Mastering Supply Chain Forecasting: Techniques, Technologies, and Pitfalls

Ever wondered how leading companies stay ahead of the curve in supply chain management? Tune in to uncover the secrets of accurate forecasting, a crucial component for success in today's fast-paced market. We promise you'll learn how groundbreaking technologies like AI, machine learning, and IoT devices are setting new standards for real-time data analysis, enabling companies to anticipate demand and optimize their operations. Discover the importance of addressing assumptions within forecasting models and learn practical strategies for mitigating risks through regular updates, scenario analysis, and stress testing.

In the second segment, we explore various forecasting techniques used in market research, from qualitative methods like market surveys and executive opinions to quantitative approaches including time series analysis and advanced machine learning algorithms. Learn how combining multiple methods can yield the most robust results and find out how to choose the right technique based on factors like historical data and forecast horizon. We'll also touch on the pitfalls of over-relying on historical data while ignoring external factors such as socio-political shifts and economic conditions. Our comprehensive and adaptable approach ensures your supply chain remains resilient and efficient. Don’t miss this insightful episode packed with expert advice and actionable strategies!
You may want to also listen to:
Chain Reaction Podcast Superforecasting and AI - Supply Chain Advantage

You can follow Chain Reaction on LinkedIn, Twitter and Facebook

Chapters

• 0:05 Importance of Forecasting in Supply Chain
• 8:29 Forecasting Techniques and Pitfalls

Transcript

Tony Hines: 0:05

Hello, tony Hines. Here You're listening to the Chain Reaction Podcast, all about supply chain advantage. Thanks for dropping by today. Great episode coming your way in just a few minutes. So stick around, stay tuned, find out more Now.

Tony Hines: 0:29

Have you ever wondered why your warehouse is over full, why your trucks are running all over the place to deliver products that aren't selling fast enough through stores, through outlets, through distribution networks? Well, could it be the forecast? Well, we're going to take a look at forecasts today and the impact that forecasts can have on your supply chain, and why accurate forecasts are absolutely essential. Well, the one thing that most organisations, when it comes to supply chains, are mainly concerned about is have they got the forecast right? And forecasting can be quite a big issue for many complex supply chains. If you've got thousands of SKUs, or millions of SKUs, then forecasting can become a very difficult process indeed. But, after all, it's digits, it's numbers on a page and we can do the maths. But it's whether the assumptions that go into the forecasting model are correct, accurate and do they lead to accurate forecasts. But it's whether the assumptions that go into the forecasting model are correct, accurate and do they lead to accurate forecasts. Well, the only way to be sure is to have real-time forecasting or forecasting closer to the point of sale, and then there's less margin for error. But how do you actually achieve that? If you're in a fast-moving consumer goods business forecasting can be quite challenging. In business-to-business it can be challenging, but not quite as challenging as FMCG products. There's usually more time and more uniformity, less risk, more stability in how goods are sold in B2B. In general, accurate forecasting, of course, is the key to any successful venture. By predicting future demand and trends, companies can optimize their operations, reduce cost and improve customer satisfaction. We're going to look at the importance of forecasting in supply chain management in this episode, but don't forget also I did an episode on super forecasting that you might want to go back and listen to, because that has some interesting points to make too. But we're going to look at forecasting here, focusing on technological advances, the use of real-time data and the assumptions that we build into forecasting models and how they change the results. Let's take a quick look at technology before delving into detail.

Tony Hines: 2:49

Advanced technologies have revolutionised supply chain forecasting. Traditional methods, which relied heavily on historical data and manual calculation, have given way to sophisticated algorithms and machine learning models that adjust in real time. Sometimes, these technologies can analyse vast amounts of data quickly and they accurately identify patterns and trends that would be impossible for humans to detect. Artificial intelligence and machine learning are at the forefront of the transformation, and AI-driven forecasting models can continuously learn from new data, which improves their accuracy over time. They're adaptable, they're dynamic models and they take into account consumer preferences and external factors. They're all poured into a mix through a funnel which generates results. Ai models, for example, can reduce errors by 30 to 50 percent, and that can lead to significant cost saving and improve operational efficiency.

