You may hear the phrase ‘predictive analytics’ used a lot these days. You may have an inkling that it is related to business intelligence, statistics or even machine learning and artificial intelligence. 

You are right. Predictive analytics is related to all those other buzzwords.

So, you may ask, “What is the difference?”

Well, all these technologies overlap in either limited or very substantial ways, but the main difference is between seeing into the past and trying to see into the future.

For instance, business intelligence can often be used to find the answers to very important questions, such as:

• What was the unit revenue for last month for Denver?
• Where did costs increase the most quarter-to-quarter?
• Is our new product release having faster or slower acceptance that the last release?

Questions that predictive analytics can answer have more to do with the future:

• Which customers are at risk of leaving for the competition?
• Which email message is most likely to prompt a click from a recipient?
• When is the best time for me to purchase raw materials for my product?

Predictive Analytics is Similar to How You Make Predictions

How these systems go about predicting the future is not that mysterious. It works very similarly to how you and I would do it on our own.  If you wanted to predict, for instance, whether an email message was going to have impact and cause a purchase from your customer, you might look at the results from the last email that you sent to similar customers.

The trick is that things change and just looking at the past is not a perfect predictor of the future. Your last email campaign may have a number of differences from your current one:

• The timing is different (e.g. late October may have different responses than a month later, during the holidays).
• The recipients may have already received the first email and may not be as receptive to hearing the message again.
• Or just the opposite—the recipients may have already received the first email and may now have just enough brand awareness to take action.

So, as your financial advisor, always tells you, “Past performance does not guarantee future results.”. To make the prediction as accurate as possible, predictive analytics gathers as much information as possible from the past in order to predict the future. Though never perfect, this can be very valuable to any business.

Predictive Analytics Works by Looking for Patterns in Your Data

Tools that fall under the moniker of ‘predictive analytics’ use all sorts of different algorithms and technologies to find patterns in your data. Some use neural networks, others use decision trees or cluster analysis, but the bottom line is that they are all doing basically the same thing.

• They look for patterns in the database that seem to be related to the behavior you are trying to predict.
• They test those patterns to see if they hold up under lots of changing situations (i.e. Are they robust?).
• They use the most robust patterns to make a prediction, weeding out the ones that are spurious or purely random.

Why Do You Need to Choose the Most Robust Patterns?

The main reason you need to choose the most robust patterns is because the other things you find will not be as predictive. For example: my twenty-five year old daughter joined a fantasy football league at work but knows very little about the game of football or its players. So she selects players for her fantasy team based on how cute they look. During the fantasy season she absolutely kills it and wins the pot at the end of the season. Everyone is amazed and asks her how she did it. When she reveals her selection strategy, none of the veteran fantasy team players emulate her strategy the next year (even though it was the best strategy this year).

Why didn’t those veteran players use my daughter’s strategy? Well they knew from year to year that this was a strategy that was unlikely to keep working. They had past experience, and they could look back and see that, while the pattern of good-looking players to fantasy winners worked this year, it would not have worked in the past.

What the veteran fantasy football players do naturally as just good common sense is the same thing that good predictive analytics tools perform for their users.  They weed out the patterns that might have worked one time but don’t seem to work continuously, and they do that by testing any predictive pattern on all of the data that they have.

When Should You Use Predictive Analytics?

The simple answer is that you should always be using predictive analytics. At its best, it can tell you exactly what to do next (like in programmed trading on the stock market). At the minimum, it can provide you with just one more piece of valuable information as you make an informed decision (just like a trusted human advisor that you can consult and either listen to or not).

What is the Challenge with Using Predictive Analytics?

The biggest challenge of using predictive analytics is in finding a way to make it easy enough to use by the business user who understands all the facets of the business.  This problem is being solved today however with self-service predictive analytics offerings from vendors such as RapidMiner, Alteryx, Angoss, FICO, IBM, KNIME, Microsoft, SAS and others.

Predictive Analytics Can Be Your Competitive Advantage

The bottom line is that with the large amount of high-quality data that is created and captured by today’s businesses and the ease of use of new predictive analytics tools, you should be using predictive analytics. Start by looking for two things:  1. A problem where small improvements can make or save a lot of money. 2. The presence of lots of high-quality data.  Then look for an approach that matches your level of skill (i.e. hire someone if you’re not sure what you are doing).

 Just like many things in life, utilizing predictive analytics may seem hard at first, but like making pancakes, after the first couple of tries, you’ll get the hang of it.