Predictive Analytics and the Age of the New Know

In today’s volatile, global economy, the ability to predict outcomes is more vital than ever. Competition has increased immensely due to expanded access to markets, changing rules and shrinking business cycles. In other words, it’s business as unusual. Among the biggest shifts is the growing power of customers to influence their buying experience, and companies are under pressure to achieve more with less.

So, what is it exactly? Predictive analytics is a blend of tools and techniques that incorporate modeling, statistics, machine learning and data mining, thus enabling organizations to identify patterns that can predict future outcomes and drive better decisions. A variety of data types are used including: transactional, demographic, historical, text, sensor, economic and unstructured to unveil and measure patterns that identify risks and opportunities. Driven by unprecedented amounts of big data, access to cheap computing power and the advent of modern tools; predictive models are becoming more efficient and accurate with improved access for all kinds of enterprises. And that’s a very good thing.

It’s interesting to note that the same technology advances that’ve empowered buyers have also provided companies the opportunity to embrace and leverage the power of information. Organizations can now effectively understand and address not only consumers, but commercial customers as well. The ability to obtain, organize, analyze and store huge amounts of data about their markets and customers is the driving lever. Once appropriate systems are in place, the advantage of deploying predictive analytics to create competitive advantage becomes clear. Enterprises will have the means to win, retain and serve customers in outstanding ways. And that’s the bottom line for every business – serve customers in ways that surpass your competitors.

To achieve this, enterprises must work to gain predictive powers in three critical areas:

1. Provide clear insights about customers and business processes. Today, many traditional business intelligence tools contain simple predictive models which expose causative trends and future projections. This valuable information is surfaced via dashboards and reporting directly to managers and executives. The problem though, this vital content often lacks the connection to business decisions, process optimization, customer experience or any other action based on the predictive insights.
2. Create intelligent, adaptable customer interactions and business processes. Organizations must use predictions to change the future, otherwise the effort becomes an exercise in futility. Since the top tools can deploy their scoring engines or models into applications where the need for insights is apparent, the impact can be immediate. Consider the ability to detect fraud at the moment of swiping, or adjusting digital content based on user context, or being able to initiate proactive customer service for at-risk revenue sources.
3. Reimagine customer engagement and instigate new digital products. The potential for predictive analytics extends far beyond what most companies exercise today. As model building and deployment continue to accelerate, application developers can embed predictive analytics into deployed applications more seamlessly than ever. When used to examine application data, predictive analytics gives developers the option to focus on features that promise the greatest customer value. They’ll also be able to predict the impact of new application aesthetics or functionality.

Today, more organizations are capitalizing on this discipline because tools are quickly being adapted to accommodate a breadth of users. While large organizations will require data scientists to do the heavy lifting, business professionals and application developers (as highlighted above) can also employ today’s tools. And the more people who use predictive analytics, the more value the organization will glean. Here are a few examples:

• Data scientists require powerful, flexible tools. The demand for data scientists is huge, but not bigger than the demands being placed on data scientists. They are being pushed to build more models, more accurately, in less time. To be highly productive, they need tools that speed the process and have the might to analyze gargantuan data sets. Once the analyses execute, the insights must be operationalized wisely and efficiently. This can happen through custom APIs, Predictive Model Markup Language (PMMLs), or other methods for creating scoring engines and embedding predictive actions in applications.
• Application developers must apply insights to enhance software experiences. Not every organization will have data scientists, but most can benefit from predictive analytics. To address this, there are new tools intended for less credentialed users. These tools use modern development interfaces familiar to application developers who have experience with integrated development environments, such as Eclipse or Visual Studio. Further, smooth integration of these insights into apps is promoted via web services, APIs or PMMLs.
• Business analysts must explore and consume reliable predictions. As the demand for predictive analytics becomes more pervasive, options are being created for even the most naïve users. Accordingly, there are tools that provide “one-click predictive modeling.” This simple to execute process runs a series of algorithms against the data and finds the prediction with the highest accuracy. To provide even more options to non-data-knowledge types, there are an increasing number of exchanges and marketplaces for prebuilt predictive applications, such as Azure Machine Learning Marketplace.