Tony Hines: 3:49

Real-time data is another game changer, because we use real-time data to replace historical data. It provides up-to-the-minute insights into the current market conditions and allows companies to respond swiftly to changes in demand, supply disruptions or other unforeseen events. There's no one model fits everybody. Of course. You have to contextualise the forecasting model for your organisation to make sure it's working in your industry and for your business. Real-time data from Internet of Things devices can monitor inventory levels across the supply chain. They can look at production rates and transportation conditions and, as they ever change which they do, of course they can update the model. It can be fed into forecasting models to provide more accurate and timely pictures of what's happening in your supply chain, and companies can take decisions to adjust production schedules or reroute shipments to avoid potential issues, as in the case of the red sea problems with the huti attacks. They can very quickly say well, you can't go through sewers because that's too dangerous, therefore go around the cape of africa, and they can calculate the cost of doing that and the additional time it's going to take and how it's going to affect all the linked parts of that supply chain.

Tony Hines: 5:08

When it comes to forecasting, of course, one of the big factors are assumptions. While technological advances and real-time data have significantly improved forecasting accuracy, it's also important to recognize that all forecasting models are based on certain assumptions, and these assumptions can impact the results and should be carefully considered when interpreting forecasts. Common assumptions include the stability of the historical trends, the reliability of data sources and the absence of major disruptions or the possibility of all those things. However, these assumptions may not always hold true. For instance, a sudden economic downturn or a natural disaster can play havoc with the model and they can alter the demand pattern, and your previous forecasts, conducted in a different context, can all be rendered inaccurate. So forecasting is a tricky game, and actually the best forecasters have probably been at the game for some time. They know the way around and they know which variables need to be watched. To mitigate these risks, companies should regularly review and update their forecasting models, incorporate scenario analysis, stress test the model and help identify potential vulnerabilities to improve the robustness of those forecasts.

Tony Hines: 6:25

Another factor is that you can run different possible models to try and get a picture on which is likely to be the most probable and reduce reliance on any single assumption. Change the assumptions in the model and see what happens. What if this happens? What if I take away that disaster or that disruption in that area? How will that change our forecast? What happens if there's a big upturn in demand in one geographical area? And so on. We can play around with the model and, of course, the best forecasts will do that. So, all in all, forecasting is a vital component of supply chain management, enabling companies to anticipate demand, optimize operational performance and enhance customer service and satisfaction.

Tony Hines: 7:12

Technological advances such as AI and machine learning, along with real-time data, have significantly improved forecasting accuracy. However, it's essential to be aware that assumptions underlying the models need continuous refinement as conditions change. So if you want to stay ahead of the curve and navigate the complexities of the modern supply chain with greater confidence and agility. You need to apply technology, real-time data and make sure that the assumptions in the model are relevant. Now, believe it or not, there are techniques in forecasting that have been around for quite some time, and the split into qualitative forecasting techniques and quantitative forecasting techniques.

Tony Hines: 8:00

Let's take a look at some of the qualitative techniques that you may be interested in. The first is the Delphi method. This involves a panel of experts who provide their opinion and forecasts independently. The results are then aggregated and shared with the panel for further refinement until a consensus is reached and you can have different rounds in this Delphi method. So there'll be a panel, they'll discuss what the issues are, they'll create a forecast and then they'll refine it in iterations.

Tony Hines: 8:29

And then there's the market survey, traditional way to gather information and data directly from potential customers about their future purchasing intentions. Of course, how much attention customers or consumers pay to these market surveys is another matter. Sometimes they just want them out the way quickly, so they'll rush through them. So you might have set up the survey with all great intent and gone through all the testing processes very carefully, and you send it out. And then, well, the customers might trash it Not intentionally, but they don't take it as seriously perhaps as those issuing the survey do.

Tony Hines: 9:07

Then there's executive opinion. Senior executives use any experience and intuition to make forecasts. Well, that depends on the quality of the executive opinion, the executive team making the decision, and how up-to-date they are. And then there's the sales force. Composite Sales teams provide estimates based on their interactions with customers and market knowledge. So you've got all those qualitative techniques. Now the one thing that's important is not to rely on any single technique but to perhaps, like all good research, take multiple approaches to the problem. That can be expensive, of course.

Tony Hines: 9:45

Then we turn to quantitative forecasting techniques, and we've got four possible approaches. There's time series analysis, tried and tested. This method uses historical data, identifies patterns and trends over time, and then they use moving averages or exponential smoothing to bring it into the current and future positions. But again, it's the quality of the historical data that assumes that the past is a continuation into the future. Regression analysis involves identifying relationships between variables. Simple linear regressions have been used for many years, where you look at two variables, while multiple linear regressions looks at more than two variables and it is probably more comprehensive. But is it any more accurate? Usually it is, but it depends on what goes into the model, the regression model.

