Fresh Tips From Gartner: Combining Predictive and Prescriptive Analytics
March 18, 2016 by Madison Jacobs
Gaining in-depth knowledge about practical business intelligence (BI) and analytics strategies is essential to achieving success in the digital age. At Gartner's BI & Analytics Summit, many leaders and analysts in the business technology space shared their insights around emerging and foundational techniques and technologies that help enterprises learn and re-master necessary skills -- and gain the competitive advantage. We attended an awesome session on combining predictive and prescriptive analytics and wanted to share some great takeaways from the presentation!
Session: Three Ways to Combine Predictive and Prescriptive Analytics
About Lisa: Lisa Kart (@Kart_Lisa) is a Research Director in the Data and Analytics team at Gartner. They focus on data science and advanced analytics, including predictive and prescriptive analytics, machine learning and optimization. Lisa has more than 20 years of experience in applying advanced analytics to solve complex business problems (Source: Gartner).
Understanding predictive and prescriptive analytics
Although predictive and prescriptive analytics are different, Lisa explained that combining the two can lead to a more positive impact on critical business decisions.
In the opening of the presentation, Lisa offered up a terrific analogy to explain the difference between the two types of analytics.
Think of it this way: a weather app on your phone provides you with a "prediction" about whether it’s going to rain next week (this is like predictive analytics). But, the weather app doesn't tell you if you need to pack an umbrella or bring a raincoat, or what you will need to remove from your bag or suitcase to make room for your raingear. It doesn't take into account other important factors required to provide a "prescription" for how to protect yourself from the rain.
"Prescriptive analytics helps us make decisions, and in many cases, it relies on some of the information we know. Some of those are fixed, some of those are things like predictions," Lisa said.
Why do we care about predictive and prescriptive analytics?
She shared that by 2020, predictive and prescriptive analytics will attract 40 percent of enterprises' new investments in business intelligence and analytics.
How do you use predictions to influence decisions?
"Organizations want to predict things," Lisa said. "Why do they want to predict things? It helps give them a forward-looking view into running their business better."
Here's an example of some things that enterprises might want to predict:
- Credit risk -- What's the credit risk of my customers if I'm going to lend to them?
- Marketing performance -- How likely are customers to respond to marketing offers?
- Customer satisfaction and retention -- Will customers churn?
- Sales -- What's the forecast for sales revenue?
- Insurance fraud -- Can we predict fraudulent claims?
And associated with these predictions are decisions.
Here's an example of some decisions companies might make based on the above predictions:
- Credit risk -- Setting the optimal price for a loan or credit card.
- Marketing performance -- Choosing the best marketing offer to send.
- Customer satisfaction and retention -- Establishing proactive steps to reduce customer churn rates.
- Sales -- Choosing the best product offering and pricing.
- Insurance fraud -- Establishing the best data to collect and analyze to investigate fraudulent claims.
"If I think about decisions I want to make, having a prediction is very helpful in making that decision,” she added. “If I think about how prescriptive analytics can help me is that I can start to look at the prediction for alternative actions I can take. I can think of all the possible decisions I can make... I can then optimize that... I can look at those together."
Breaking down predictive and prescriptive models
Here're the three main ways we predict things:
- Predictions - Finding the probability of a specific outcome (Example: Responsiveness to a marketing offer)
- Forecasting - Predicting a series of outcomes over time (Example: Sales revenue)
- Simulation - Predicting multiple outcomes and highlighting uncertainties (Example: Confidence interval or distribution around an outcome)
And, here're the prescriptions you can apply to develop a solution based on predictions:
- Business rules -- Predefined framework for choosing among alternatives
- Optimization -- Outcome driven, constraint-based evaluation of an interdependent set of options
Three of the most common ways of combining predictive and prescriptive analytics
- Predictions + rules
- Forecasting + optimization
- Simulation + optimization
Our Favorite Use Case: Predictions + Rules
Here’s Lisa’s example of combining predictive and prescription analytics from Wells Fargo:
Wells Fargo combined both types of analytics to help determine the best way to treat delinquent accounts in real estate collections. They used their existing predictive risk models with a decision tree product from FICO to help them develop an effective collection strategy that they could apply to multiple accounts and portfolios.
"Rules are good for applying consistency across many customers or many situations. They can also use this strategy to make sure that they're not calling too many people on the first day because they don't have enough resources to do that," Lisa said. "The goal is to identify swap sets, so people to accelerate that are going to result in better performance of collecting accounts and some to decelerate that might end up paying on their own or that might be good customers that we don't want to anger."
Why combining predictions and rules made sense
Lisa explained that it made sense to combine predictions with rules because Wells Fargo had different models they wanted to combine. Once they had a better feel for the short and long-term delinquency rates, the best action to take became fairly obvious to them. It gave Wells Fargo a repeatable set of rules they could apply to multiple portfolios.
Lisa concluded that the nature of the problem determines what type of predictive model you choose to take: predictions, forecasting or simulation. The complexity of the problem determines if you use should use business rules or optimization to make a decision.
You can read Lisa's report on how to get started with prescriptive analytics here.
The first step to effectively leveraging analytics is getting access to business critical information. Watch our newest video and find out how Captricity can help you get your hands on the customer data you need to develop and facilitate data analytics.