Demand sensing

what’s it all about?

With global market volatility at unprecedented levels, supply chain optimisation has never been more important, delivering the right goods at the right time for customers. This has been brought into stark focus with the current COVID situation, causing massive surges in demand for products (such as toilet paper earlier on this year) while demand for other products fell off a cliff. To provide an example, my go-to retailer for office wear has sent through so many promotional reductions their shirts must be selling for less than cost price.

How can organisations with complex international supply chains respond effectively to such sudden and major disruption? That’s where demand sensing comes in. Demand sensing differs from traditional demand forecasting by leveraging machine learning to accurately understand the impact of external factors on demand in the short term – around 4-6 weeks and utilises new sources of data. These new sources of data, known as demand signals, can be both internal variables such as sales data from stores and external sources such as social media and web site interactions, and are analysed in almost real time. This means organisations can use new sources of data to help better understand what shocks to demand and supply are doing to their supply chain, and how to react to critical and unexpected events in the marketplace. As a result, demand sensing works alongside more traditional demand forecasting as we know it, rather than replacing it.

A quick introduction to machine learning-based forecasting

Machine learning-based forecasting automates supply chain planning processes, meaning that there is not the need for high-touch, manual processes to come to decisions on what stock levels are required at a particular time for an organisation and why.

Machine learning-based forecasting reduces the human interaction required throughout the planning cycle, reducing time spent on the aggregation, cleansing and analysis of data to provide decisions in seconds and without the biases which can emerge in human data analysis. As a quick comparison, some R-based packages are able to carry out 1.5million forecasts per hour, underlining the speed of results in a machine learning-based system.
With information at their fingertips, employees have greater capacity to be able to work on more skilled tasks, such as exception management and innovation initiatives. Additionally, with more and more complex supply chains spanning countries around the globe, machine learning-based forecasting can join the dots across a range of data sources and promote collaboration in planning across territories. Furthermore, territories can be even more granular in the way they understand demand and how they plan for this with the ability to even drill down to specific Stock Keeping Units (SKUs). This added assurance provides organisations with inventory performance raised by up to 20%, while hitting or exceeding target service levels. Additionally, in monetary terms, companies have seen an average of 24% improvement in their gross profit advantage across their supply chain. With higher degrees of certainty, organisations are able to reduce logistics costs as there is less need for short-term changes to a supply chain and the expediting of goods and transport which reduces bottom-line spend on supply chain and even manage capacity of factories more effectively by reducing expenditure on overtime and short-term agency employees.

How does demand sensing add value to the machine learning-based forecasting equation?

Demand sensing helps organisations to improve the accuracy of machine learning based systems to enable a better reaction to sudden changes in behavior in the short-term. Therefore, organisations can overlay demand sensing onto existing machine learning based forecasting systems and see a dramatic improvement in short-term forecasting when there is volatility in the market. Of course this is a major challenge for organisations currently, and demand sensing brings an alluring solution to the issue by typically offering a 30 – 40% reduction of forecasting errors.  The two pieces of technology complement each other and improving the capability of an organisation to have clearer understanding of current positions and how to respond.

Why is demand sensing so relevant in a post-COVID world?

The COVID-19 situation has highlighted and sharpened the business case for machine learning driven forecasting and demand sensing. With planning cycles being reduced into days and weeks rather than months, organisations need to be sure that they are making the right decision in response to the challenging circumstances in the short term. Organisations can rapidly respond with a higher degree of confidence to these shocks using Machine Learning/Demand Sensing.

What are organisations doing in this space now?

With a high degree of confidence in short-term forecasting thanks to demand sensing, organisations can reap significant benefits. This is particularly relevant for organisations within the Fast-Moving Consumer Goods, Consumer Products Group, Supermarkets and Fashion industries, where speed and shorter planning cycles are critical. Nike, who just last year acquired the AI platform Celect, have made a major push to transfer their supply chain into a digitally enabled solution, with an emphasis on demand sensing with the view to build a digital operation that “others won’t be able to match” according to John Donahoe, the CEO of Nike. And they aren’t alone – with estimated global annual spending on AI by retailers exceeding $7.3 billion by 2022, there is already significant investment in this advanced technology. We at Capgemini have seen companies begin to take the steps to land pilots and start scaling, with clear benefits on show in speed, accuracy and quality of data across a variety of sectors with complex and multi-national supply chains.

It should be noted that accuracy is one thing but being able to act on it is quite another! Organisations need to adapt and be agile so as to respond to more granular and precise planning outputs and of course ensure they need to set up their pre-built models correctly with quality data in the first place. But harnessing the power of these latest planning techniques will bring significant competitive advantages in our ever more complex and unpredictable world.

Author

---

Ryan Abraham

Ryan is a consultant within the Operations Transformation team at Capgemini Invent and has worked across a variety of sectors in transformation projects. Ryan has a keen interest in automation and is part of the internal automation workstream within OT.