Tony Hines: 10:34

Then there are causal models, and causal models assume that the variables to be forecasted are influenced by one or more other variables. So there's an interaction between the variables. Sales might influence advertising spend, economic conditions and seasonality All in the mix. So you have to, in a causal model, pick out the variables that closely inform the way the model behaves, and accurately, of course. Then there's machine learning algorithms. These are advanced techniques like neural networks and decision trees that can analyze large data sets to identify complex patterns and make accurate forecasts. So there we have it Qualitative, quantitative and a different approach for the different types of techniques employed under both of those umbrellas.

Tony Hines: 11:23

Each of these techniques has its strengths and weaknesses, and the choice of method often depends on the specific context and available data. That's a very important issue. You don't always have the available data or access to the data you'd like to have, so you have to substitute in some way. Combining multiple methods will provide more robust forecast, as we've said, and it's important to think that one through and how much benefit it brings vis-a-vis the cost, choosing the right forecasting technique for your business involves considering several key factors to ensure the method aligns with your specific needs and context.

Tony Hines: 12:01

A guide might be, first of all, to look at the following seven steps. Determine the purpose of the forecast. Understand what you aim to achieve with that forecast. Are you looking to predict sales, manage inventory or plan for production? The purpose will influence the choice of technique and also the choice of data. Secondly, assess historical data. Evaluate the amount and quality of that historical data. Some techniques require extensive historic data, while others can work on limited data sets. Thirdly, consider the forecast horizon. Is it short-term, medium-term, long-term? And that might determine the type of analysis that you wish to use, adopt and the techniques that you want to select for the process of the forecast.

Tony Hines: 12:47

Complex methods like regression analysis or machine learning are very useful for long-term forecasting. Short-term forecasts might be more easily done with moving average techniques. Fourthly, we can think about the accuracy of the forecast and determine the level of the accuracy that we actually want to achieve. High accuracy might involve using advanced techniques such as machine learning, whereas less critical forecasts might be adequately served by simpler methods, and that keeps the cost down. You've always, fifthly, got to look at the cost and benefits, the resources that you'll need to conduct the forecast, and you have to weigh those up against the benefits, as we've said. The sixth thing is to understand the context. Take into account the specific context in which the forecast will be used. Factors such as market volatility, seasonality and external economic conditions influence the choice of the technique, and seven combine multiple methods Always an important thing to consider and it brings strength to the forecast. If you use combinations of forecasting methods, it's likely that the forecast will be more robust and you can see what each particular technique is bringing to the table, and you're likely to get more reliable results, and qualitative insights from experts, along with quantitative data, can be part of that multiple method mix.

Tony Hines: 14:18

There are, of course, a number of pitfalls that you need to avoid, and this is what experts talk about when they talk about making forecasts. The first thing is don't over-rely on historical data. If you do, it might be very misleading, because you can't assume that the past is going to continue into the future, particularly in rapidly changing environments or in an industry that's changed a lot. Well, you might have changed your processes and the offerings that you make to the market. You might have a lot of variety. You might not be comparing like with like when you're using that data, so you need to complement any historic data with current market conditions and any customer insights that you can glean about your products and services.

Tony Hines: 15:06

Another thing is ignoring external factors. Don't overlook what's going on externally. It's very important to think about the nature of geopolitics and how that is changing world commerce, and you can't just do this in a vacuum. Supply chains don't operate in a vacuum. They operate in the real world, where there's socio-political shifts, technological advancements or global economic challenges, all influencing how the market works. So you need to regularly look at the opportunities and threats and your particular strengths and weaknesses strengths and weaknesses keep a watchful eye on global and local trends and incorporate any of the relevant external factors into the forecasting model.

Tony Hines: 15:49

The third problem is over-complication. Sometimes, if somebody's technically competent, they might decide to build a mega-complex model and that can be difficult to understand. It's difficult to maintain. And also sometimes when that individual who set up that mega model moves on to another organization, maybe to your competitor the model falls apart because nobody else understands it. Nobody else is able to replicate it. So keep it simple. Keep the forecasting model as simple as you possibly can, without it being too simple. Focus on the key variables that materially impact the forecast. It's about materiality how material are the variables you're looking at? If you do an analysis of any forecast and say you've got six key variables, if you extend that to 100 key variables, you'll simply dilute the model and it may not bring anything extra in terms of accuracy to the forecast, but it will cost you a lot of money and it might be too complicated for those interpreting the model to make sense of. So always strive for simplicity in models. Better to go for simplicity and adjust.

