Making Sense of Analytics – Establishing a Framework

The world of big data and predictive analytics is hot right now. I can’t open my inbox without receiving a barrage of emails from various companies promoting a new analytic solution or feature set that can solve a business problem. How can a business executive seeking to foster a data driven and analytical culture within his team or company make the right decisions while managing the inherent risks? Most executives today know they should be doing something in the data analytics space because everyone else is – or says they are. How should they synthesize all this information and focus on what will help them solve their most pressing business problems?

The analytic framework should initially address a range of questions which will help guide the resources that are required to bring to bear on the business problem. There are several key steps in understanding this framework.

Firstly, start at the end point – what business problem are you looking to solve and why? In general, the types of problems to be addressed may be categorized in one of four areas, as set out below.

Next, consider whether the nature of the problem is strategic or tactical in nature. For example, an objective to analyze and improve the budgeted marketing allocation between various media channels may only be required once every quarter or even annually. On the other hand, an objective to predict fraudulent transactions, or to customize pricing based on specific constraints or identify customers at high risk of defecting to a competitor is more tactical and must be done in real time and at scale.

This in turn, gives rise to a need to understand the resource requirements necessary to implement an analytic solution. The depth, breadth and complexity of skills required to address the problem needs to be understood. Do the resources already exist in-house or do they need to be acquired? The skill set for a self-serve, data discovery driven business intelligence initiative is vastly different to the need for a complex statistical analysis. While this may sound simple, the technical capabilities for BI and BA applications have begun to overlap in recent years and it can be easy to believe that a quantitative savvy business analyst has the expertise to manage projects that require far greater data science expertise. Further, what technology platform should be used? There is a myriad of solutions in the marketplace today, ranging from industry heavyweights to start-ups. The industry analyst firms can provide a framework to assist in making the best vendor selection.

Finally, the business case for any analytics endeavor should be based on an ROI analysis. Management should be focused on addressing the big problems that can help drive revenue, minimize cost and eliminate inefficiencies and yield the best return.

Having selected the focus area for analysis, management needs to identify the people, tools and processes that will be deployed. While the choices made in this regard will be specific to the needs and available resources of the company, management should be sensitive to establishing a collaborative, multi-disciplinary team that eliminates silo friction and better aligns the strategic goals with operational activities. With the framework established, the business owner should have some understanding of the underlying analytic elements of the project – these are set out below.

Historically the industries most receptive to predictive analytics were those that are heavily regulated or have a high volume of transactions that should be handled in a consistent manner but that is no longer the case.

Once the area of focus is identified then the analyst can begin to identify the most useful data to bring to bear on the problem. This is where some domain knowledge is helpful. Analytics can do much of the heavy lifting, but understanding the industry and what data may have predictive value is key to extracting the most effective insights. The data may be structured or unstructured and may be generated from the company’s internal ERP systems, be it financial, transactional, profile related etc. or from external sources such as web data, public record data or data purchased from independent vendors.

Much has been written about the evolution of analytics from descriptive analytics, which analyses data and generates metrics describing what has happened and is the value proposition of most business intelligence applications; to predictive analytics, which applies statistical modeling techniques to data to predict the likelihood of future events and on to prescriptive analytics which prescribes the most appropriate course of action. Management’s decisions as to which level of analytics best meets their objectives will start to target those analytics vendors that best meet those needs. However, there has been a trend towards convergence between BI functionality and analytics as BI vendors incorporate automated modeling behind the scenes to provide more predictive insights to their traditional measurement and reporting functions.

While more traditional BI activities may be performed by business users charged with a broad mandate to discover relationships and establish performance metric reports and dashboards, or what Gartner refers to as “citizen data scientists”, generating more complex analytic insights may require a deeper understanding of data science and modeling techniques. Data scientists have the expertise to apply the most effective modeling techniques and algorithms to different types of problems. Further, they may have some level of domain knowledge that provides them with additional insight.

Of course, once the discovery and modeling work is completed and some insights are generated, they need to be deployed. Whether it is targeting a specific offer to a prospect, allocating a marketing budget, predicting whether a customer will attrite or some other finding, those decisions must be acted upon to create business value. Processes should be established to ensure that these decisions are implemented in a consistent, repeatable and scalable way. The financial impact of the decisions should also be measured so that the initial strategic objective can be evaluated.

In summary, any analytics project should be viewed as part of a broader shift to a data driven analytically based decisioning culture. While the initial focus will likely be narrow and concentrated on a specific business issue, business leaders should recognize that the technology and skills associated with an analytics culture will increase as the disciplines become more broadly spread across a range of business activities and processes.