Demystifying the dark science of data analytics
- 17 October, 2017 21:00
Although few will publicly admit it, data analytics remains something of a dark science for many IT leaders, replete with mystical methods and seemingly inscrutable practices. Yet despite its somewhat enigmatic reputation, analytics has repeatedly demonstrated that it is a proven science, a powerful tool that generally leads to significant improvements in productivity, efficiency, sales, profits and other key business metrics and goals.
Today's analytics revolution caught many senior IT leaders blindsided, observes Michel Ballings, an assistant professor of business analytics at the University of Tennessee. "Only recently has computing become powerful enough to perform advanced data analytics," he says. "[Most] senior IT leaders graduated way before the data analytics revolution."
Advanced analytics is essentially a research skill, and most IT leaders and executives have never been professional researchers, notes David Johnston, lead data scientist for ThoughtWorks, a technology consulting firm. "These skills are most common in the academic community, and for this reason most successful data scientists and analytics managers are former academics." As a result, many old-school IT leaders view emerging analytics initiatives with a combination of bewilderment and dread.
[ Keep up to date with the 10 hottest data analytics trends today (and 5 going cold). | Bolster your career with our guide to the big data certifications that will pay off. | Get the latest insights by signing up for our CIO newsletter. ]
Embracing the unknown
Enterprise IT leaders need to recognize that analytics will present a challenge to their stable and well-structured departments. "There is a shift in skill sets needed within large organizations to drive data and analytics-enabled solutions," says Justin Honaman, a managing director in Accenture's digital technology advisory service. "Many traditional IT skill sets do not work with the new analytics engines, code bases and data management structures evolving quickly in the marketplace, fueling a need for new talent."
Seth Garske, executive director of marketing science at HackerAgency, a direct marketing firm, notes that many IT chiefs may find themselves unsure of the best way to approach analytics, especially given its overlap with other departments. "While analytics tends to be technical in nature, it really is more of a business function, similar to finance or accounting," he explains. "This misunderstanding has created some pretty ugly turf wars in some organizations."
Dan Magestro, senior manager of advanced analytics for West Monroe Partners, a business and technology consulting firm, believes that if IT leaders invest more time learning how analytics works, as well as the ways analytic processes can be used to help their organization, it would increase adoption and defuse in-house struggles over roles and responsibilities.
Deeper analytics knowledge can also help IT leaders understand why the approach often seems so mysterious. "Data science, in its best form, is an extremely creative endeavor," Johnston says. "There is not necessarily a need for managers to understand the internals of every analysis, just as owners of a software project need not understand the underlying technological internals." What matters most, Johnston says, is seeing the value created.
Unlike IT, where solutions are often obvious and widely adopted by enterprises worldwide, analytics processes are frequently unique and individualized. "Choosing the best analytical method is sometimes straightforward, sometimes art," Magestro says. "For example, looking for cause-effect relationships in data usually means some kind of regression, and looking for similar characteristics in large customer datasets likely involves clustering algorithms." When optimizing a marketing budget, an analytics expert can select from countless methods that can do the trick. "In such cases, it's often more important that a method is used properly and with good assumptions than whether it’s the 'best' method," Magestro says.
Experts differ on whether enterprise analytics initiatives should be centralized, either within IT or a standalone analytics department, or spread across individual business units. Many believe that IT is best positioned to serve as an analytics advocate and technology supporter, not as a base of all enterprise analytics initiatives. "There is no reason why data analytics should be siloed inside one department," Johnston says. "Rather, it’s a set of skills that should be encouraged to grow throughout the enterprise."
Since data science is a rapidly evolving field, there's a considerable advantage in having multiple teams collaborating and learning from each other, even if there is a bit of friendly competition among them, Johnston claims. "Different teams will do things in different ways, resulting in faster exploration of the entire field in order to better discover the kinds of methodologies most suited for one’s business environment," he notes. "Such cross-pollination of ideas can be further encouraged by rotating people in and out of different teams."
The analytics Center of Excellence (COE) model — a group or team that leads and coordinates analytics initiatives across the enterprise — has been discussed for many years, yet has found little support versus embedding analytics talent within business units, Honaman notes. "Within traditional IT, there is typically a centralized model for consolidating and managing operational data while providing access to that data to analytics resources within the business." Yet, with a few exceptions, this approach doesn't mesh well with the unique and specialized analytics needs of individual business units.
Magestro says that the case for centralizing analytics is strongest in two instances: when data or skills synergies are widely spread across different business functions, or when less mature business functions could benefit from the expertise a central team can provide. "We see cases where a central analytics team can be a temporary catalyst for growing functional analytics, in which case the central team might be best for highly specialized needs, such as machine learning or artificial intelligence," he says.
Regardless of how or where they originate within the enterprise, all analytics projects require strong and knowledgeable leadership. "The key is to have a good skipper at the helm," says Anirudh Ruhil, a professor of leadership and public affairs in the online Master of Public Administration program at Ohio University. "You also want the team leader to have a wealth of proven experience, because that is ultimately the gauge of a good analyst."
Exactly who leads an analytics project depends on an enterprise's level of analytics maturity, its industry and heritage, its leadership strength and the specific business areas that drive its strategy and growth, Magestro says. "In some sales-driven businesses, a senior leader of marketing analytics might make more data-driven decisions than any other area," he notes. "In companies with a strong central data management function, an IT leader might be best positioned to elevate analytics capabilities."
Competition for analytics talent is intense. "Most companies have sensibly given up on the idea of hiring the hot-shot data scientist with three PhDs to perform magic and instead have built teams of more junior but competent people," Johnston says. He believes that such a strategy that can succeed almost anywhere if certain principles are followed. "You must empower them to succeed," Johnston says. "Give them data, cloud computing and whatever tools they require."
Johnston also suggests that management should keep a loose rein on analytics teams. "An effective team that is delivering at a high rate of productivity does not require a small group of people to assert authority over any others," he states. "Naturally, the more tenured people will take on roles of slightly higher responsibility, simply because they are more experienced and likely more knowledgeable of how to complete the task at hand."
Business unit representatives should be involved in analytics project planning from the very start, from identifying which metrics to track through vetting the data visualization dashboards, says Phil Schmoyer, a manager at management consulting firm Navigate. Training business mangers and staff to interpret data, on the other hand, shouldn't be a major concern. "If an analytics group is performing optimally, it shouldn’t need to train people to interpret the results," he says. "[The tools] should be designed to be intuitive to the business unit digesting the information."
Dispelling the mystery
Honaman recommends that IT leaders set aside their analytics fear and skepticism and focus instead on the challenge at hand: getting actionable insights into decision makers' hands. "Marketers, supply chain practitioners, finance leaders and operational executives all have a stake, interest and investment in analytics, and this creates new complexity and politics as it relates to the space," he says. IT's role is to help, in whatever way possible, enable the creation of clean, available data and insights to fuel and drive business decisions.
Rather than dismissing analytics as an inscrutable mystery, IT leaders should pitch in to help data experts gain access to the data and tools they need to succeed. "Far too often, we see lack of success caused by bureaucratic constraints imposed on a data science team rather than lack of talent," Johnston says. He notes that it's not unusual, for instance, to see organizations making data access too painful to make it worth pursuing and putting tight limits on the tools that are made available. "These constraints can easily reduce productivity by factors of three or more," Johnston says. "The kind of ambitious people you want to keep won’t remain long in such an environment and natural selection will leave you with an analytics team made up of docile and uncreative people who won’t deliver the value expected from them."