If you work in the technology space, the chances are you’ll be bombarded on a daily basis with vendor-led messaging on big data. If you believe the hype, these technologies can do everything from make our cities smarter to improve our health, and help us understand our customers better. But your big data projects will only succeed if you pay careful attention to the data itself and how you interpret it.
To put it another way: it doesn’t matter how big it is, it’s what you do with it that counts.
Let’s start with some definitions for descriptive, prescriptive and predictive analytics. Descriptive analytics can be thought of best as displaying data you already have – this could be data on how many page views or click-throughs have been generated in a certain time frame. For this reason, it tends to be historical – think about the average temperature for a particular month, for example. The challenge with descriptive analytics in a business context is that you end up making decisions based purely on historic data – which is a bit like driving your car forwards by looking in the rear view mirror.
Predictive analytics are all about forecasting. They won’t produce a definitive statement on what will happen but instead will help you predict what’s likely to happen. A great example everyone will be familiar with is the weather forecast. Number crunching predictive analysis here will provide the meteorologist with a best guess; a forecast of the weather the next day.
Prescriptive analytics involves giving advice to people consuming the data. It’s all about going one step further than predictive analytics by actually providing concrete steps that they can take. If we continue the weather theme, it could be a personal assistant app advising its user to take an umbrella with them when they leave the house, because the forecast is for rain.
Predictive vs. prescriptive?
All of these types of big data analytics might be useful to your role. Predictive analytics are great to use for scenario planning – providing a range of outcomes to choose from and plan against. With these options you can work through to a natural conclusion to predict things like what sales will be like next year, or how many visitors you’ll get to your website. The model is designed to give you answers – or at least best guesses – to your question.
Prescriptive analytics are more nuanced than predictive models and require a greater understanding of the business context. In short, the outcome depends on the use case, so in the weather example Siri might tell you to take an umbrella, but if it’s applied at the Met Office, the outcome could be to issue a flood warning. And in a retail context it could be intelligence to recommend you raise the price of umbrellas in store because it’s forecast rain.
Don’t forget the data
The truth is that if you’re going to succeed at big data you need to choose your analytics tech carefully. Many of the products out there are solutions in search of a problem – so it’s really important to do your research and make sure you invest in something that will actually be useful for your business.
The other key is to pay close attention to the data. So often organisations fail because the data itself is inconsistent, inaccurate or incorrect. This doesn’t just mean someone has entered it incorrectly. It could also be that you’re using the wrong kind of data. People that use big data successfully usually spend time working on the dataset itself as well as the model.
To give an example of this, we operate primarily in the realm of marketing and customer data, most commonly including names, contact details, job titles and the like. In this Customer Relationship Management context, we find that this type of static data is often inaccurate, subject to manual error on input, but also is weak for predictive purposes. It is the wrong type of data to accurately predict purchase intent, for example. While job titles might help you target the right audience, this approach cannot indicate a changing likelihood to purchase.
In this type of predictive analytics modelling it’s vital to use behavioural data, which changes every time a customer interacts with the business. Using up-to-date data on their interests and interactions will produce a much more accurate and valuable result, and will also enable prescriptive output, such as telling a salesperson what to say to a hot prospect, or a marketer what topics to write about in the next newsletter.
This unstructured interest data holds the key to understanding and influencing the customer journey, and it is growing in volume as well as importance. In fact Marc Benioff, the CEO of Salesforce recently announced that it outweighs structured data in their CRM ecosystem by 5:1.
The first step is to find technology that can predict outcomes by turning behavioural data into this interest and intent data, and then you’ll gain a level of insight into your customers you never thought possible. Armed with this insight, you are able to enter the world of prescriptive analytics, where you don’t just know what might be around the corner, but also what to do about it right now.
This article is from the CBROnline archive: some formatting and images may not be present.