At Algo.ai, we’re proud to stand at the forefront of AI research and development. That’s why we’re excited to share some of our research on prescriptive analytics — combining ideas from machine learning and operations research to develop a framework for using data to prescribe optimal decisions.
Descriptive analytics is one of the first steps in data processing: summarizing historical data to produce useful information and prepare data for further analysis. Predictive analytics utilizes statistical techniques such as data mining, predictive modeling and machine learning to analyze historical facts and make future predictions. Prescriptive analytics takes elements of both descriptive analytics and predictive analytics to turn data into optimized predictions.
Operations research and management science typical starts with models and aims to make optimal decisions, leaving the data itself as an afterthought. Conversely, machine learning and data science often begin with data in order to make decisions.
The opportunity that these methods present is to take the significant amount of available data — often big data in electronic form — and develop a theory that unifies operations research and management science that takes data from predictions to prescriptions. The more data we have, the more optimal decisions can be made.
In one real-world example, take the case of a global Fortune 100 media company that sells nearly 2 million different CD, DVD and Blu-ray titles at over 50,000 retailers worldwide. Faced with limited retail shelf space, an endless array of titles and uncertain demand for new releases, how can they best choose which titles to order and in what quantities to maximize the amount of media they sell?
Using a wealth of data and our prescriptive analytics methodology, this company can use prescriptive analytics to fill in gaps throughout the decision-making process to find the best titles to promote. Data was obtained from all available sources, including four years’ worth of sales data spread across their 50,000 retail partners.
Next, data was sourced from online sources such as IMDb and Rotten Tomatoes to determine factors that indicated a movie’s popularity — its actors, Academy Award nominations and box office statistics. Finally, more data could be pared down by location to discern a particular title’s popularity in any given region.
With prescriptive analytics, all data can be considered useful. Even though some data may be more helpful than other data, it can help inform the best solutions and open up previously unseen ways of thinking.
For complete insights into how prescriptive analytics works, download our scientific research paper.