Thanks to new tools that are more powerful and easier to use, predictive analytics is moving from the labs of data scientists and onto the desks of business analysts. In the hands of these users, predictive analytics is helping answer questions and predict behaviors in areas including marketing, HR, and fraud detection. In a webinar on July 8, 2014, Fern Halper and Allen Bonde discussed these new tools, as well as the skills and knowledge that business users need to use the tools effectively.
Halper, Research Director for Advanced Analytics, TDWI, began the discussion with her own definition of predictive analytics: “A statistical or data mining solution consisting of algorithms and techniques that can be used on both structured and unstructured data to determine outcomes.” The emphasis on outcomes is the key differentiator between business intelligence (BI) and predictive analytics, Halper explained; BI helps users understand what has happened, while predictive analytics helps users predict outcomes based on data.
For example, BI can tell you how many telco customers have churned in a year, while predictive analytics can help you identify which customers are most likely to churn. (In the coming year, Halper said, we’ll also hear more about prescriptive analytics, which suggests or automatically initiates actions based on predictive analytics.)
What has changed?
In past years, predictive analysis was performed by highly trained statisticians and mathematicians using sometimes-arcane scripting languages. But new analytics solutions with graphical user interfaces and pre-built algorithms are giving rise to a new class of user who has more business knowledge – and often less mathematical or statistical expertise. (Adoption of these tools doesn’t lessen the value of statisticians and mathematicians, Halper emphasized in the Q&A following the webinar; it just means their expertise can be applied to thornier, more complicated problems.)
Halper believes three skills can help users frame problems for predictive analysis: Critical thinking, meaning the ability to formulate meaningful questions based on data; domain expertise, which enables a user to understand the relevancy of data and identify data outliers; and data sense, which includes the ability to understand unstructured data and to think in terms of targets and calculated variables.
Halper also identifies three skills used by business analysts to explain and defend the predictive analytics models they create: tool knowledge, which users can develop through online training; knowledge of techniques such as decision trees, linear regression and cluster analysis; and storytelling ability, which is sometimes necessary when explaining predictive models to business leaders.
Organizations can quickly roll out individual analytics projects that deliver value, Halper concluded. (In the webinar, she offered specific guidelines for choosing and launching small predictive analytics initiatives.) But broader adoption – that is, getting an entire organization to embrace predictive analytics – doesn’t happen overnight. “It takes time to build trust, show results and get [an analytical mindset] to permeate an organization,” she said.
Allen Bonde, Vice President, Marketing & Innovation, Actuate keyed into Halper’s idea of launching small analytics initiatives and posed a question: “How do you get started with predictive analytics, especially when you’re not a data scientist?” he asked, and offered three high-level observations in response:
- Fast is the new Big. Speed is critical when testing models, refining questions, and adjusting models based on results and feedback. “Speed enables iteration, and iteration enables better performance and insights,” Bonde said.
- Staying Focused is Key. Too much data and too much speed can lead to scattershot efforts. Organizations should focus first on problems where real impact is possible and where results affect the broadest set of business users. After all, predictive analytics isn’t an intellectual exercise. “It’s not only about discovery and development of models,” Bonde said. “It’s about operationalizing those insights.”
- Picking the Right Tool. A predictive analytics tool that will be used by business analysts needs to help users access data, understand patterns in that data, and deliver insights from that data. Bonde detailed these three interrelated challenges in the following chart.
Pre-built algorithms, such as the k-means clustering, Decision Trees, and Association Rules found in Actuate’s BIRT Analytics solutions, help to shorten time-to-value when companies give predictive analytics tools to business analysts. Having these capabilities built in and easily accessible enables both speed and focus. “That combination really makes predictive analytics more practical,” Bonde said. “This drives adoption – which, in turn, drives innovation.”
The complete webinar with Fern Halper and Allen Bonde is available for free replay. You can also learn more about BIRT Analytics and download a free trial version today.
Crystal ball photo courtesy of Christian Schnettelker