Descriptive vs. prescriptive vs. predictive analytics explained

Why so many types of analysis?

Every day, we work with diverse data: from accounting systems, customer management software, log files of our internet sites, shipping and logistics reports and even sources external to our organization such as supplier prices and exchange rates. Your business has many data sources, which afford many opportunities for analytics.

At one time all this information may have been brought together by IT teams into a standardized system such as a data warehouse to be used by a few specialists who built reports and dashboards and analyses for business executives. We used to talk about the enterprise "single version of the truth" with an emphasis on consistency and standardization.

Today, however, the users of our data are also very diverse. Even if you have a data warehouse, self-service software for visualization and reporting enables business users at all levels to do their own work, often with few dependencies on IT. But we see new specialisms emerging too, such as data scientists and data engineers.

This new landscape of data and a new, diverse population of people who we broadly call information workers, has created many patterns of analysis. Three of the most important you will hear about are descriptive, prescriptive and predictive analytics, but we could also add diagnostic and real-time analytics as interesting variants.

Give me the facts: What is descriptive analytics?

The most common analytic pattern is what we call descriptive analytics -- descriptive, because it gives an account of what has happened in your business. We can present this in many different ways: as reports, dashboards or visualizations. Yet they all convey what has happened in your business.

I use the past tense -- what has happened -- because descriptive analytics only describes what has already occurred. The reports may detail what occurred yesterday, or last month or over the last year or several years, but we're looking back over history. Gathering the data takes at least some amount of time: the latency of the report.

Sometimes (especially if we work in a fast-moving environment such as manufacturing or financial services) we may want to see much more up-to-the-minute data. Such a demand requires the somewhat specialized discipline of real-time analytics. I say specialized because pulling together the right data quickly enough proves technically demanding. Presenting such data to users in a useful manner poses a particular design challenge and enabling responses and actions to what we see in real-time data also requires specific software integrations. For these reasons, just speeding up your descriptive analytics does not truly give you real-time analytics.

What went wrong, what went right: What is diagnostic analytics?

If we look back, with descriptive analytics, over what has already happened, we will likely ask next a very human question: Why did it happen? Most often, we will ask, "What went wrong?" Less often, I'm sorry to say, we may ask "What went right?" We tend to be more surprised by our failures than our successes. As I said, these are very human questions.

These questions bring us to a pattern which we call diagnostic analytics. We try to understand not just what happened, but why it happened. This requires us to have a model of the business process under analysis which enables us to dig into relationships between activities in the process.

A descriptive report, for example, may list all our customer accounts, their purchase orders and related invoices -- a format useful for seeing top customers, orders increasing, slow payments and so on. However, the report has no model behind it which understands that sales calls are made by specific salespeople and typically lead to purchase orders, which in turn lead to invoices. Without this model, the report is only descriptive of what has happened.

However, with a model we might be able to diagnose that in the current year, fewer sales calls were made, but by more senior staff, which led to fewer purchases but larger invoices. This may be a very useful insight because it tells us not only what happened but also, to some extent, why. To be diagnostic, analytics must understand our business processes and how different elements of the process connect. To perform true root-cause analysis, layers of diagnostic analytics may be required to reveal patterns lying deeper in our business dynamics.

What needs to be done: What is prescriptive analytics?

As you can see, descriptive analytics can prove useful, but a smart use of diagnostic analytics can help us to understand the dynamics of our business more effectively. Nevertheless, our restless human curiosity and the demands of business leaders will surely take us on to the next question: What do we do next?

If today your business runs on data, tomorrow it will be running on analytics. Understanding the various types of analytics will help you map out your journey and ensure better business outcomes.

Now we need some guidance -- some insight into how we might modify our actions and processes to make better business decisions going forward. For this, we have another type of analytics, known as prescriptive analytics, which prescribes or advises us what to do. Various patterns of advisory analytics can be found in budgeting and planning tools, healthcare applications or even IT applications. The critical advantage lies in comparing how different choices could impact potential outcomes and highlighting the best options. In budgeting, we may see how different allocations across business functions may affect profitability. In healthcare, we may see how different care options may improve recovery. In IT management, prescriptive analytics can help us allocate on-premises and cloud resources effectively as we expand the number of users.

Look to the future: What is predictive analytics?

I am sure that today you have business reporting that looks to the past. It is descriptive. Possibly your enterprise planning or business intelligence software includes (or enables) diagnostic and prescriptive analytics. The final piece of the puzzle asks: What happens in the future? Whether we are concerned with tomorrow's chance of sunshine, or the price of oil next year, we may think of this as a great unknown, but analytics can really help us here too. Even though the results never prove completely accurate, we can use predictive analytics technology to drive future outcomes.

At the most basic, predictive analytics looks at the patterns of the past and projects them into the future. Based on what we have seen already within our business, we can make some useful guesses. You have had this experience with your online shopping -- customers who bought this, also bought that. Recommendations such as these are based on a prediction, which is based on a statistical model of what past customers bought.

In practice, our predictions may demand considerably more sophistication: For example, after a certain sequence of error codes, over some period of time, this piece of equipment failed and should be maintained immediately. You can see that predictive analytics use case is somewhat more complex, but still works by projecting what we know from the past into future scenarios. The algorithms which make these advanced analytics possible may be well-known -- projecting time-series such as sales figures or prices may use algorithms that many analysts learn at college. Similarly, recommendation engines are quite common and well-understood by many developers. However, more advanced uses may require cutting-edge machine learning technologies such as neural networks.

The analytic journey

This spectrum of business and technical capabilities can make analytics seem like a daunting prospect. However, as you look over your current business capabilities you may find you are already leveraging more analytics than you expect.

Your business reports -- operational, financial, managerial -- reflect a classic form of descriptive analytics. If you have business intelligence tools in place, your analysts likely do some diagnostic analytics, even informally. Your budgeting and planning tools may enable some prescriptive analytics. Taking stock of where you are not only helps to understand what analytics tools and skills exist, but also gives you a good perspective for planning future investments.

One thing is sure -- if today your business runs on data, tomorrow it will be running on analytics. Understanding the various types of analytics will help you map out your journey and ensure better business outcomes.