A few weeks ago, we had the opportunity to attend the International Symposium on Forecasting (ISF) in Dijon. It was great to engage with forecasting experts from industry and academia and to gain fresh perspectives on supply chain demand forecasting. One of the things that makes ISF stand out is its holistic approach to this topic. Yes, there is a significant focus on the technical aspects, such as new model architectures. But we all know that the forecast generated by a forecast engine is seldom the final forecast that ends up being used as input to make decisions within a business. Demand planners play an important role in enriching this “baseline” forecast by (if done well) adding information that was not captured in the model. This aspect was also given considerable attention at ISF – thanks among others to ongoing research by Robert Fildes and Paul Goodwin. 

At EyeOn we are also passionate about optimizing the demand planning process as a whole; making sure the overall process is efficient, and each component is adding value. In this blog, we will present our views on some effective best practices. 

Evaluating the quality of the entire demand planning process 

Most organizations measure the quality (i.e., forecast accuracy and bias) of the end result of the demand planning process; often referred to as the “final” or “consensus” forecast. However, these metrics alone don’t capture the value added throughout the process. For instance, you might be satisfied with an 80% forecast accuracy, but if the baseline forecast accuracy was 85%, you’ve invested valuable time and resources only to diminish the forecast quality. This example highlights the importance of tracking the Forecast Value Add (FVA) of enrichment: that is, to what extent are demand planners improving the baseline forecast. 

Monitoring only the FVA, however, isn’t enough. Imagine a scenario where the FVA is 10 percentage points, indicating that demand planners are excelling at improving the baseline forecast. Does this mean the overall demand planning process is flawless? Not necessarily. It might be that your forecast engine is underperforming; leading demand planners to spend considerable effort on enrichments that a higher-quality forecast engine could have handled more efficiently. In other words, if the forecasting engine were better, the planners wouldn’t need to spend as much time on adjustments. This example emphasizes the need to also evaluate the quality of your forecast engine by comparing it to a simple benchmark (e.g., by comparing the forecast accuracy of your forecast engine to a naïve forecast). In short, to assess the effectiveness of our overall demand planning process, we must evaluate the quality of all the individual components. 

But there’s more to consider 

While having these basic insights is a good starting point, they may not necessarily offer guidance on how to improve the quality of your enrichments. For this, you need to track details of the enrichment process, such as the number and type (e.g., direction, magnitude) of enrichments. This allows you to gain perspective on: 

• Which type of enrichments have historically been associated with positive/negative value add 

• Which parts of your product portfolio benefit most from demand planner interventions 
• The time invested in the enrichment process 
• Potential biases in enrichments

From insights to action 

Insights are valuable, but they should be in service of improving the process. Data-driven insights in the performance of the baseline forecast engine and the behavior of the demand planning team should provide concrete suggestions for process improvements, ultimately leading to: 

• Improved quality of enrichments, resulting in higher forecast value add and a better demand plan 
• Enhanced efficiency of your demand planning process through the adoption of a more targeted enrichment strategy
   

How EyeOn can support 

Making the value-add in forecasting tangible is at the core of what we do. And we offer multiple ways to get experience with it or even get started in a fast way.  

• Play our Forecast Game: a business game where you compete against others to create the best possible forecast, applying best-practice principles around forecast enrichments. Measuring and learning on forecast performance is at the heart of the game set-up.  

• Use our Smart-touch dashboard: a ready-to-use dashboard that provides all the insights discussed in this blog. Let’s connect your data and unlock direct insights to improve your forecasting and demand planning process.  

• Get your copy of our Statistical Forecasting e-book: ‘Statistical Forecasting as your steppingstone towards AI’. In the e-book, we explain how to set up statistical forecasting to create an optimal foundation for machine learning-based demand planning. Download it here. 

If you’d like to learn more, feel free to reach out to Rijk van der Meulen or Erik de Vos.

The post Are all stages of your demand planning process adding value? appeared first on EyeOn.

In our previous blog on the value of statistical forecasting, you’ve discovered that introducing statistical models in your demand planning setup will help you kickstart your forecast automation journey. In this blog, let’s take a closer look at what it means to incorporate statistical models in demand planning. First, it is good to emphasize that statistical forecast models alone are rarely enough. If you look at the entire portfolio of the company, there will be parts that will already benefit from statistical forecasting alone, while there will also be parts that require enrichment of the forecast to cover all demand dynamics.   

