It goes without saying that the Covid-19 pandemic is the largest health crisis that the NHS has ever faced. Arguably it has also presented one of the most unprecedented challenges ever faced by the retail industry. While two very different sectors with their own unique threats, there are some valuable lessons that retailers can glean from healthcare on one problem in particular: demand forecasting and operations planning.
Let’s start with some context. The Covid-19 pandemic has driven a massive surge in demand for critical resources across nearly every healthcare system in the world. As part of the data response to the pandemic, we worked with the NHS to develop a predictive tool called the Early Warning System (EWS). It is a first-of-its-kind AI solution that is based on a technique called Bayesian Hierarchical Modelling (BHM). Using aggregate data (e.g. Covid-19 positive case numbers, 111 calls and mobility data), the EWS is able to accurately forecast cases weeks in advance so hospitals can divert staff, beds and vital equipment, such as PPE, where it is needed most.
To make accurate predictions, standard approaches to forecasting typically require years of historical data and consistent patterns and trends. Neither exists for a situation as unprecedented and rapidly-changing as the pandemic. What makes the EWS so special is its ability to bring together external contextual information and learn local patterns between geographies. This allows for accurate predictions, down to a hospital level, even as things are evolving extremely rapidly.
The NHS response to the pandemic is a great example of how organisations can use data and machine learning to make accurate predictions about the future and to inform more effective decision-making – even during highly volatile conditions. The data techniques applied in the EWS – and the predictive analytics used to better manage life and death decisions – carry important lessons for retailers as the economy opens up.
With over 80% of the UK adult population now fully jabbed, countries coming off the red list every day, and case numbers falling, there is hope that a “new normal” is emerging. Weddings and group get-togethers are firmly back on the agenda, shopping trips planned, and consumers are looking for ways to spend or mis-spend their lockdown savings.
This is good news for retailers – and businesses at large. But moving from economic stagnation to a huge boom of economic activity over such a short timespan is not normal and introduces masses of uncertainty. Few brands (if any) have ever managed a transition like this before and it is still not clear which sectors will see the bulk of the benefit as the economy bounces back. Will shoppers flock back to high streets? Or will they jump on a plane in search of sunshine? What is more, the recent supply chain disruption and labour shortages exacerbated by COVID and Brexit has caught many retailers off guard. The Office for National Statistics recently reported that 27% of food and accommodation firms have experienced lower than normal stock levels and Richard Walker, Managing Director of Iceland, has warned that these shortages could even “cancel Christmas”.
From flu vaccine suppliers to food and drinks brands, firms are struggling to meet customer demand, and the impact of these challenges are reaching consumers too, with the latest figures from the BRC and NielsenIQ revealing a 0.4% month-on-month rise in UK shop prices in August. Predicting how all of these uncertainties will play out poses a huge challenge for retailers. If brands want to navigate the uncertainty successfully, prepare effectively, price optimally and allocate resources accordingly, it’s critical that they set themselves up properly when it comes to the way they use data. Ultimately this means deploying similar capability that the NHS did in developing the Early Warning System.
One of the biggest obstacles for forecasting demand is data. Most demand forecasting approaches used by retailers are heavily reliant on historical data, using past trends and cycles in purchase behaviour to predict future trends. Some companies might have incorporated some external data sources to get a better handle on the influence of macroeconomic effects or seasonality, but even these companies still run into challenges. This is because traditional systems used for demand forecasting used by many retailers fail to account for the issues of data sparsity, scarcity, and uncertainty that inevitably exist in the real world.
Over the last year for example, we have seen volatility in global supply chains driven by both the pandemic and Brexit, as well as dramatic and unexpected shifts in consumer behaviour. What this means in practice is that traditional statistical approaches to forecasting will struggle to account for sudden pivots in purchasing trends that the economic recovery will bring, risking inaccurate demand forecasts, and ultimately excess supply chain costs, and either excess inventory or insufficient stock.
The key to navigating uncertainty for brands is to focus on building the best possible understanding of the underlying factors that drive demand for their products. AI and machine learning are a great solution for this as they offer the possibility to incorporate a much broader range of internal and external sources of data than traditional techniques. The latest generation of ML methods are able to drive as much as a 20 percent improvement in the accuracy of demand forecasts. For example by incorporating: high-street footfall, consumer search activity, website engagement, and weather data in addition to traditional internal data like sales data and marketing activity. Crucially, machine learning can learn the subtle correlations between products, geographies, and this rich range of data, which is extremely valuable in making more accurate and granular forecasts during times of uncertainty.
