The Role of AI in Predictive Analytics: Is it AI or maths that is revolutionising businesses data-driven decisions?

In the vast landscape of business technology, the surge of AI (Artificial Intelligence) has become akin to a modern gold rush.  The term is everywhere you look, promising predictions and forecasts that will provide veritable mines of information to improve the company’s bottom line.  So before you get swept up in a whirlwind of promises and shiny applications, just remember that at the heart of predictions are the powerful disciplines of mathematics and statistics.  In fact whether it’s a model, algorithm, machine learning, deep learning or even the fabled “AI”, at its core there is mathematics and statistics. 

Before AI became an everyday term, business of all sizes were using predictive analytics across a wide range of functions: predicting sales, inventory management, chatbots to interact with customers, workflow management, driver route optimisation and many other areas where maths and statistics could find patterns within data.  These used a wide range of approaches such as regression analysis, time series analysis, and decision trees and others besides.  Such approaches are still valid – they are robust and rigorous approaches that do not pretend to mimic human intelligence or learn on their own accord (i.e they are not AI!) 

There is no denying that newer approaches such as machine learning (ML) has increased the breadth and depth of insights and information available to businesses.  What’s important to remember is that commonly used ML techniques such as random forests or scalable vector graphs aren’t magic.  Rather they are maths, driven by data and a defined set of human parameters.  They can’t learn by themselves and aren’t applicable out of their training data set.  Machine learning and statistics combined can deliver very powerful results which Bays uses in our Home Hazard Predictions

Deep learning, whilst being what many may think of as “AI” is still grounded in mathematics and statistics, with the algorithms using layers of neural networks to mimic (as far as possible) the human thought process.  These techniques need large amounts of data and are most commonly used in areas such as image or language processing, which might not be considered traditional predictive analytics for businesses. 

AI is touted as a panacea for almost everything, but instead of believing all the hype it’s important to remember a few key points when looking to create evidence based decisions: 

  • Most insights, trends and patterns associated with business intelligence come from a spectrum of techniques, all based in mathematics and statistics 
  • Machine learning and deep learning can offer new insights and information, but don’t forget to start with the rigorous and robust approaches offered by mathematical and statistical modelling. 
  • Before starting on any predictive analytics, take time to evaluate your data (volume, completeness, storage, accessibility and terms of usage) 
  • Remember to assess the computational costs of any chosen technique 
  • Determine how important it is that the developed tool is explainable (ML tools are not completed explainable in how they come to a result) 

If you take time to look at the full breadth of predictive analytic tools and techniques, your business will be able to better align their decision-making processes with their actual needs.  Using more traditional approaches can not only be cost-effective but also explainable and truthfully sufficient for the business need.  Moreover, developing an inhouse understanding of predictive techniques will allow your business to develop a clear strategic investment plan of where to use ML and maybe even deep learning to achieve your goals which need deeper, more nuanced analysis. 

Predictive analytics remains a cornerstone of informed decision-making in the digital age.  However, by understanding the rich tapestry of mathematics, statistics, machine learning, and deep learning, businesses can employ the right combination of tools to harness its true power.  That results in data-driven decisions being solid, robust, defendable and based in maths, not magic.

By Sophie Carr

Leave a comment

Your email address will not be published. Required fields are marked *