The typical setup with statistical forecasting is the forecasting approach that combines statistical forecast models with human enrichment. This requires a certain setup in terms of process, people, data, and tools. Let’s focus here on what the core forecasting setup would look like.  

In a setup with statistical models in demand planning, the starting point will always be historical sales quantities. Consecutively, several core activities will need to be part of the forecasting setup.   

1. Ensure full sales data availability 
2. Validate the categorization of the portfolio and the demand 
3. Clean the sales history so only baseline related sales remains 
4. Organize seasonality detection and application 
5. Select the best-fit statistical model based on a back-tested performance measure 
6. Organize for forecast enrichment 

The potential of using statistical forecasting in a demand planning process is mostly determined by the value it adds compared to the current way of working: where does best-in-class statistical forecasting already outperform the current process, or is it on par with a fraction of the effort? But to get the most out of statistical forecasting, it is best to first understand what drives demand in the business.   

Before you start with statistical models in demand planning, we encourage you to consider a few pre-analysis steps. These steps will help you understand how to set up your statistical design. Often this is related to what you already know about your portfolio, but seeing it quantified gives you the confirmation you need to move forward. We distinguish 4 pre-analysis activities in the creation of the statistical forecast. These 4 pre-analysis activities are outlined below.

1. Business characteristics 

Statistical forecasting requires structured master data. Usually this means developing a hierarchy setup both for the product and customer angles that matches the business for demand forecasting. Often ERP hierarchies are used at the start, and equally often it turns out that these hierarchies contain gaps to perform well in forecasting. This should not be surprising as ERP hierarchies are constructed to serve the key purposes of the ERP system and not necessarily demand planning.  

Before you start with statistical forecast models, execute a hierarchy screening and validate that you are combining the right products, that you have a consistent hierarchy built-up and that you know how to deal with complex products. By doing this, it will open the discovery on the hierarchy level on which you will probably install your statistical forecasting.  

Next to the product and customer aggregation level decision you need to take before setting up statistical forecasting, you also need to determine your forecast time bucket set-up. Whether you use calendar month, week, day or something else, at least get your definitions straight and align them to how your business works.

2. Categorization

We want to start with the element of categorizing your portfolio. In essence, categorization will not make you a better forecaster, but it should make you realize how diverse your portfolio is and what the implications are for forecasting.   

To explain this further, we want to introduce the concept of a Demand Forecasting Unit (DFU). The level at which a statistical forecast is made, i.e. a product or a combination of product and region or even product, customer and region, is called a Demand Forecasting Unit (DFU). The higher the level of aggregation, the fewer the number of DFUs and the more accurate the statistical forecast for each DFU. However, you may want to select a lower level of aggregation if it better suits the planning steps in your S&OP process (for example, promotion planning, your supply, raw material, or capacity planning). A categorization matrix distinguishes between New Product Introductions (NPIs), End of Lives (EOLs), and active Demand Forecasting Units (DFUs).   

NPIs are defined as DFUs that have been introduced in the last few months or are about to be introduced. With limited data points, statistical forecast models cannot provide a reliable future statistical forecast based on DFU data alone. In a setup with statistical forecasting, you are left with a few options:   

1. Forecast them fully manually 
2. Forecast them based on a set of defined parameters like distribution, shelf space, loading in period.  
3. Forecast them using reference products on which you apply enrichment 
4. Forecast them using NPI profiles based on historical sales patterns from other NPI 

EOLs are defined as DFUs that have not been sold in a certain period or that have been marked for discontinuation. Forecasting is not a priority for these, although it helps to determine the service and inventory risks you may face. Forecasting should be kept simple, ideally without a lot of enrichment, and with a clear point in time when these will no longer be forecast, indicating the actual point in time when the DFU will stop. EOLs are determined by the specific dynamics of your business and can often be based on agreements with your customers. EOLs will require some manual intervention and especially follow-up to ensure a smooth phase-out.  

