Predictive analytics: definition
Predictive analytics is a broad term for using historical and current data to make projections about what might happen in the future.
Making predictions about what’s next, about the future, is hard-wired into the human brain. We take our past experiences, quickly assess how some past experience is similar to the current situation and use that information to make an educated or experience-based guess about what’s likely to happen next.
Predictive models are an advance of human “gut” predictions, because they are objective, repeatable, based on lots of information and use statistics to identify and organised what matters most in making the prediction most likely to be correct.
It goes without saying that the more data you have, the more accurate your predictions will be – so organisations are increasingly looking to collect more data on their employees, customers, products, and brands in order to make accurate predictions.
Until recently, that kind of data was in limited supply. However, with the emergence of Big Data, predictive analytics has become not only more accessible, but more powerful and advanced than ever before. We now have the capability to collect huge volumes of data, along with the processing power to analyse rapidly and easily.
How does predictive analytics work?
Here’s a brief overview of the technology behind prediction analytics. If you’re less interested in what’s under the hood and simply want to know how you can harness its power for your business, feel free to skip ahead.
Making predictions from data involves constructing a mathematical model (AKA predictive model). This is a tool for finding out what you want to know based on historical data, the target outcome, and the known facts about the scenario.
You can think of a predictive model as a mathematical representation of reality. Like a scale model or architectural model, it replicates a real-world scenario or idea and scales it down so that only the parts you’re interested in are included.
To develop the model, begin by gathering all the data you have on the variables that you think might predict some outcome of interest.
If you are training a predictive model using machine learning, you’ll need to have some ‘source of ground truth’ to train it against. This is basically a dataset of the outcome of interest, so that the model can learn from what happened in the past.
When you then introduce new data, or the “test data set,” you don’t provide the model with your outcome information and you see how well it predicts it based on the ‘source of ground truth,’ looking at precision and recall metrics or other model accuracy measures, depending on the analytical approach you’ve taken.
For example, let’s say you’re developing a predictive model for finding out whether the sun is going to shine on a certain day. To train it, you’d provide data that covered things like:
- How often it was sunny on the same date in past years
- What the weather was like leading up to sunny days in the past
- Any known weather systems such as storms or areas of low pressure that affected sunshine
- And so on…
It’s important to split the dataset into training and test sets so that, when you give it your test data set — all the predictors without the knowledge of whether it was sunny that day or not — you can assess how well it predicted a sunny day.
The model would also need clear parameters to define the outcome you’re interested in. For example, you might specify the hours of sunshine and the temperature range that would qualify a day as sunny.
At the end of the training period, your model would (hopefully) be able to predict that, for example, sunny days are most likely after a thunderstorm, and happen more often now than they did 50 years ago.
As you can see from even this very simplified example, designing a mathematical model and training it to work well is a complex and time-consuming process. It usually requires the skills of data specialists to perfect. Fortunately though, there are accessible solutions available that give you all the power of a mathematical model without the need to develop one yourself.
What are the business benefits of predictive analytics?
With so many possible applications of predictive technology, the benefits are theoretically endless. Here are a few common use cases for predictive models in commercial settings.
- Planning ahead
Maybe the most obvious benefit of predictive analytics for business is its ability to help you see into the future and plan accordingly. Predictive technologies can tell you what’s likely to be over the horizon so that you can prepare in advance and adjust how you allocate your resources. For example, let’s say you are a fashion retailer. Your predictive model tells you that natural materials are about to rise in popularity. You can start working with designers and manufacturers who make these kinds of clothes, and cut back on your synthetic lines.
- Time-saving and efficiency
Businesses can turn a lot of the work involved in low-risk, routine decision-making over to predictive technologies, freeing up humans for more valuable or high-risk strategic tasks. For example, a predictive model can do much of the work of generating a credit score or deciding whether a straightforward insurance claim can be paid out. It may need some human support for outliers or complex cases, but it will take a lot of work off employees’ plates.
- Fraud detection
The strength of a predictive model is its ability to recognise patterns, which means it can also spot when something is out of place. Predictive technology can help businesses detect unusual patterns of behaviour that might indicate fraud. For example, if a banking customer based in the US suddenly seems to be making purchases on many other continents in a short space of time, the company can intervene to make sure the account is secure.
- Predicting customer churn
Predictive models can learn the behavioural patterns that precede customer churn, and flag up when they’re happening. By acting promptly, a company may then be able to retain the customer by taking action.
