learn more about what we can do for you.