Understanding Descriptive, Predictive and Prescriptive Analytics
At the heart of an effective business strategy is the use of data and analytics. Knowing answers to questions such as – what do customers want, how do customers move through the sales cycle and what prices are considered fair, offers a competitive advantage.
The ability to translate data into business value, however, comes from being able to extract valuable insights from analytics. That can only happen if organizations have the ability to take general purpose analytics and optimize it for their unique business situations.
Knowing the following types of analytics and their capabilities can serve as a useful starting point:
- Descriptive analytics sorts through data from various sources to give a view of the past.
- Predictive analytics uses computer modeling and information from the descriptive process to give insights into trends and probable future outcomes.
- Prescriptive analytics uses techniques such as game theory and simulation to make recommendations based on past performance and future expectations.
Determining Which Class(es) of Analytics You Need
Descriptive Analytics – What has Happened in the Past?
Just like the name implies, this is a form of descriptive statistics. They yield quantified, summarized information about past events – how sales compare to the number of visitors over time, how different venues are performing relative to each other, or how the ratio of profit/revenue differs between online and offline sales.
These analytics are typically available in the form of standardized, canned reports and can mainly be accessed through a point-and-click style dashboard.
Above is an example of the kind of report descriptive analytics generates. From this chart, one could infer that except for Mall D, the visitor count for all other malls has been consistent from July through December. Performance issues at specific locations can now be diagnosed more easily, or the data can be cross-referenced with website visits or other metrics.
One of main uses of descriptive analytics in retail is to find links between buying habits and other behavior. It is now easy to determine how many people responded to a coupon promotion and what their average spend was, how effective and economical different forms of advertising were, how store traffic (including metrics such as the time spent in store, which zones were visited, etc.) influences sales and what level of engagement your efforts on social media have achieved.
Outside of retail, the same techniques have also found several useful applications at sports stadiums, airports and hotels. Banks use them to assess individuals’ credit risks and identify marketing opportunities, while medical researchers use them to track the spread of diseases.
The power of descriptive analytics does not lie in making it easier to calculate averages or draw graphs. The main difference between it and traditional statistics is that technology can provide us with huge amounts of raw information, especially when tools such as RFID tags, WiFi access points, Bluetooth beacons and QR codes are plugged into the system. What it does is analyze a mass of data and present it in a highly simplified form and find correlations between unconnected pieces of data. Also, it can display information at a highly granular level, such as “average dwell time for males aged 20-25 on Saturdays”.
Predictive Analytics – What is Likely to Happen?
Predictive analytics uses repeatable and mathematically justifiable methods to make credible predictions about the future. For example, it can answer a question such as – which locations are likely to double their sales over the next two years based on historical data?
Most people know how to extrapolate one variable against another. Predictive analytics, however, uses far more complex models, including those based on machine learning. It typically uses historical data both as part of its training set and as input, correlating multiple variables over time and testing its hypotheses and assumptions as it goes.
Applying predictive analytics requires a greater knowledge base than simply interpreting the trends of the recent past. Ideally, some experience in computer science and statistics should be available in-house. Even with a powerful analytics engine and a large amount of data, it’s frequently possible to get the answer you want by simply tweaking scenarios drastically and often enough.
Done well, predictive analytics can be really useful. Every manager has probably lost sleep over questions such as – whether the revenue projections for the next quarter are remotely accurate? Whether it is best to favor one revenue stream over another? or Whether a round of layoffs will simply result in having to re-hire people in a month’s time?
Predictive analytics also have more direct, tactical applications. It may, for instance, be used to estimate sales at various price points, the number of visitors to be expected on a given day based on date, weather and recent social media sharing related to your brand, or whether it is the right time to run a sales promotion. It does, however, by definition stop short of making suggestions. Instead, its purpose is to take much of the guesswork out of making important decisions and allow managers to anticipate the most likely contingencies.
Prescriptive Analytics – What Actions Can be Taken?
Prescriptive analytics identifies the steps to be taken to achieve a specific outcome. It can probably be compared to hypothesis testing, where one course of action is explicitly recommended over another based on the evidence available. It is the most complex and advanced form of analytics.
One of the most well-known applications of prescriptive analytics is dynamic pricing. Based on sophisticated modeling of consumer behavior, it can predict the increased likelihood of a transaction taking place if a discount is offered. Based on that likelihood, it adjusts the price accordingly. Prescriptive analytics uses historical data, expected changes in the environment and advanced modeling to present optimized actions for the company to achieve what it wants. These suggestions also take uncertainty into account.
Implementing prescriptive analytics correctly in different situations is often a challenge. Certain algorithms, such as those based on the logistic equation or production scheduling, maybe well understood but there are others business processes that are more complex. Not only does an accurate theoretical model have to be constructed of the business process being simulated but all inputs, outputs and interrelationships have to be defined mathematically and in a way that can be programmed efficiently.
If this can be done, though, prescriptive analytics offers enormous advantages. In the first place, all suggestions are based purely on data and logic, without any sort of gut feel or bias entering into them. Perhaps even more importantly, different departments and individuals within the same company will all have different views, priorities and concerns. This makes consensus and even a common understanding of a problem difficult to reach. Where a decision is complex and much data is available, prescriptive analytics may well produce results far superior to any meeting.
Don’t Be Analytically Challenged
In a world where data is ubiquitous, a company that is able to make correct and evidence-based decisions quickly is likely to outperform any less-informed competitor.
At the moment, descriptive, predictive and prescriptive analytics are all somewhere between the early adopter phase, and the time where it will be a standard feature of retail and other sectors. While the challenges involved in installing and utilizing analytics may seem a little daunting, not doing so is likely to be far worse in the long run. Losing market share will only be the beginning.
Proximity MX, our location analytics software that uses WiFi infrastructure to provide actionable insights into customer behavior and helps you engage with them real-time.
Ask for a free demo to learn how Proximity MX can help you with WiFi analytics.