For the group of mature DFUs, a category is determined that represents the Pareto of total value (margin, revenue, or sometimes volume if there is no price or value information) with ABC categorization and demand volatility with XYZ categorization. On the axis of the ABC categorization, the smallest set of DFUs that generate 80% of the total value (or other) gets an A, the next 15% gets a B, and the remaining gets a C. From another dimension, the X category products are relatively stable and therefore the easiest to forecast using statistical forecast models, and the Z category shows the highly volatile, often intermittent items.   

The ABC-XYZ categorization focuses on which DFUs to spend the most time on and which DFUs can ideally be left untouched by statistical forecasting. This is not meant to be the holy grail, but for anyone who wants to understand where statistical models in demand planning can help and where you need to organize differently, this is a good place to start. Identifying the part of your portfolio where forecasting may not be the solution can already trigger thinking about what to do next with supply, inventory, and service setups.

3. Outlier cleaning 

The goal of statistical forecasting is to find historical patterns to build future forecasts. Extreme outliers caused by one-time events (such as Corona, the Suez Canal, a shift in product due to supply issues, etc.) should not be included in the baseline statistical forecast. Therefore, an element of outlier and event cleaning should be present in any setup with statistical models in demand planning. Without cleaning on historical sales, all sorts of effects will remain in your data to be considered by the statistical models and affect the baseline forecast. Think of promotions, out-of-stocks, one-off marketing events, one-off market impacts, forward buying due to price changes; all these effects have happened in the past, but there is no guarantee that they will happen again in the future.    

If you want to discover what statistical forecast models can do for you, it is important to understand the outliers in your sales data. Are they related to buying behavior, events, or things outside of my control? Do we have data that we could use to facilitate some automated cleaning of outliers, such as out-of-stock information? Getting an understanding of outliers will help you determine how to organize yourself for cleaning in an efficient way that helps statistical forecasting to drive quality. Remember, as with any system that uses data, “crap in = crap out”.    

To conclude this topic, just a few thoughts on cleaning sales data. A first step is to remove relevant outliers that you can associate with known drivers. For example, promotions are usually known, so if you do not want the statistical forecast models to predict them, you should clean for them. In an ideal world, you would then clean out the exact effect of the promotion, but in the real world, this data is often not available. So, consider ground rules that use the event data to clean for specific products, customers, and time periods where there might have been an event effect.   

The second level of cleaning is related to unknown outliers. This is where automated outlier cleaning comes in. The intent is not to clean heavily, but only to take out the truly exceptional sales that are very different from the normal sales pattern. Usually this means cleaning within a bandwidth, where the way you define the bandwidth can be done in multiple ways. Just make sure that the approach you use is transparent.

4. Seasonality & trend detection    

The final piece of insight to gain when starting with statistical models in demand planning is the importance of seasonality and trend detection to your business. One of the core expectations of statistical forecasting is to capture seasonality patterns and identify growth trends as part of the baseline forecast. However, it is easy to see trends where there are none, or to mistake event-driven sales for seasonality.   

Let’s look at seasonality first. Seasonal patterns can be identified on the outlier-cleaned actuals. This is usually done at the DFU level and a higher level of aggregation, often product group (versus product) or region (versus country). Seasonality is often detected monthly but can also be relevant weekly. Usually, a company already has a good sense of when sales are higher or lower due to seasonality. However, testing this statistically can provide surprising insights. Seasonality may not be statistically evident at the aggregate level where it is expected, or seasonality may appear where it was not previously expected. Digging into the details can then reveal an underlying cause for this seasonality.   

In essence, seasonality must represent the sales pattern associated with the characteristics of the product. Ice cream sells more in the summer and hot soup sells more in the winter. In several product categories, event planning may cause sales that look like seasonality. For example, if deodorant is promoted in March, this does not mean that the sales would have been higher without the promotion. Probably the sales would still increase due to the warmer weather, but in a smoother way.

Second, there is the growth trend element. For each DFU, trend detection shows whether it is statistically significantly trending up or down. If so, it provides details such as increase/decrease and slope. For example, if a company has an ambitious volume growth target for the next few years, a thorough trend analysis will provide interesting insights. It can determine which part of the portfolio is already showing an increasing trend to validate the client’s gut feeling and identify future growth champions.     

Often the first instinct when looking at sales data is to see trends, but this is often wishful thinking or driven by the growth target that needs to be met. Doing the quantitative analysis can show where the trend is and how to best set up the trend adoption in your models.  

