Predictive analytics is helping businesses in fields as diverse as healthcare, retailing, hospitality and insurance peer into the future to optimize inventories, manage staffing, enhance customer engagement, set pricing and achieve many other profit-building goals.
Thanks to steady advances in artificial intelligence (AI) and machine learning (ML), predictive analytics is growing increasingly accurate and insightful. Yet many businesses continue to view predictive analytics with various degrees of skepticism, convinced that the technology remains too complex, too disruptive and too expensive to incorporate into routine use.
While doubters can be firm in their beliefs, one thing they often fail to recognize is that predictive analytics is a tool that’s best applied, at least initially, in small measures. “Getting started in predictive analytics is a lot like learning to swim,” explains Ellen Houston, applied data science lead at Civis Analytics, an Eric Schmidt-backed data science software and consultancy firm founded in 2013 by Dan Wagner, chief analytics officer for President Obama’s 2012 re-election campaign. “If you dive directly into the deep end, you may not make it very far,” she quips.
Sriram Parthasarathy, senior director of predictive analytics for Logi Analytics, a predictive analytics platform provider, suggests that skeptics get their feet wet by using the technology to find the answer to just a single predictive question, utilizing historical data that’s readily available. “Once you’ve started demonstrating the ROI of answering that problem, you can over time add more data to improve the model and incorporate new insights into other parts of the business workflow,” he says. “Success with these initiatives will provide your business a competitively differentiated application and will drive more revenue.”
Here’s a guide on how to get started with predictive analytics at your organization, from the ground up.
Build your predictive analytics team
Most enterprises that are serious about using predictive analytics form a cross-functional planning team to identify potential opportunities and develop targeted strategies. Obtaining leadership buy-in is essential, suggests Anton Berisha, senior director of clinical analytics and innovation at LexisNexis Risk Solutions. “Then form a small but capable predictive modeling team and build up from there.”
Team members should include:
- Predictive modeling experts
- Content experts, preferably with some analytics experience
- Data and database analysts
Successful predictive analytics planning teams typically consist of a partnership between quantitative practitioners and the relevant stakeholders, says Eric Felsberg, principal and data analytics group director for nationally-ranked law firm Jackson Lewis. “This enables those with subject matter knowledge to work closely with the analytical professionals to develop tools, methods and solutions that are quantitatively robust, [and] are informed by those with experience on the subject matter to solve the relevant business challenge.”
It’s important to understand that predictive analytics is not just crunching numbers, observes Ye Zhang, CTO of Katabat, an international management consulting firm.
“Instead, what you really need is to distill data into actionable strategies that can generate incremental revenue,” he states. “For that purpose, the awareness and involvement of the whole organization is always preferred.”
Define the business problem and find the right tools
The team must always seek a clear understanding of the business problems that predictive analytics will be expected to address, notes Nachum Shacham, a data scientist at analytics software provider Teradata.
Key questions to consider include:
- Are the problems defined accurately in terms of business process and goals?
- Can the impact of predictive analytics be quantified, including expected benefits and costs?
- What are the risks from prediction errors (i.e. false positives and false negatives)?
- What are legal risks and liabilities that can arise due to prediction bias?
As various predictive analytics applications and models are viewed and vetted, team members should keep in mind that it’s not necessary to re-invent the wheel. “You should think about acquiring existing tools/models/vendors from your industry and verify your ROI model as soon as you can,” Zhang advises. Core technology competence in predictive analytics is very valuable, but it is also very expensive to acquire. “It’s okay to first leverage existing technology to build a proof of concept to verify the ROI, then decide whether/how much of the predictive analytics stack you build is worth the effort of keeping it in house,” he notes.
Build internal capabilities
Although predictive analytics application and model vendors are becoming increasingly niche-focused, it’s still hard to mount a serious initiative without having at least some in-house professionals on hand who know how to modify and tweak apps and models to meet specific business forecasting needs.
