Demystifying Advanced Analytics: Which Approach Should Marketers Take?

“Advanced analytics” has been the new buzzword on every organization’s mind for the past several years. Recent advancements in machine learning have promised to optimize every arm of an organization – from marketing and sales to supply-chain operations. 

For some, investments in advanced analytics have been worth the hype. Those who succeed can gain a sizable competitive advantage from a relatively cheap and non-disruptive investment. For others, advanced analytics are far from an instant fix-all. Without honing in on the appropriate time, place, and method, data becomes worthless. The price of data collection is also getting cheaper, misleading organizations into valuing quantity over quality. 

Real, sustainable success is best achieved through the synergy of machine learning and human capital. The goal of advanced analytics should therefore be to emphasize the humanity within organizations – not to replace it. While the pitfalls should be acknowledged, incorporating advanced analytics doesn’t have to be daunting. Outlined below are three broad, widely used approaches to advanced analytics, as well as a brief guide on when to employ these approaches. 

Descriptive Analytics

Descriptive analytics, or “business intelligence,” recognizes patterns in historical data. Simply put, this method of advanced analytics answers the question, “What happened?” Descriptive analytics are:

• Accessible: Input and output performance variables are visualized as dials on a dashboard. Managers use historical trends as rough guidelines, before adjusting the dials as they see fit. 

• Backed by concrete, empirical data: The past is unchanging and verifiable. Because of its objectivity, descriptive analytics has ingrained itself into the daily tasks of the average business manager, and that won’t change anytime soon. 

• Hard-capped by the capacity of the human brain: We lack the memory and processing power to manually sift through large swaths of granular data and therefore become dependent on data aggregation. In descriptive data, the facts we base our decisions on might be concrete, but the decisions themselves remain a guessing game. 

• Over-reliant on internal transaction data: Internal data is readily available and cost-effective but encourages managers to pollute the objectivity of their descriptive analysis by mixing in their own anecdotal experiences. This creates biases which stymie innovative thinking, as managers prune data to reinforce their traditional beliefs. 

Predictive Analytics

Predictive analytics also sifts historical data for patterns yet differs from descriptive analytics by building models that can predict future outcomes, rather than relying on the intuition of managers. Prescriptive analytics are:

• Rationalized by immediate impact: The potential for “quick win” opportunities coupled with its ability to spit out frequent insights and suggestions makes this method a favorite for offer management, CRM segmentation, and maintenance. 

• Structurally flawed: It is impossible to predict with complete accuracy a variable as complex as demand – there are too many moving parts. Even individual components, like weather, competition, and supplier performance, cannot be predicted with absolute certainty.

• Limited in number of inputs: For a model to be valid, its variables must be independent of one another. Therefore, common predictive techniques such as regression, clustering, and time-series forecasting use very few variables.

• Constrained by the growing divide between data scientists and business scientists: Data scientists strive to increase the accuracy of their model, while business scientists are concerned with emphasizing the added value of predictive analytics – often these incentives are at odds with each other, resulting in both false positives and false negatives.

Prescriptive Analytics

Out of the three methods, prescriptive analytics is the closest to complete automation, as machines make decisions that are based on managers’ defined objectives. This method answers the question, “What needs to be done?” Prescriptive analytics are:

• Output oriented: Optimizing the business impact of decisions becomes the goal, rather than in input-oriented methods where the accuracy of decision variables is prioritized. The model builds on each inaccurate experiment, inching closer to an optimal outcome. 

• Tasked with balancing market prediction and uncertainty: Market prediction drives the expected revenues, while uncertainty drives the expected costs. In contrast, predictive analytics tunnel-visions the a deterministic impact, while ignoring the level of error in demand uncertainty.

• High-cost and logistically complex, yet high reward: Expensive, dedicated software and hardware solutions are required to translate management strategies into mathematical, machine-friendly optimization objectives and business rules.

• Dependent on specialized human expertise: There is a tall barrier to entry for data scientists capable of running prescriptive analytics models. These specialists enable the model to dynamically adjust its recommendations according to the management’s specifications – ensuring optimal outcomes and systematic adherence to all rules and constraints.

Case Study: Price Markdowns to Remedy Excess Inventory

An organization operated gift stores at museums, and therefore had to deal with seasonal and unpredictable demand. This organization had over 100,000 SKUs, making discounting difficult to optimize. Which of the three approaches proved to be most successful?

By using descriptive analytics, markdowns were calculated by multiplying the proposed markdown percentage by the number of units on hand. However, this approach proved inadequate as it solely relied on historical inventory data and didn’t consider customer or context-related factors impacting consumer demand.

Unsatisfied, managers moved to predictive analytics. They used regression-based techniques to discount products with the highest price elasticity, calculated by factoring in historical sales volumes and prices. They simulated scenarios to determine the optimal markdown mix based on strategic objectives – considering factors like margin and inventory level. However, the models were only marginally better than the descriptive model, due to various factors beyond price affecting sales. 

Eventually managers settled on a prescriptive analytics model, which was able to consider a vastly wider range of factors influencing consumer behavior. To reach this conclusion, this organization had to redefine the role of their management and prioritize asking the right questions rather than having all the answers. The company learned that humans excel at intuition and dealing with limited data, while machines are better suited to make decisions in data-rich environments.

Finding the Right Approach

Distilling this down, organizations should conduct a careful cost/benefit analysis to determine which analytic approach to take – factoring in the relevance of available data as well as the strength of the business case.

Descriptive analytics is often the best fit when there is limited data, coupled with low complexity. Prescriptive analytics can only be justified if the frequency of decision-making and complexity of decision is high, which gives the model access to data that is both granular and relevant to the problem at hand. Predictive models are the middle ground between these two extremes and perform best in areas where there is little forecasting uncertainty. 

There are some business problems that cannot be solved with advanced analytics because the flaws are inherent, not a result of inefficiency. Ultimately, the strength of the business case is defined by the amount of inefficiency that data-driven insights can address – and how much the humans at the center of the decision are served by the approach.