Ready to kickstart your automation journey with statistical models in demand planning?

A high-quality statistical forecasting approach will help identify past demand patterns and apply them to future forecasts more accurately and efficiently than manual methods, freeing up planners to enrich the forecast where necessary. But most importantly: it will kickstart your forecast automation journey. 

In our e-book, we guide you on your journey to getting the basics of statistical forecasting right. You will discover: 

• How to get started with statistical forecasting 
• Practical steps on how to get the most out of your statistical setup & effectively tackle challenges 
• How to turn your statistical forecast setup into AI-based forecasting 

In doing so, we will provide you with the steppingstone to optimal “smart-touch” forecasting: a balance of machine and human efforts to deliver the best possible forecast in this ever-changing environment. Get your copy of the e-book here.

The post Getting started with statistical models in demand planning appeared first on EyeOn.

In one of our previous blogs, we introduced the various types of data that are essential to enable smart-touch demand forecasting: facts, plans, and indicators. To shed more light on what these types of data can bring in a smart-touch forecasting set-up, we will dive deeper into this topic in this blog.  

Accurate data is essential within a smart-touch demand planning set-up (including machine learning). Data provides the foundation for building accurate models and thus increasing performance. The accuracy of the models used depends on the selection and quality of demand drivers. By using various types of data, such as facts, plans and indicators, these models learn to recognize patterns and make more accurate predictions. As the availability of structured data within large multinationals grows, the need arises to investigate what data archetypes to use for what purpose. 

Demand planning data archetypes 

Data available for demand planning can be classified into three categories: facts, plans and indicators. 

To make this a bit more tangible, we will touch on each data archetype in the next paragraphs. We will indicate typical examples of each type. To make it easier for you to recognize these types in your own organization, we will touch on some characteristics per data type. We will look at: 

• Data ownership: who owns the data, who unlocks the data or where can you find the data 
• Data availability: how frequently is the data available 
• Data granularity: what is the level of detail in the data type 
• Time horizon: for which horizon (short, mid or long) and in which direction (backward, forward) is the data available  

Facts 

The first data archetype essential for demand planning is factual data. This resembles data that is available within your company and can be taken off the shelf usually (e.g., from your ERP system, data received from your customer). Usually, this type of data is very structured, with a lot of detail and available in high frequency. 

Factual data can both be historical data, such as historical sales or growth numbers. It, however, can also be future data, such as orderbook and contracts. Let’s say you have an agreement with one of your customers to deliver 100 tons of a certain product; you can already consider this in your demand plan. In the overview below, you can find some examples of factual data. 

Many companies already use some factual data in their demand planning process. Historical sales are often fed into a statistical forecasting engine to predict future demand. We also see more and more that other types of factual data are also integrated into the demand planning process. For example, order book can easily be included as a driver in a machine learning model (as our client Cargill did). Overall, factual data is the data archetype that is most easily available and can be a great stepping stone towards smart-touch forecasting. 

Plans 

The second data archetype is related to plannable data. This is data of activities that you are foreseeing for the (near) future. This data is typically related to future events, portfolio changes or activities that are specially generated as part of plans to influence demand.  

Next to internal customer data, it could also be related to customer activities or changes that your customers plan to execute. In general, this data is updated regularly but less frequently than factual data. Updates come from review moments, new information that is processed or because of business planning processes such as S&OP / IBP. 

This type of data is often already included in the demand planning process for many companies. The most obvious example is related to portfolio changes. You can think about including ‘phase-in and phase products’ as the first step in your historical data cleaning process. If you perform this in the right way, you can take the history of the preceding product and turn that into your succeeding product. This will make your baseline model much more accurate.  

Overall, plannable data is the data archetype that requires good alignments internally between different parties such as marketing and sales. Think carefully about your business planning framework in which you combine making the plans with a data integration set-up that brings efficiency and transparency. It could also require collaboration with your customers on the activities that they are performing.  

Indicators 

The last data archetype is indicator data. This concerns all forms of data that give an indication of the direction demand is likely to go in. Often these can be % indicators or expectations that determine a position versus a specific situation. Indicator data can be extremely varied. Typically, this data is more aggregated and less frequently available. 