- Customer experience
Predictive technology can help businesses provide a personalised experience to customers by learning what they like and anticipating what they may want next. It can also boost the customer experience more generally by building an understanding of typical consumer behaviours and preferences that businesses can use to help them plan and design experiences.
- Text analysis
Free text (e.g. responses typed into an open field on a survey or as part of a customer review) is a form of qualitative experience data. It is information-rich but harder to process than numbers and rating scales because it varies in form and structure. Predictive technology can process text data at scale and identify clusters of words and phrases that represent certain sentiments or ideas. It can then generalise them to create a big-picture analysis that can be understood at a glance.
Predictive analytics: the time is now
Right now, we’re living in a sweet spot for predictive analytics. The technology is affordable, the know-how is accessible and there’s enough historical data around to make truly valuable predictions for business, governance and organisation of everything from ecological conservation work to education and healthcare.
But although these capabilities are more accessible than they used to be, now is also a time when the mastery of predictive analytics still offers companies a competitive advantage. Most businesses are aware that predictive technology is of value to themselves and their customers, but not everyone is using it – yet.
Predictive analytics tools
At Qualtrics, we’ve developed a suite of predictive analytics tools that put the power of advanced predictive analysis in the hands of just about anyone – no data science knowledge required. Here’s an introduction to the Qualtrics iQ predictive engine.
Using machine learning, Text iQ parses text from surveys, reviews, social media and just about any other natural language source, unearthing big-picture trends and patterns. It can help you understand how customers are feeling about a product, brand or experience. It can highlight the topics that matter most to your audiences, or pinpoint where to focus when you’re improving customer journeys.. And because it’s fully integrated with the wider Qualtrics platform, it makes taking action on your findings absolutely seamless.
Your business undertakes a wide range of activities, all in the name of customer satisfaction and profitability. But not all of them are created equal. More often than not, the success of your business rests disproportionately on certain factors more than others. Driver iQ uses powerful statistical analysis to uncover which aspects of your business matter most – so you can devote more of your energies to the drivers of key metrics like customer satisfaction, repeat purchase, brand advocacy and more.
There’s a whole world of statistical tools and processes out there. But knowing how to run them correctly, or even which ones you need to be running in the first place, is a mystery to many of us. Without statistical expertise on hand, many businesses skip the stats and miss out on the benefits. Stats iQ changes all that. It automatically selects the appropriate statistical tests you need to run, does the grunt work and delivers the information you need, simply expressed in plain English.
Predict iQ uses advanced deep learning to build a detailed picture of customer behaviour, so you can anticipate customers leaving your company even before they do. When you know a customer is at risk of leaving, you can reach out and repair the relationship before it reaches breaking point. Predict iQ helps you understand the drivers of churn, too, so you can reduce the chances of it happening in the future.
If text data is information-rich, voice data is an information goldmine. Voice interaction is one of the oldest and best-established channels for customer experience, but until recently the insights it holds have been difficult to access and use.
Voice iQ cracks the code, taking unstructured voice data and turning it into insights across EX, CX, BX, PX and more. It uses data mining, voice recognition and a sophisticated index of known customer effort markers to interpret and categorise your voice data.
The result: a clean, structured dataset that you can smoothly integrate with your wider experience management programs.
Glossary: Words and phrases related to predictive analytics
If you’re new to predictive analytics, you’re likely to come across a lot of unfamiliar language which can quickly start to feel confusing. Here we break down some common words and phrases in layperson’s terms. Some of these are related to predictive technology, and others may even be used synonymously or interchangeably with the term predictive analytics.
Technologies with this capability can learn and develop their understanding spontaneously, for example, a streaming service that learns what kind of film you like and recommends similar titles is using machine learning. There are two main types of machine learning – supervised (which requires human instruction) and unsupervised.
Artificial intelligence. While machine learning is based on the idea that machines should be able to learn and adapt through experience, AI refers to a broader idea where machines can execute tasks “smartly.” Artificial Intelligence applies machine learning, deep learning and other techniques to solve actual problems.
This is an advanced type of machine learning. A system – usually an artificial neural net – analyses information using multiple, sequential layers of processing. The layers work together to build a sophisticated understanding of complex ideas, images or processes. Deep learning can be used to analyse photographs and identify human faces, for example.
Data mining is the process of exploring large volumes of data and discovering patterns and trends within it. It’s a necessary part of predictive analytics, but not the whole story.
An algorithm is a set of instructions a computer follows in order to complete a task. In the context of predictive analytics, the algorithm is the basis of a predictive model – it becomes a model when trained on data.