Data Science Primer: Predictive Analytics—How to Increase Profits Using Data You Already Have

Author by Brian Goodwin, PhD

Beyond Business Intelligence
 
Forward-thinking business leaders track KPIs and use tools like Power BI to create snapshots of what’s happening in the business from moment to moment.
 
Business intelligence is crucial because it helps leaders quantify their success and visualize relevant statistics in a meaningful way. But whereas business intelligence focuses on the present, predictive analytics looks toward the future.
 
At Concurrency we take companies beyond business intelligence by employing computational techniques to arrive at forecasts, or other predictions. Our methodologies carry a statistical rigor, which minimizes exposure to human intuition and helps you see correlations between seemingly unrelated parameters.
 
Putting Big Data to Use
 
Big data is big business. While most companies don’t technically collect “big data” (the strict meaning of the term normally involves leveraging terabytes or even petabytes), it’s fair to say that most companies DO collect far more data than they know how to use. Some companies are venturing into the Internet of Things, gathering gigabytes per second from sensors deployed on just about anything. Others gather sales data, advertising metrics, customer profiles, or other types of information. We want to leverage that data. That’s where a data scientist enters the picture, making use of predictive analytics and other techniques to make your existing data work for you.
 
Step One: Risk Estimation
 
In a nutshell, risk estimation is the science of estimating probability of failure—such as estimating the risk of default for a given borrower. Depending on this probability, you may be able to make changes to mitigate this risk.
 
Risk estimation doesn’t prescribe specific actions (this is the next step!), but it’s essential to wise decision-making. For example, risk estimation may help you identify overexposure to particular types of risk. It takes factors like credit exposure, business disruptions, and supply chain issues into account to help you balance profitability with risk exposure.
 
Historically, accurately estimating risk has been difficult because big data computational tools haven’t been available until recently. Most companies have “data warehouses” with mountains of unused information. Data scientists use machine learning to winnow through the chaff. The uncovered patterns expose dangers and opportunities.
 
Step Two: Predictive Analytics
 
Where risk estimation shows you what could happen, predictive analytics is creating a model for what will happen.
 
We first use data mining, sometimes known as KDD (Knowledge Discovery in Databases) to explore data attributes and its various distributions within a broad parameter space. If we seek to optimize the inventory position, say, we use machine learning to forecast demand, model historical shipment data, and model other variables that influence demand. Sophisticated mathematics are used to determine what is relevant, what is “noise,” and what is likely to happen next. Through a process called back-testing, we ask, “if we had used our algorithm in the past, would the various sites have had enough supply to meet demand?” If so, we deploy the machine learning model as an AI with the objective of inventory optimization.
 
Step Three: Prescriptive Analytics
 
With prescriptive analytics go beyond the prediction and use it to estimate the best combination of actions that should be taken in light of the current circumstances.
 
In prescriptive analytics, we utilize an artificial intelligence (AI) to prescribe specific actions to take based on knowledge of past and current data. For example, perhaps we’ve used predictive analytics to estimate the machine longevity on a factory floor. For example, the AI may discover that machines from provider B fail 20% faster than the machines from provider A or C—but only above a specific level of humidity. We now have an automated AI to alert us when to order replacement machines to minimize downtime, or which machines not to order during the humid months!
 
Outcomes like these are economically meaningful for companies. What’s more, they are within the grasp of many organizations because they’ve already made tremendous progress in gathering large amounts of high-quality data. Many organizations are generating incredible amounts of potentially valuable information—such as through cloud-connected sensors—but have yet to take the next step. That is, they are still unable to resolve patterns into actionable insights that can flow from their data centers right into their boardrooms.
 
Many organizations can gather large amounts of data, but the real moment of digital transformation arrives when you can commandeer that data.
 
Author

Brian Goodwin, PhD

Data Scientist

Tags in this Article