Looking at market data, you could think about oil prices for customers in the process industry, but also about the evolution of the flu for the life sciences industry. One of the most common examples is weather data to predict the number of ice creams sold. If you had this external data available, you would be better able to predict possible peaks and dips in your demand. However, the biggest issue with this type of data is the lagging effect. This means that external data is not always available when you need it. For example, weather data is only accurate as of 14 days prior. If your supply chain has a lead time longer than 14 days, this data is less interesting to include. 

Where do these data archetypes add value to demand planning? 

The data archetypes mentioned above have very different characteristics. Because of this they will add value at different time horizons within your planning horizon. Factual data is typically adding value on the short horizon in which you perform demand sensing. E.g. using forward looking order book data to determine what your likely sales will be in combination with your forecast data. This can help to sense your demand and tune the short-term demand signal, thus impacting service and optimizing inventory. To really get the value out of these types of data, a certain level of automation is required, thus allowing the focus to go taking decisions with high speed.   

On the other end of the spectrum, long-term indicator data is mostly focused on more strategic decisions or mid-term decisions. Growth outlooks in your customers’ industry can give indication of demand changes you will see further out. Using such indicator data to anticipate can drive key decisions on network design. Integrating indicators is best organized by means of a scenario planning approach, which is reviewed at planned moments as part of your IBP process.  

In the mid-term, we tend to see that plannable data has the biggest impact. It is mainly focused on the tactical horizon, which is aimed at the improvement of demand planning and forecasting. Applying information on sales and marketing plans makes your forecast more realistic and thus allows to steer supply and inventory management decisions. Regular review of the forecast with the latest information is the best approach here. Avoid continuously reviewing the forecast because of plan changes, as plans tend to change multiple times. But also avoid never reviewing, as initial plans are highly unlikely to stay stable.  

Ready to elevate supply chain forecasting and unlock the full potential of demand planning?

Despite the steps made in the road towards smart-touch demand planning, many organizations do not unlock the full potential of the data they have available. To unlock this data, you need to know what you can do with it.  

That’s why we’ve created the smart-touch forecasting roadmap. 

In this roadmap, we’ll dig deeper into the hype surrounding complete forecast automation, advocating the crucial role of human expertise when applying statistical and machine-learning models. And we’ll provide you with practical steps to accelerate progress in your supply chain forecasting maturity level. Get your copy.

If you would like to further discuss the data you have available and how to use it to make better predictions, please reach out to our forecasting specialists: Erik de Vos or Bregje van der Staak. 

The post The power of data in demand planning: facts, plans, indicators appeared first on EyeOn.

In today’s world, it is hard to find anyone who would deny the importance of demand planning and forecasting. Focus on high inventory volumes is no longer an effective solution. Costs should go down while service should go up. We need to handle more last-minute requests for more complex portfolios. And we need to do it all while managing the entire supply chain in a sustainable way. Without a solid demand planning & forecasting strategy, we would be stuck in a never-ending loop of firefighting.   

Despite the abundant evidence that demand planning & forecasting is beneficial to any organization, many planning teams are still stuck in the labor-intensive full-touch forecasting phase. What is stopping us from realizing the true value of demand planning and forecasting?  

In this blog, we will touch on the right vision of forecasting, which we call smart-touch forecasting. As any development should be linked to your context, we will offer some key questions to consider before you start developing your forecasting strategy, and how the answers can guide you towards a stronger approach to forecasting development. 

Let us focus on smart-touch planning 

As we explain in our e-book “Navigating the era of smart-touch forecasting”, smart-touch planning and forecasting is any planning set-up that balances human expertise with the available machine intelligence within the realm of demand planning. In this set-up we distinguish 3 key elements that make it smart-touch: 

• Automate the obvious 

• Recommend the probable 

• Flag where human intervention is needed. 

In smart-touch we assume that both the human and the machine add value to your forecasting strategy. Automating, recommending, and flagging are skills you can expect from the machine, while interpreting and deciding is where you would look to the human. It is important to emphasize that smart-touch planning and forecasting goes beyond mere forecasting algorithms. It means implementing planning processes that derive decisions from forecasts, considering all available and relevant data. It involves redesigning roles and responsibilities across the organization, with planners focusing on key forecasting decisions and business functions providing valuable input to the machine. And it means thinking carefully about what data can add value to your forecast and how to unlock that potential.   