Predictive analytics professionals come from various backgrounds, including computational biologists and chemists, astrophysicists, mathematicians, computer science majors, quantitative social scientists and, of course, academically trained statisticians, notes Hyoun Park, founder and CEO of Amalgam Insights, a technology consulting and management firm. “The important part is to look for analysts who are problem solvers and dig into data, rather than pure number crunchers.”
Existing employees are often highly suitable predictive analytics training candidates, since most are already familiar with enterprise goals and practices. Management can follow several approaches to training, including in-house lessons, external education programs or online classes. “A special consideration for predictive analytics training is verifying that trainees have the tools and applications they need to carry out their duties after the training,” Shacham says.
Engage departments and end users
Many enterprises maintain a separate data science business unit that’s responsible for acquiring, developing, customizing and implementing predictive analytics applications, models and tools. “However, it is not uncommon for data science teams to work closely with IT departments to implement the computational infrastructure required so that they can deliver analyses and tools to the relevant stakeholders,” Felsberg observes.
In some instances, third-party vendors, managed by relevant stakeholder departments, assume responsibility for virtually all analytics operations. “What department should be responsible for predictive analytics operations depends on the scale of the operation and on the long-term vision of how analytics and predictive methods will influence the business,” Felsberg says.
Although developing, configuring and customizing predictive analytics tools can be highly challenging, using them shouldn’t be. “Predictive tools designed for business stakeholders should, in fact, be easy to use,” Felsberg says. The difficultly, he adds, lies in ensuring accurate interpretations of analytical results. He feels that any training provided to end users should focus on helping individuals make accurate interpretations, allowing results to be used effectively and with minimal risk.
Often, however, little or no training is necessary. “Many of today’s predictive analytics tools have become little more than drop-down functions in a very friendly GUI interface, so it’s becoming easier and easier to train employees on these tools,” observes Andrew Pearson, managing director of Intelligencia, a software consulting firm that advises clients in the gaming, lottery and sports betting industries. “I work in Asia in industries and countries that lack basic employee education, and yet some of these tools are within the reach of people who don’t have degrees in analytics and barely have high school educations,” he states.
Enterprises just beginning with predictive analytics tend to fall into the same traps. Parthasarathy notes that dated, inaccurate and poorly formatted data is the most common obstacle new adopters face. “If you’re putting dirty data into the application, you’re going to get inaccurate insights out of it,” he explains. “Data readiness is a critical component of preparing for predictive analytics, alongside ensuring that you’re gathering data points that best address the business question you’re looking to answer.”
Launching a predictive analytics initiative simply to demonstrate that the enterprise is “advanced” or “ahead of the curve” usually turns out to be a costly and inefficient exercise. “To successfully leverage predictive tools, one should have a specific business challenge to address first, then look to see if the data is or may be available,” Felsberg suggests.
Another common challenge organizations face is finding a way to effectively share predictive insights with colleagues who can use the information to make better business decisions. “You can gather all these great insights through predictive analytics, but your efforts will be fruitless if the right people aren’t receiving this information in an actionable way,” Parthasarathy cautions. “The best way to make certain that predictive information is shared with the right people is to incorporate these insights into the applications people use every day in the context of their workflow.”
Many organizations also fail to realize that predictive analytics demands continuous reinvestment. Applications and models must be periodically revisited and updated, or they will fall out of date. “You have to keep adjusting your predictive models to fit current business conditions,” Zhang notes. “It is by no means a one-time investment.”
A large number of businesses also don’t fully understand the holistic view predictive analytics can bring to their organization. “The reality is, predictive analytics can infuse an entire customer journey, from customer acquisition to customer intelligence to the customer experience and potentially, customer churn,” Pearson says. “Customer journey information feeds into labor management, supply chain management, as well as a whole host of other departments.”
For more on avoiding predictive analytics pitfalls, see “7 tips for overcoming predictive analytics challenges.”
A final takeaway
The fields of AI, ML and deep learning are creating extraordinary opportunities in predictive analytics, Pearson observes. “Hadoop data lakes and many of the open source tools are allowing companies to implement far cheaper analytics tools than were available just a few short year ago,” he explains. “It’s a fascinating time to be in this business.”