A Rainy-Day Guide to Predictive Analytics

Small talk always seems to start with observations around the weather.  So, as we attempt to boil down a complex topic like the applied machine learning techniques to develop predictive analytics, let’s start with the weather.

Photo courtesy: PIxabay

Photo courtesy: PIxabay

A simple thing like a rainstorm can be a great example for how predictive analytics will change every day life.  So, let’s get wet!

Descriptive analytics: “It rained last week.”

Your traditional reporting solution does a good job at showing you a specific event.  It’s great for telling you what happened, but doesn’t provide much information on why it happened, or more importantly, why it might happen again.  That’s great for showing historical data, but this method has left it to the analyst to recognize trends, find patterns and turn that historical data into action. 

That’s great for knowing it rained last Tuesday, but doesn’t tell you that April is the rainiest month of the year or that historically, there’s a 90% chance of rain on any day in April.   It’s just another data point in what ends up being a guess and gut feel to predict what might happen tomorrow.

Diagnostic analytics: “Historically, it rains most days this April.”

By taking your traditional reporting solution a step further by aggregating data, reviewing trends, and drilling down into the details, you can start gleaning meaningful facts from the data.  Now, instead of a just telling you what happened, you can start to understand why it happened and, potentially, what might cause it to happen again. This allows the user to start making decisions about the future, but takes additional input and analysis to make those decisions. 

In this case, based on historical data, you know that it rains on 90% of the days in April, and because next Tuesday is in April, you can be 90% assured that it will rain next Tuesday.  This is great information to have, but still needs human input to take any meaningful actions in the future – like cancelling your golf game. 

Predictive analytics:  “There’s a 90% chance it will rain next Tuesday.”

Statistical forecasting methods review historical data to recognize patterns and use those predict future demands. It takes into account seasonal trends, correlative factors, and historical order patterns. These statistical models have gotten quite sophisticated as long as correlations are predictive of future performance.  Of course, your mileage may vary.  As a result, these statistical models need human intervention to account for unrelated data such as promotions, or competitor moves in the market.  These factors need human adjustments to account for these trends, and at some point there’s a judgment call involved.

This is like the weather forecast on the 10-day horizon.  The forecasts are fairly good these days, so if next Tuesday calls, then it’s time to cancel your golf game.    However, it’s still up to you to make the adjustments.

Prescriptive analytics: “Order umbrellas to arrive next Tuesday.”

The next generation of analytics will include the proliferation of prescriptive analytics.  Prescriptive analytics will utilize many historical data points, find intricate patterns, correlate with multiple future predictive data points and utilize that data to develop not just patterns, but also take actions automatically.   These next generation tools will integrate tightly with transactional systems to streamline and automate decision making based on machine learning and artificial intelligence (AI) technologies.

In the future, predictive analytics will review the weather forecast, identify that it will rain next Tuesday, automatically cancel your golf game, see that you don’t have an umbrella, order it and then notify you that a new umbrella will arrive the day before the rain hits.   It automates the decisions, create solutions, and leaves the humans dry and warm.

Result driven data

Yeah this all sounds great, but does it really work?  It sure does.  See the examples below:

Statistical forecasting shows the results with a wider band of possible outcomes.  In order to accommodate those outcomes there is some hedging that needs to occur, but all those hedges get expensive.

Prescriptive analytics, utilizing machine learning and AI algorithms will narrow that group of possible outcomes to a more manageable set.  These manageable outcomes allow for surer bets that allow for automation of reactions.  Those hedges seem less expensive and the decision making can be automated.

Now what?

These capabilities are not science fiction, they are here today.  With the advances in application development, it is quick and manageable to develop customizable predictive demand solutions that integrate easily with any supply chain planning system.  This includes a plug and play approach to platforms from any vendor. 

For example, SAP’s new Leonardo platform is a perfect example of future architecture of supply chain planning.  Leonardo and SAP’s cloud foundry platform allow for customizable applications to plug into the larger cloud infrastructure to allow for an easy way to provide custom solutions to any platform.

This the future and I can’t wait for our umbrellas to show up. We’re geeks that told the packages to come on Tuesday.

0 comments on “A Rainy-Day Guide to Predictive Analytics

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.