Obviously, data and techniques are important pre-requisites in set-ups that are aim for smart-touch and a level of machine automation that is linked to it. In the more advanced forecasting strategies, machine learning will be used to make more accurate predictions. But smart-touch can go further than just forecasting. For example, you can think about the following:

Knowing where to start with your forecasting strategy 

Understanding smart-touch is one thing, but how do you get started? We will touch on 4 challenges you will face if you want to use machines in forecasting. As you read the 4 challenges, try to see if you can recognize them in your own organization. The first one you recognize is your starting point.  

Challenge 1: Motivation for more forecasting automation 

Let’s think about the last time you were motivated to change something fundamentally. What triggered the motivation to actually adopt something new?   

• Were you inspired by someone who shared something new?    
• Did you have a positive new experience?  
• Did a negative experience trigger you to change?  
• Were you challenged to adopt a change? 

Whatever your trigger was, the same motivation hurdle exists for demand planning & forecasting. Moving towards more automation and smart-touch in your forecasting strategy requires motivation to change and start trusting machines. Motivation to accept and understand what machines are good at and to act as a complementary planner where machines struggle.  

Challenge 2: Figuring out which data will allow machines to add value? 

As mentioned earlier, data will be central to any smart-touch predictive system. Machines need data to do their job. Obviously, sales data is a key source for any forecasting setup, but there is so much more. By breaking down your business into the key elements that drive your demand, you will see which data you can leverage. E.g. if I work in an ice-cream business, it is likely that season, weather, portfolio, … will affect my demand. Once you know which elements influence your demand, you can start thinking about how to organize the data around it. Different types of data will require different efforts to unlock and will add different value to your forecasting. Stay tuned for our next blog on data types for forecasting.   

Challenge 3: Having an integrated forecasting strategy to develop more advanced setups.  

Advanced forecasting strategies like smart-touch planning, will require thinking further than putting in place some tools. We cannot repeat it enough: a tool alone will not fix your forecast problems. A tool with more advanced forecasting capabilities and smart-touch setups will help. But, a tool with poorly organized data will be useless. So, start thinking about your data structures and data flows. Some of this will also require reflection on planning processes and roles, as some data only exists because people put in the right effort.  Connect the dots between process, roles, data, and tools, and develop a plan to put your smart-touch forecasting set-up in place. 

Challenge 4: Having a smart-touch forecasting personal development approach  

Finally, once you have developed towards a more smart-touch forecasting strategy, it is a matter of also connecting to a more personal development approach. Using smart-touch forecasting is a combination of processes, data, and tools, but also of people acting their smart-touch roles. Think about making smart-touch roles tangible enough so they can become part of personal development plans. Have I stopped my old ways of working? Am I using the exception-based alerts to my advantage? Am I focusing on value-added enrichment rather than overall forecasting?  

The challenges as your guide for forecasting strategy development

A solid approach to smart-touch forecasting would require you to cover your angles on the 4 challenges and start with the first challenge you encounter.  

Bring in inspiration, experience, and training whenever you see that people are not yet open to moving towards more automation in forecasting. Organize data discovery to get your organization aligned on what data has value in forecasting and excited about the potential it holds. Begin tool development only after you have taken an E2E view of understanding your forecasting processes, the roles involved in forecasting, and the data structures required. And top it off with an ongoing personal perspective that encourages continuous development and opens the door to more smart-touch thinking.   

At EyeOn, we encounter these challenges on a daily basis, and we have developed our approaches to cover the 4 angles. Connect with our expert team or already start your discover with our forecasting self-assessment.  

Get your copy of the e-book here.

The post Smart-touch forecasting: Practical questions to improve your forecasting strategy appeared first on EyeOn.

Why publish a blog on statistical forecasting for demand planning when artificial intelligence and machine learning seem to be popping up everywhere as the ultimate holy grail? Well, because to apply machine learning in a way you can benefit from it, companies need the experience of having worked with a best-in-class statistical forecasting approach as a solid steppingstone toward the next level of demand planning. And many companies haven’t reached that level yet.   