Tony Hines: 17:02

Another pitfall is not to review the forecast, just keep running it time and time again without updating variables. Data sources and some of the assumptions that you've already built into the model may no longer be relevant. So you need to look at things like fresh data, consumer trends and changing market conditions. Another problem with forecast is confirmation bias, where you subconsciously seek out information that matches your existing beliefs about the market and that can skew the result and it can be very misleading. So don't get tied to or wed to a particular model. Challenge those assumptions that you're making and value other perspectives. Look for others who've got different views on the matter. Diverse perspectives are good. This is why it's often useful in a forecasting team to have diverse participants who've got different experiences and different backgrounds, who can bring something different to the assumptions on the table.

Tony Hines: 18:00

Over-optimism or over-pessimism is also a problem in forecasting. If you're too optimistic or you're excessively cautious, it could change that forecast immensely. So ground the forecast in the data, not in the optimism or the pessimism, and use scenario planning to prepare for the best case, the worst case and the most likely outcomes in any situation. Another problem can be poor data quality. Inaccurate or incomplete data can severely disrupt the forecast itself. So making sure that you've got accurate data that's readily available and can be updated is important, and make sure that the data you're using is valid for the situation.

Tony Hines: 18:41

Another factor that people mention is the lack of stakeholder involvement. If you exclude stakeholders, you might be biased or incomplete in your analysis of the problem for the forecast. So involve different departments, involve partners, involve the stakeholders and ensure that all angles are covered in making a comprehensive forecast. Other issues might be to do with seasonality forgetting about the seasonal patterns of particular demand, and it's essential to get that view of seasonality in the model, if it exists and also there might be cycles. Some products, you only sell them every one or two years. I mean, I might have different customers that keep a continuous line moving, of course, but if it's cyclical you know somebody only has to buy something from you every couple of years then you need to take that into account. Cyclical and seasonality issues have to be brought into the model, and the other thing about many models is that they become inflexible and they fail to adapt when conditions shift. So when you get new information, make sure there's a way that that new information, that new data, can be brought inside the model and you can update your model and change it. So those are just a few things to keep in mind, and here are some examples of businesses that have adaptable, flexible models when it comes to forecasting, and I'm just going to tell you about one or two examples.

Tony Hines: 20:21

Do you remember during the pandemic, when they had the image of the Mona Lisa with toilet paper and they were saying that toilet paper demand had gone through the roof and there was a shortage? Well, amazon, with its AI-driven predictive forecasting model, was quick to respond to this unexpected surge in demand. It was 213%, apparently, for toilet paper sales, but their model adapted swiftly to the new demand levels on real-time data using AI, and they were able to meet customer demand, even though this was an unprecedented situation. Walmart has used big data and advanced analytics to improve its demand forecasts. It analyzes vast amounts of data from various sources, which include sales transactions, weather patterns, social media trends and customers changing habits, and Walmart can predict demand with high accuracy. That helps them to optimise inventory levels, reduce stockouts and improve overall service to customers, maintaining supply chain efficiency. Then there's the classic fast fashion companies like Zara, and in the past they've used real-time data from stores to generate forecasts based on the changing patterns. They understand customer preferences fast and they can adjust their production and inventory modelling very quickly. And that's, of course, what fast fashion was all about the throughput model based on real-time data.

Tony Hines: 21:49

It's important in grocery retailing too, because you want to predict demand for perishable items like milk and fruit and vegetables, so that you're not left holding excessive amounts of stock past the sell-by date, because if you do, that's waste. You think you had all that cost to get those goods on ships to stores and available for customers, but then, because you've overestimated on the forecast, you've got too much in store and it doesn't sell. So forecasts are important. Happy forecasting. Well, that's it for this week.

Tony Hines: 22:36

I hope you've enjoyed the episode on forecasting and managed to glean some tips that will help you produce your next forecast. At least it's food for thought, isn't it? Some of the issues and some of the thinking around the forecast, around the process of doing a forecast, and so I hope that's helpful, and if, of course, you're a student of supply chain, I'm sure the talk about forecasts will help you put forecasts into context in supply chain strategies. You can see they are very important indeed. Well, we have over 250 episodes of the Chain Reaction Podcast, and you can stop by and pick those up on demand. Play on demand Pod Cast. Yeah, I'm Tony Hines, I'm signing off and I'll see you next time in the Chain Reaction Podcast, and don't forget to subscribe so you'll be first to know when a new episode drops Bye. For now, so you've been listening to the chain reaction podcast, written, presented and produced by tony hines.