Why are product and geographical trends important? To use the classic lemonade stand example, let’s say you are a lemonade seller but are trying to add brownies to your range. In the first instance, you will have zero data about demand on your street for brownies. However, because you have ongoing sales and lots of historical data for the lemonade that you already sell, you will pretty quickly be able to start seeing how sales of brownies correlate with your lemonade sales. This creates an opportunity for you to get better at forecasting demand for brownies than you would have been able to if you were starting from scratch. This is where product trends can be key to demand forecasts.
Now, imagine you are going to start selling lemonade on different streets and maybe even in the next town over. Clearly, the demand dynamics are going to differ somewhat – there might already be a lemonade stall in the area or different numbers of dog walkers, but the base level data such as weather and time of year will probably have a similar impact on demand in these different places. Similar streets are also likely to share similar demand profiles. This creates the opportunity to learn about how demand might pan out in your new target streets more quickly, thanks to what you already know about the first. This is how geographical trends can come into play.
With this knowledge, brands aren’t restricted to long-term forecasting based on a high-level understanding of customer behaviour and aggregate sales data. Instead, they can monitor these more granular metrics in real time, forecast short term demand fluctuations, and react accordingly by making optimisation decisions around pricing and logistics in a precise and highly-targeted way.
That real-time insight is vital now more than ever; with the gap between restrictions having been eased but before any “new normal” being established likely to be the most uncertain period yet of the last two years.
The Covid-19 pandemic has exposed weaknesses in forecasting systems and the instincts of traditional ways of doing business across the world, but it has also presented an opportunity to re-set, innovate, and radically improve the economics and efficiencies of forecasting tools. If the success of the NHS Early Warning System is anything to go by, the best brands are getting ahead by using these techniques to predict demand in a way that is more robust, more flexible, and built on a solid foundation of data analysis. If it saved lives for the NHS, it will save money and improve customer experience for retailers.
Author:
Josh Muncke, Director of Consumer Business, Faculty
You are in: Home » Guest Comment » GUEST COMMENT Demand forecasting: What can retailers learn from the healthcare sector?
GUEST COMMENT Demand forecasting: What can retailers learn from the healthcare sector?
by Josh Muncke
It goes without saying that the Covid-19 pandemic is the largest health crisis that the NHS has ever faced. Arguably it has also presented one of the most unprecedented challenges ever faced by the retail industry. While two very different sectors with their own unique threats, there are some valuable lessons that retailers can glean from healthcare on one problem in particular: demand forecasting and operations planning.
Let’s start with some context. The Covid-19 pandemic has driven a massive surge in demand for critical resources across nearly every healthcare system in the world. As part of the data response to the pandemic, we worked with the NHS to develop a predictive tool called the Early Warning System (EWS). It is a first-of-its-kind AI solution that is based on a technique called Bayesian Hierarchical Modelling (BHM). Using aggregate data (e.g. Covid-19 positive case numbers, 111 calls and mobility data), the EWS is able to accurately forecast cases weeks in advance so hospitals can divert staff, beds and vital equipment, such as PPE, where it is needed most.
To make accurate predictions, standard approaches to forecasting typically require years of historical data and consistent patterns and trends. Neither exists for a situation as unprecedented and rapidly-changing as the pandemic. What makes the EWS so special is its ability to bring together external contextual information and learn local patterns between geographies. This allows for accurate predictions, down to a hospital level, even as things are evolving extremely rapidly.
The NHS response to the pandemic is a great example of how organisations can use data and machine learning to make accurate predictions about the future and to inform more effective decision-making – even during highly volatile conditions. The data techniques applied in the EWS – and the predictive analytics used to better manage life and death decisions – carry important lessons for retailers as the economy opens up.
With over 80% of the UK adult population now fully jabbed, countries coming off the red list every day, and case numbers falling, there is hope that a “new normal” is emerging. Weddings and group get-togethers are firmly back on the agenda, shopping trips planned, and consumers are looking for ways to spend or mis-spend their lockdown savings.
This is good news for retailers – and businesses at large. But moving from economic stagnation to a huge boom of economic activity over such a short timespan is not normal and introduces masses of uncertainty. Few brands (if any) have ever managed a transition like this before and it is still not clear which sectors will see the bulk of the benefit as the economy bounces back. Will shoppers flock back to high streets? Or will they jump on a plane in search of sunshine? What is more, the recent supply chain disruption and labour shortages exacerbated by COVID and Brexit has caught many retailers off guard. The Office for National Statistics recently reported that 27% of food and accommodation firms have experienced lower than normal stock levels and Richard Walker, Managing Director of Iceland, has warned that these shortages could even “cancel Christmas”.
