Four Steps to Get Started With Predictive Analytics
Data and analytics have been a hot topic throughout the marketplace for the past few years, but many industries still are working on strategic ways to access and integrate analytics functions. The latest push, of course, is to ramp up capabilities so that you aren’t just mining insights from historical data, but predicting future behavior and trends.
For pharmaceutical companies and healthcare providers looking to tap predictive analytics to gain a competitive advantage, here are four steps to get started:
1. Compile the data. The first step is to create a consolidated repository that’s accessible across different groups in the enterprise. Thanks to Big Data constructs, one can easily accommodate data from the internal sources (ranging from call centers to direct shipments), data from clients (such as patient information), syndicated data (for example, sales or managed care) and data from third-party sources (such as email marketing vendor and Medscape).
2. Query the data. The second step is to make it as easy as possible for the widest group of users to query the data. The days of technical queries routed through IT are long gone. For fast results, users must be able to query data with as close to a natural language interface as possible, preferably one that interprets needs and suggests appropriate data sets. Equally important is the ability to conduct “business or domain friendly search,” which allows users to assemble a query using characteristics gleaned from the data, categorizing healthcare providers by value or by insurance plan, for example. Queries should be as simple as, “Identify healthcare practitioners with high sales and large percentages of patients on favorable plans.”
3. Augment the data. New data arrives constantly, so scalability and extensibility are crucial. It must be relatively painless to extend the enterprise repository to accommodate and ingest new data sources, whether structured or unstructured. That means having connectors to a wide set of data types and formats, including social media, as well as relying on metadata-driven transformation to reduce or eliminate manual processing required in a standard ETL (extract, transform, load) process. Equally important is the capability to easily configure and perform quality checks on the data to maintain high quality and avoid duplication.
4. Enable low-cost, high-speed analytics. Companies should be cognizant of the value of reusing and re-purposing analytical processes and investments in order to cut costs and time to insights. This is facilitated by the creation of a framework for analytics, starting with business rules engine and moving to advanced analytics that accommodates plug-and-play algorithms. Ultimately, a user should be able to assemble applications for a specific need in a user-friendly way from components processing data (e.g. text analytics) or a pre-packaged machine learning tool (e.g. collaborative filtering).
By following these four steps, pharmaceutical companies can create the foundation for predictive analytics across a number of departments, from sales and marketing to research and customer service. For instance, consider the challenge of a pharmaceutical company facing diminishing market share for a particular customer segment. The sales team probably would wonder what’s driving the loss of market share and how it’s distributed across insurance plans. To answer this question, the company might use predictive analytics on a big data platform and compile a variety of data sources, especially customer demographics, market share, customer's responsiveness to past campaigns, composition of the provider's practice, and sales representative feedback. This helps identify potential recommendations in the same way that Netflix and Amazon use algorithms to suggest movies and books. The analysis might reveal that the product has a disadvantaged formulary access—lower ranking on the list of medications available—for a majority of customers in a specific market when compared to the competition, and that healthcare practitioner has high affinity for peer recommendations but has not received any material in past few months that highlights peer support.
Predictive analytics can suggest a course of action that not only improves customer engagement, but also suggests ways to undermine the competitors’ formulary advantage. It also can harness feedback from the sales representative to inform an “intelligent recommendation engine” so that the “next best actions” provided to sales teams and contact centers continue to improve over time.
Other examples of predictive and prescriptive analysis include:
- Customer segmentation, retention, cross-sell opportunities
- Marketing budget estimation for new drugs
- Revenue forecasting based on drug efficiency and outcomes
- Forecasting daily patient load by diagnosis and protocol based on historical data
- Signal analysis to predict Adverse drug events
- Predicting and preparing for disease spikes based on past disease outbreaks
- Forecasting sales based on past marketing ROI data
- Guidance to provider on best suited clinical protocol based on past symptoms and outcomes
It's important to note that providing simplicity to users on the front end almost always entails a degree of complexity on the back end—requiring knowledge of domain, third-party data sources, open-source technology, statistical analysis, machine learning/AI (artificial intelligence) techniques and more. As these capabilities are in high demand, it’s often difficult to find staff with these multiple skills. However, creating a viable, rewarding predictive analytics engine can give companies a competitive advantage—and, based on the pace of change in the data and analytics space, soon could become table stakes.
Sanjay Bhasin is a seasoned Healthcare Analytics professional and a principal of Smart Analytics that has created solutions for Healthcare, Consumer Goods and Financial Services leveraging Predictive Analytics, Machine Learning and Business Intelligence technologies. We welcome your questions or feedback on this topic.
