Artificial Intelligence and Machine Learning: What does it (AI)L mean?

Written by Inigo Bridle
Integrated Business Planning Consultant at I-Plan. Implements centralised supply chain, sales and operations planning and demand and supply planning. Strategic planner with a track record of successfully implementing business turnaround and market growth projects in a highly competitive environment.
21st October 2020

What Does It (AI)L Mean?

There is an ever-increasing amount of interest and content appearing around Artificial Intelligence (AI) and machine learning (ML). In the UK, BBC 4 (1) has shown several programmes about the subject, The Royal Society ran a year-long series of events called You and AI (2) and in the current pandemic, AI and ML are being mentioned in relation to predicting local outbreaks to developing vaccines. 

With all the interest, I have been trying to understand what all the fuss is about and how can it help with the development of Sales and Operations Planning software and processes? 

To begin with, a definition of AI and ML would be useful. These two items when discussed seem to be interchangeable, but is this true? Intelligence is the ability to acquire and apply knowledge and skills. Artificial Intelligence is generally defined as intelligence demonstrated by machines, rather than natural intelligence demonstrated by animals. Machine Learning is the description of the process by which machines acquire the skills and knowledge to be applied. Therefore, AI and ML are linked but not the same. So, what next?

In the world of Sales and Operations Planning, (S&OP) the goal is to determine the best possible plan for the business into the future, balancing demand and supply to achieve maximum profitability in most cases. We know, however, that trying to predict future events is not an exercise that is 100% accurate. There is variability within any process and S&OP is no different. Given the variability, could these new ideas of AI and ML help at all? Of course, is the simple answer.

The techniques by which machines learn with AI is applied in the main via mathematical algorithms. These are a process or set of rules to be followed in calculations or problem-solving operations. The likelihood is that the algorithms used in ML for applications could predict future behaviour given a certain set of parameters with which to operate. If this sounds familiar to those with experience in forecasting software tools, it is because that is exactly what tools like I-Plan are doing today and have been doing for over twenty years.

Can we also use ML and AI to help with the profitability aspects of planning the operations? For a given set of variables and constraints could an algorithm calculate the most profitable solution for a business? To quote the song, “Stop me if you think you’ve heard this one before..” 

The interesting thing about what is happening now, ‘ the peak of the hype curve..’, as it was described by Dame Wendy Hall (3) at an event in 2018 (4), is that the discussions happening now about the application of the tools are based on academic research from 15 to 20 years ago. For tools such as I-Plan, it is precisely this research that was quickly adopted to solve the complexity found in continuous manufacturing Sales and Operations Planning calculations.

There are areas for improvement in learning and new techniques are being developed all the time and these should not be ignored. In round Four of the Makridakis Competition (5), which looks at forecasting performance, the findings included: 

  • Hybrid methods, utilizing basic principles of statistical models and ML components, have great potential. 
  • Combining forecasts of different methods significantly improves forecasting accuracy, and  
  • Pure Machine Learning methods are inadequate for time series forecasting. 

The Round Five results (6) published on 6th October 2020 conclude that ML methods are in use today in retail sales forecasting as a sector. Other interesting findings include the fact that the winner of the competition was not an experienced forecaster, but rather a student with a high experience of using ML to analyse data. Overall, the findings from this round of the competition were that all the top performers included some elements of ML in their solutions. The competition concludes that to improve forecast accuracy the following processes are key:  

  • Combining
  • ‘Cross-Learning’ 
  • Cross-Validation

It is interesting to see these practices described as useful for ML processes as for any business operating and S&OP process, these processes are also key for generating a collaborative and validated plan for the business. These headings could also be applied to people as well as to the software utilising the data. 

The management by ML for these processes is in part due to the development of technology and computing power. This means that the huge data sets available for learning are now able to be managed and utilised at greater and greater speeds. The theories and methodologies being used are not necessarily cutting edge, but now the tools exist to allow these practices to reach their full potential.   

These developments in technology and understanding can only bring benefits to the adoption of the techniques and technology required in complex decision-making processes such as Sales and Operations Planning.  







(6) Makridakis, Spyros & Spiliotis, Evangelos & Assimakopoulos, Vassilis. (2020). The M5 Accuracy competition: Results, findings and conclusions. 

More information: 

Give I-Plan a go

Get started with a free demo