From flu vaccine suppliers to food and drinks brands, firms are struggling to meet customer demand, and the impact of these challenges are reaching consumers too, with the latest figures from the BRC and NielsenIQ revealing a 0.4% month-on-month rise in UK shop prices in August. Predicting how all of these uncertainties will play out poses a huge challenge for retailers. If brands want to navigate the uncertainty successfully, prepare effectively, price optimally and allocate resources accordingly, it’s critical that they set themselves up properly when it comes to the way they use data. Ultimately this means deploying similar capability that the NHS did in developing the Early Warning System.
One of the biggest obstacles for forecasting demand is data. Most demand forecasting approaches used by retailers are heavily reliant on historical data, using past trends and cycles in purchase behaviour to predict future trends. Some companies might have incorporated some external data sources to get a better handle on the influence of macroeconomic effects or seasonality, but even these companies still run into challenges. This is because traditional systems used for demand forecasting used by many retailers fail to account for the issues of data sparsity, scarcity, and uncertainty that inevitably exist in the real world.
Over the last year for example, we have seen volatility in global supply chains driven by both the pandemic and Brexit, as well as dramatic and unexpected shifts in consumer behaviour. What this means in practice is that traditional statistical approaches to forecasting will struggle to account for sudden pivots in purchasing trends that the economic recovery will bring, risking inaccurate demand forecasts, and ultimately excess supply chain costs, and either excess inventory or insufficient stock.
The key to navigating uncertainty for brands is to focus on building the best possible understanding of the underlying factors that drive demand for their products. AI and machine learning are a great solution for this as they offer the possibility to incorporate a much broader range of internal and external sources of data than traditional techniques. The latest generation of ML methods are able to drive as much as a 20 percent improvement in the accuracy of demand forecasts. For example by incorporating: high-street footfall, consumer search activity, website engagement, and weather data in addition to traditional internal data like sales data and marketing activity. Crucially, machine learning can learn the subtle correlations between products, geographies, and this rich range of data, which is extremely valuable in making more accurate and granular forecasts during times of uncertainty.
Why are product and geographical trends important? To use the classic lemonade stand example, let’s say you are a lemonade seller but are trying to add brownies to your range. In the first instance, you will have zero data about demand on your street for brownies. However, because you have ongoing sales and lots of historical data for the lemonade that you already sell, you will pretty quickly be able to start seeing how sales of brownies correlate with your lemonade sales. This creates an opportunity for you to get better at forecasting demand for brownies than you would have been able to if you were starting from scratch. This is where product trends can be key to demand forecasts.
Now, imagine you are going to start selling lemonade on different streets and maybe even in the next town over. Clearly, the demand dynamics are going to differ somewhat – there might already be a lemonade stall in the area or different numbers of dog walkers, but the base level data such as weather and time of year will probably have a similar impact on demand in these different places. Similar streets are also likely to share similar demand profiles. This creates the opportunity to learn about how demand might pan out in your new target streets more quickly, thanks to what you already know about the first. This is how geographical trends can come into play.
With this knowledge, brands aren’t restricted to long-term forecasting based on a high-level understanding of customer behaviour and aggregate sales data. Instead, they can monitor these more granular metrics in real time, forecast short term demand fluctuations, and react accordingly by making optimisation decisions around pricing and logistics in a precise and highly-targeted way.
That real-time insight is vital now more than ever; with the gap between restrictions having been eased but before any “new normal” being established likely to be the most uncertain period yet of the last two years.
The Covid-19 pandemic has exposed weaknesses in forecasting systems and the instincts of traditional ways of doing business across the world, but it has also presented an opportunity to re-set, innovate, and radically improve the economics and efficiencies of forecasting tools. If the success of the NHS Early Warning System is anything to go by, the best brands are getting ahead by using these techniques to predict demand in a way that is more robust, more flexible, and built on a solid foundation of data analysis. If it saved lives for the NHS, it will save money and improve customer experience for retailers.
Author:
Josh Muncke, Director of Consumer Business, Faculty
Read More
You may also like
Register for Newsletter
Receive 3 newsletters per week
Gain access to all Top500 research
Personalise your experience on IR.net