There's no longer any debate over the importance of advanced analytics in modern enterprises. There's also no question that data quality is critical to the accuracy and trustworthiness of analytical results.
Unfortunately, ensuring and sustaining that data quality -- through proper analytics governance -- is often a weak link in many organizations' analytics chains. There are several reasons why, including the relative newness of today's advanced analytics, which has led to a lack of universal best practices and defined supporting roles.
But there's another, less obvious reason why analytics governance is often lacking: a false confidence. The implementation of advanced analytics invariably follows the implementation of more conventional analytics, which require pretty straightforward -- even simplistic -- governance. Put simply, it's easy to become complacent and to underestimate the urgency of keeping a close eye on data governance when it's time to introduce complex models in the enterprise.
Here are some of the most important points to pay attention to in the business analytics governance process, starting with how it has changed.
Most organizations that have implemented analytics in decision support got started with numbers culled from internet activity. Studying customer behavior in order to anticipate customer need rapidly became a standard practice, and more complex applications of analytics evolved from that point.
In this, the enterprise had it easy, governance-wise -- the data collection and analysis for that type of analytics was relatively straightforward and easily monitored. It didn't hurt that the data's cost of acquisition was practically nil.
But as analytics became more complex, so did analytics governance. Why? Because the creation and support of complex models frequently requires a broad range of inputs whose connections between one another can vary in strength and relevance over time. Those variations must be detected and incorporated into the model to maintain the integrity of its performance.
Put differently, the relationships between interacting variables in a model will certainly change over time, with some becoming more significant and some less so. It is necessary to prune models like plants in a garden, trimming away those that fall from significance and promoting those with increasing impact. The difficulty there is twofold: achieving vigilance in refreshing models and realizing that bad data will lead to modifications in models that degrade their performance.
Building a new framework
Today, analytics governance straddles the chasm between business intelligence (BI) and business analytics (BA), and success hinges on understanding the difference between the two. BI looks backward, BA looks forward; BI is about understanding past events and performance, while BA is about predicting what will happen next.
As supply chain expert Robert Handfield of North Carolina State University said, BI is needed to run the business, while BA is required to change the business. And, as we know, in today's digital era, the survival of a business is often dependent upon the speed with which it can adapt.
Mastering and actualizing these distinctions requires enterprise-wide competence and involvement -- and a new business analytics governance framework. That framework needs to include the following components.
Universal data quality standards. It seldom occurs to those in charge that it is not enough to employ data analysts; data quality analysts are becoming increasingly important. The maintenance of data quality is becoming an art in itself as the internet of things continues to creep into data collection milieu. The establishment of enterprise-wide quality standards, as well as personnel to enforce them, is an essential step.
An analytics governance board. It's a common mistake in the enterprise to assume that analytics are the purview of IT. That's not true; it is a dynamic partnership between IT, executives, line managers and field personnel. For advanced analytics and model deployment to be strategically meaningful, alignment between all participants must be maintained, with executive sponsorship and representation from each line of business and department involved.
This becomes especially critical when enterprises deploy organization-wide data collection infrastructure, methodology and supporting expertise to keep the cost of analytics operations in line.
Perpetual monitoring and maintenance. The constant flow of data is not just essential for the use of a predictive model, but for its maintenance as well. Equally perpetual mechanisms must be put in place to monitor model performance. Model maintenance should also be frequent.
New roles. In addition to the new role of data quality analyst, a number of other new jobs -- as well as some redefined old ones -- will play a part in business analytics governance moving forward.
The data steward, for instance, is responsible for maintaining the fitness of data for use in the models. Increasingly, data stewards are surfacing on both the business and technical sides of the enterprise. The data custodian covers the security, transport and storage of data with an eye on maintaining its integrity.
The business analyst continues to hold sway over the impact of model predictions on strategic planning, but with the rapid growth of advanced analytics, detailed knowledge of the models is required for business analytics governance.
Finally, there's the changing role of the business manager, who typically implements strategy and offers recommendations to executive leadership. The role is now more dependent than ever on BA and quality data. Today's business managers must learn to curb their intuition and place their trust in objective information -- hard to do at all, let alone in the swiftly shrinking windows of opportunity that are common in today's business climate.
Such trust is often hard to come by, particularly when the manager is a veteran of many years. It may be that the cultivation of that trust is, for the enterprise, the hardest challenge of all.