by Thomas Bosshard
A widely acknowledged, purpose-based classification divides analytics it into four different types: descriptive, diagnostic, predictive or prescriptive analytics. This division is based on the questions the types try to answer:
These descriptions are useful in determining how advanced a business intelligence solution must be for it generate the business value you want.
As Figure 1 shows, the categories also differ in the extent the decision-making process is left to human judgement as opposed to fully automated decisions. The more automation, the more complex the analysis is. Therefore, each of these types is one step up from the former in regard to complexity and the size and variety of data sources needed to create value. The most basic form is descriptive analytics.
Both descriptive and prescriptive analytics draw on historical data within the organization to describe variation, for example a company’s year-over-year growth rate. In a marketing context, this could be a description of a customer base’s demographics or about how it changed over the years.
The business intelligence gained from descriptive analytics is typically communicated with visualization tools, a typical outcome would be a report or a dashboard consisting of bar, pie or line charts. However, other than delivering general summaries of the past and current state of affairs in an organization, these reports give no indication about why something happened the way it did. Establishing the cause for the observed phenomena is left to the person interpreting the report who them mostly takes educated guesses about cause-and-effect relationships.
Many organizations still mainly use descriptive analytics, which is perfectly adequate to get a quick view of what went right or wrong to date. However, this does not represent fully data-driven decision-making. Descriptive analytics needs to be combined with other forms of analytics in order to explain trends and help decide what to do about them.
Diagnostic analytics goes beyond basic reporting of past performance and looks for patterns and interdependencies in the data that can explain why something went right or wrong. It is usually combined with descriptive analytics and draws from the same data source. However, it provides the context that make conclusions about cause and effect more reliable. In other words, it gives meaning to the results of a descriptive analysis, for example by correlating different data sets such as demographic and behavioral data or by singling out the more important reasons why something is happening.
Most traditional business intelligence packages offer this combination. The outcome is also typically a dashboard, but one from which more meaningful conclusions can be drawn from due to diagnostic intelligence.
While both types of historical analytics offer a deeper, hindsight understanding of past trends and the reasons for them, knowledge of future trends are just as important for a business to thrive. Predictive analytics and prescriptive analytics, the most sophisticated of the four types, promise this.
Predictive and prescriptive analytics calculate probabilities, i.e. the likelihood of something happening in the future. For these forecasts to be reliable, giant amounts of data needs to be mined (“big data”). These can include numerous different sources of external data such as posts on social media or competitive intelligence. The data volumes analyzed are in fact so staggering that only very complex predictive models can make sense of them.
An already widespread area of application is sentiment analysis. Also known as opinion mining, predictive analytics is set to replace traditional methods of market research as the amount of data it sifts through is huge and the result thus considered more reliable than for example focus groups. It enables businesses to gauge what people think and, as a result, to take a more pro-active approach in spotting business opportunities even before they arise.
This type of “advanced” analytics also uses existing historical data to train predictive machine learning (ML) models. This culminates into the most sophisticated type analytics – prescriptive – that serves as a basis for deciding how to react to forecasted trends. Prescriptive analytics is considered to be fully digitalized decision-making and the nearest approach to artificial intelligence.
If applied with the right data, the insights gained from prescriptive analytics can represent valid recommendations of the actions a company can take in the future to boost to business outcomes