While the tools and techniques can serve a myriad of goals, they should never drive the analysis. The lifecycle of effective predictive analytics begins with the obvious – by asking specifically what business problems require solving.

• Interested in attracting new customers? Who isn’t? New customer acquisition campaigns supported by targeted response modeling could bring in more customers for the same marketing cost.
• Is avoiding risk a top priority? Determine which customers have a high likelihood of defaulting by executing models on claims and risk of loss.
• Want more profit from a particular set of customers? Companies can increase profits from current clients by analyzing cross and up-sell targeting models.

Once the larger questions are answered and the models executed, the process begins a continuous loop. This is no one-time operation. Organizations must rerun their analysis on new data to ensure the models remain effective. They’ll also need to respond to changes in customer desires and the impact created by competitors. Many firms are analyzing data weekly, and some even continuously. Be advised however, once the direction is established, the truly game-changing insights are won by asking deeper, more creative questions. After the question(s) has been defined and refined, they can be answered in a continuous predictive discipline by moving through these six steps:

1. Uncover data from a variety of sources. Don’t limit where you look for data. Whether internal (data siloed in enterprise applications) or external (government data or social media), valuable data can exist in numerous hard-to-access locations. Explore the data from various sources via one of many available advanced visualization tools.
2. Wrangle the data. A key challenge for predictive analytics is data preparation. When it comes to preparing the data; e.g. calculating aggregate fields, stripping extraneous characters, merging multiple data sources, filling in missing data, and so on; it’s estimated that many users spend near 75% of their time in this stage. This should be an area of targeted focus.
3. Construct a predictive model. Predictive analytics tools include dozens of varying machine-learning and statistical algorithms. These offer knowledgeable application design and development (AD&D) pros and data scientists a wealth of options to choose from. The type and completeness of data, as well as the kind of prediction desired, should point towards the best algorithm(s) to select. Typically, the analysis is run on a subset of data called ‘training data’ and set aside ‘test data’ to evaluate the model.
4. Assess the model’s accuracy and effectiveness. Predictive analytics has nothing to do with absolutes. Rather, it’s purely concerned with probabilities. To assess the predictive accuracy of the model, data analysts apply the model against the ‘test data’ set. If the model meets the defined requirements, then that model is a reasonable candidate for deployment.
5. Use the model to deliver actionable remedies to the business. To be valuable, predictions must eventually enable opportunities or circumvent negative outcomes. This requires a trusting exchange between the creators of the models and business peers. Model creators must learn from their partners on the business side which insights can lead to viable action, and business peers must learn to trust in the predictions and those who created them.
6. Monitor and improve the effectiveness of the model. A predictive models’ effectiveness will likely degrade and/ or improve over time. As a monitoring mechanism, newly accumulated data is habitually rerun through the algorithms. If the model becomes less accurate, it will need to be adjusted (by tweaking parameters in the algorithms) or the scope of the data will need to be expanded.

The power of data is real and it’s here to stay. But to fully leverage predictive analytics and apply it to winning in this volatile, high-tech global economy; organizations must turn the lens within. Leadership style and the underlying culture are sure to make or break the effort. Continuing an old-time hierarchical, top-down leadership model is the best way to get the least value from this technology. The problem is lack of velocity. Since speed to market is everything, the previously tried-and-true process of synthesizing data and communicating to top management, then waiting for them to create directives to send back down the chain is impractical and backward-thinking.

As Thornton May, self-described futurist and confirmed big brain, puts it, “We are living on a hinge of history, on the cusp of a new age, the age of the New Know. The New Know is all about moving your organization beyond just having the data, to knowing what you need to know when you need to know it.”

Predictive analytics is inarguably the best means to that end.