The ability to foretell the future can take many human twists and turns for good, for evil, to save lives, to satisfy greed, to alter the inevitable. For businesses, success in sales, marketing, operations, product development and ultimately customer satisfaction rests with the ability to look ahead and determine the right course of action.

Businesses don't have access to a souped-up DeLorean that can travel to the future and back. So the next best and most accessible time machine is none other than predictive analytics -- and its newest relative, automated predictive analytics. Advances in predictive analytics -- namely artificial intelligence to tackle and trawl lakes of big data, machine learning to seek out patterns in relevant information and improve outcomes, and visualization to distill those findings into easy-to-understand charts and dashboards -- can ultimately result in credible predictive models that legitimize actionable intelligence and encourage businesses to make bold and ultimately rewarding decisions.

No need to convince IT and business leaders at small companies and large enterprises in a multitude of manufacturing and services sectors. TechTarget's 2018 Priorities Survey placed predictive analytics among the top software initiatives planned for implementation this year by corporate decision-makers, along with cloud-based apps, business intelligence and big data analytics.

But there's a catch

While many companies see predictive analytics tools as an indispensable competitive weapon, "adoption remains elusive" for about half the organizations that have made initial plans to implement the technology, according to TDWI's Q2 2018 "Best Practices Report: Practical Predictive Analytics." Based on responses from data scientists, analysts, engineers, corporate IT and business leaders, the report indicated that if users had fulfilled their plans to implement predictive analytics tools, then 75% to 80% of organizations would be using them. Yet, in reality, just 35% to 40% have followed through on their plans.

This gap between planning and implementing predictive analytics is mainly due to shortages in the data science skills needed to build models and put them into action. Add to that data infrastructure issues, lack of funding and unrealistic executive expectations. Mandates sometimes can come from the top to implement predictive analytics without addressing and defining an actual business problem to be solved.

"It's the result of a combination of lack of skills; lack of skills with tools, including data-related tools; and corporate culture," said TDWI analyst Fern Halper. "And, of course, it is one thing to build a model that might provide some actionable insights -- like customers who do X, Y or Z might be at high risk of churn. It is another thing entirely to put those models into production to drive action."

Models are complex, sensitive and need TLC

The iterations involved in building, refining, tuning and rewriting predictive models, not to mention managing ever-increasing amounts of unstructured and real-time data flowing into corporate coffers, are all seen as impediments to the modeling process and putting insights into action.

"More organizations need to think through how they put models into production for action," Halper acknowledged. "Some organizations put models into production and forget about them. At one of our conferences, a company mentioned that they hadn't looked at a model in production for seven years. ... Operationalizing the model is what provides the most value."

That's where automated predictive analytics tools can play a multifaceted role. Most respondents to the TDWI survey believe the biggest benefits of automated predictive analytics tools is that more people can now build models and build them faster -- a basic necessity for any company working to keep up with and even surpass the competition in all aspects of its business.

"Right now, we see that only 16% of respondents to our surveys are using the easy-to-use tools -- automated machine learning, for example -- to build models," Halper reported. "That number may increase -- another 40% claim that they will [be using these tools] over the next few years -- but who knows?"

Automated doesn't mean automatic

Even with automated predictive analytics tools to help turn the unskilled into moderately and passably skilled modelers by easing the complexities of modeling, there needs to be an applied set of checks and balances, what Halper calls "the missing piece": managing and monitoring those models.

"As tools become easier to use, it is important to put controls or a QA process in place to make sure that 'good' models are put into production," Halper advised. "If someone builds a model, but doesn't know how to interpret output, for instance, that is a problem if the model is going to be put into production. They need to register models and include metadata about those models, so that they know more about the models in production. They need to monitor the models once they are in production to check for degradation."

To address the skills gap in building and monitoring predictive models, Halper said companies "are taking a multipronged approach. They are hiring some data scientists -- typically those with experience -- and supplementing [them] with others in the organization that they are looking to upskill as well as potential data scientists right out of universities."

And they'll need continuing education. "If you're going to use certain algorithms," Halper conjectured, "even if [the automated tools] are easy to use, you still need some training on the algorithms, as you will have to interpret the output, defend the model and perhaps be part of the monitoring process."

While some progress is being made toward closing the gap between the desire to adopt predictive analytics and actual implementation, it's anyone's guess as to how long it will be before most companies take full advantage of these sophisticated tools. Care to make a prediction?