End User Business Analytics ASCUE 2015
Steve Knode, PhD Collegiate Faculty, UMUC http://www.umuc.edu/analytics/academics/index.cfm [email protected]
Business Analytics Descriptive: [discovery phase]
Looks at past data and tries to find important relationships to gain insights as to how to approach the future. Tries to answer the question, “What has happened?” Uses data visualization techniques along with data mining techniques to unearth key relationships. Key terms: dashboards, data visualization, data mining, knowledge discovery
Predictive: [prediction phase] Applies modeling techniques to find a quantitative algorithm that relates inputs
and outputs to provide actionable insights. Tries to answer the question, “What will likely happen?”. Uses machine learning and other modeling approaches to quantify the relationships. Key terms: neural networks, regression, machine learning, classification, decision trees
Prescriptive: [operationalizing phase] Tries to answer the question, “What do I do when the expected or predicted
happens?” Prescriptive analytics focuses on optimization techniques and suggests actions designed to improve business operations. Attempts to leverage what has been learned from descriptive and predictive analytics to develop improved business solutions. A large part of prescriptive analytics includes the use of ‘business rules’ to embed the analytics findings into operations. Key terms: business rules, decision management, optimization models
Source: Davenport, T., & Kim, J. (2013). Keeping up with the quants. Boston: Harvard Business Review Press
How is Analytics used?
Source: Aligned Resource Optimization , retrieved from: http://www.sas.com/resources/whitepaper/wp_4183.pdf
How is Analytics used?
Source: Aligned Resource Optimization , retrieved from: http://www.sas.com/resources/whitepaper/wp_4183.pdf
Current sweet spot
How is Predictive Analytics used?
Associations: e.g., linking purchase of diapers with beer
Sequences: e.g., linking events in order or together, such as
graduating from college and buying a new car
Classifications: e.g., recognizing patterns, such as the signs
of customers who are most likely to leave the company for a competitor; applicants as low, medium, or high risk; nature of insurance claims as normal or suspicious
Forecasting: e.g., predicting buying habits of customers
based on past patterns
Estimation: e.g., estimate the probability of positive response to a direct mail campaign; Estimate customers’ lifetime value to the enterprise.
Prediction: e.g., predict customers who are likely to attrite; predict the number of customers who will accept an introductory zero interest credit card offer and not repay within the time limit of the offer.
What types of decisions can predictive analytics help with?
Source: Wessler, M. (2014). Predictive Analytics for Dummies. : Wiley. Retrieved from: http://media.wiley.com/assets/7225/54/9781118859643_custom.pdf
Some Recent Success Stories:
Source: Goldstein, M. (Webinar, Apr 16, 2015) Beyond the numbers: Using Prediction to Save Lives. Retrieved Apr 20, 2015, from http://www.information-management.com/web_seminars/beyond-the-numbers-using-predictive-to-save-lives-10026665-1.html
How does the use of analytics improve decision making?
Helps “frame” the decision
Provides “transparency” of decision process
Enables the handling of (much) more data
Speeds up decisions (?)
Fosters innovation
Enables the use of more complex models
Performs complex calculations
Standardizes approach to decision making
Provides an audit trail for decisions
Improves/changes the business model
What’s Changing?
Analytics for the end user
Inexpensive (or even free) software
Easy-to-use software
Elimination of the tedious calculations
Focus is on understanding and explaining
Turning analytics into actionable results
Analytics “in the cloud” and “mobile”
Analytics anywhere, any time, ondemand
The use of realtime data
Internet of Things
Analytics models
Decision trees
Regression
Neural networks
End user tools used at UMUC
Predixion – Developing predictive analytics models
http://www.predixionsoftware.com/
BigML – Decision Trees and Clustering
https://bigml.com/
Palisade – risk analysis
http://www.palisade.com/
Predictive Analytics – end user software
Decision Trees
Easiest to develop, understand and explain
Often very accurate
Require much less data massaging
BigML software
Free for use (some size limits)
Fast, easy-to-learn, powerful features
Runs on BigML servers
Excellent for Decision Trees and Clustering
Results turned immediately into actionable rules
Predictive Analytics