The Four Types of Data Analytics

Terms like business intelligence and data analytics mean different things in different contexts and the only way to hack through the Forest of Jargon is with a Machete of Specificity. Whether you’re building an analytics tool, shopping for a business intelligence application, or just looking to get a better handle on IT terms, it’s useful to be familiar with the analytics spectrum. In this article, we’re going to focus on disambiguating the term data analytics by breaking it down into types and aligning those with business objectives.

Data analytics is the process of extracting, transforming, loading, modelling, and drawing conclusions from data to make decisions. It’s the drawing conclusions bit that BI tools are most concerned with, as the extracting, transforming, and loading steps generally happen at the database level. There are four ways of making sense out of data once it’s been formatted for reporting, and these are descriptive, diagnostic, predictive, and prescriptive analytics.

Descriptive and diagnostic analytics help you construct a narrative of the past while predictive and prescriptive analytics help you envision a possible future. The above diagram shows examples of features that would fall into each of the four categories, along with the types of questions those features are designed to help answer.

Descriptive analytics comprise your reporting bedrock; no BI tool is complete without it. Descriptive analytics is designed to get you basic expository information: who, what, when, where, how many? A canned report, in this instance, is a report that has been designed for you and just delivers the necessary information around a subject. A canned sales report for a magazine publisher might display the upcoming issues, the sales representatives that have sold advertising in those issues, how much they’ve sold, how much more they need to sell before they reach their sales quotas, and total information. An ad hoc report, by contrast, is one you’ve designed yourself, giving you the leverage to ask more specific questions. An ad hoc sales report at the same magazine company might include the types of advertisers each sales representative sold to, separated by industry vertical, in addition to all the other sales information included on the canned report.

Diagnostic analytics help you answer the question of why something happened. Queries and drill-downs enable you to request more detail related to a report, which can help explain surprising values. Referring back to our fictional magazine publisher, let’s say a sales representative named Chris has sold considerably less than usual on a particular issue. A drill down showing employee work hours might reveal a three-day gap wherein Chris was out of town, explaining the discrepancy. Chris’s manager trusts that he’ll be able to make up for lost time but could use an alert tool to notify her if Chris’s number doesn’t hit a certain mark by a specified date.

Predictive analytics help you identify trends in relationships between variables, determine the strength of their correlation, and hypothesise causality. Statistical modelling and predictive modelling go hand in hand; the former enables the latter. Let’s say that our fictional magazine publisher is considering expanding their distribution to one of two new territories, and they want some information about which will give them the highest boost in revenue. They could use statistical modelling to determine how closely revenue correlates with, for example, an area’s population, median gross income, number of homeowners, etc. and with what degree of certainty. Then, using predictive modelling, they could plug in the statistics of the two territories to extrapolate a plausible revenue value for each region.

Prescriptive analytics is where artificial intelligence and big data come into play. Whereas statistical modelling is more about assessing correlation to prove or disprove a hypothesis, machine learning is about predicting outcomes based on numerous variables. Big data, staggeringly large sets of information often reflecting crowd behaviour and sourced from outside the company in question, is essential to machine learning because it is complex enough to refine the artificial intelligence’s decisions over time. AI can give probable answers to what-if questions such as, What if we run an article about reclaimed furniture in our next issue? and suggest courses of action, like to write about reclaimed patio furniture in particular, which will be trending next month for the target demographic.

There’s usually a positive correlation between a company’s analytic needs and its size. International corporations operate on a massive scale, so every action becomes magnified. Predictive and prescriptive analytics prove extremely valuable in the upper tiers of such organisations, but lower-level managers concerned with day-to-day operations will likely have no use for statistical modelling and be better served using a platform specialising in descriptive and diagnostic capabilities. Whatever your requirements, it helps to know the lay of the land as well as a bit of the local parlance as you navigate the data analytics landscape.