To begin our discovery of the value of statistical forecasting in 2024, let’s start with EyeOn’s concept of Smart-touch Planning & Forecasting. We believe that the best forecasting results are achieved by working in an environment that balances human and machine effort. At the core of smart-touch forecasting is the belief that a machine will always be better at recognizing patterns in data, while humans will always be needed to bring intelligence that cannot be found in the available data. This assumption then leads to 3 principles: use the machine to automate the obvious, use the machine to recommend the probable, so the human can decide, and use the machine to flag where intervention is needed.

Learn more about EyeOn Smart-Touch Planning & Forecasting.

From manual, full-touch forecasting to advanced setups with machine learning 

When we observe how demand forecasting is typically done, we tend to see 3 types of setups. The first one is the full-touch forecasting setup. In this setup, the demand forecast is the result of a lot of manual translation of demand information into what is hopefully a forecast. The usage of machine capability we tend to see in this approach is the use of Excel, which at least allows for quick calculations, if Excel can handle the size of the data. Of course, keeping the forecast up to date and aligned is a real challenge in this setup. On top of that, bias creeping into the forecast is inevitable, as there are too many factors to manage in too little time.    

At the other end of the spectrum, we are starting to see setups that use machine learning. Here, the forecast is the result of a machine learning engine that can use more than just sales history, with human intervention only when necessary. As the machine can see the connection between multiple demand driver inputs, such as promotions, portfolio changes, price changes, weather, and economic indicators, it will provide a more complete forecast with greater accuracy and almost no bias. The human then focuses on two things: filling in where the machine lacks data and deciding how to run the business based on the forecast. Getting to a working machine learning setup, however, requires a solid data approach and business understanding that builds confidence in what the machine will deliver. Moving from manual ways of working to trusting a machine can be quite challenging, often due to a lack of understanding of what drives demand. 

This lack of understanding on what drives the demand, and the consequential gaps in processes and data that will often be there, is where we see the setup with statistical forecast models that can be the bridge to more advanced forecasting. In this setup, the machine intervenes already to unlock pattern recognition based on historical data to provide a strong baseline forecast as a starting point for the planner. Because the machine intervention is focused on historical patterns, it remains understandable to the planner which means he/she can build confidence in using machines for forecasting. In the same time, by bringing more focus on value-adding enrichment, transparency in the process and data gaps is created, which forms the basis in moving to forecasting with machine learning.

Statistical forecasting as the gateway to advanced, machine learning based forecasting 

Let’s pause to consider the value that a statistical setup can bring to an organization moving toward more advanced forecasting. To do this, let us first break a demand forecast into two parts: 

1. What you know based on history: In any forecast, there will be a part that we can relate to the historical patterns we see. In its simplest form, this means identifying patterns like seasonality, trend and level in your own sales history and deriving future demand patterns from that. In a more advanced form, it could mean using covariates in your forecast set-up like sales data from your customers. 
2. What you know about the future: Since the past is no guarantee for the future, in most cases demand forecasts will need to be supplemented with elements that will be different in the future. Different plans, changing economic outlooks, expected market growth, and unforeseen events; all can and should lead to at least a validation of the forecast and in most cases an enrichment of the forecast.
   

Harnessing the power of statistical forecast models 

Now that we know the breakdown, what is the value of statistical forecast models in your forecasting setup? In summary, we would say:  

1. Statistical forecast models better capture historical sales patterns; therefore, they improve accuracy and reduce bias. Statistical forecast models will play a role in the part of the forecast you connect to your history. Historical sales patterns, such as baseline, trend, and seasonality, are something that any planner can probably identify by looking at the data but turning that into a good forecast is another matter. The most common statistical models make this easy, and the resulting forecast will be more accurate than any manual baseline. 
2. Statistical forecast models free up time to focus on value-adding enrichment of future elements by removing the obvious reasoning about historical patterns. The time that planners spend analyzing baseline sales patterns, defining a baseline forecast, and manually applying elements such as seasonality and trend can be eliminated with statistical forecast models. This creates the opportunity to move from pure number crunching based on the past, to reasoning with the business about what to add to the forecast based on the business information you have. So, statistical forecast models automate the obvious historical sales reasoning and allow planners to focus on what they can add based on information about the future.
3. Statistical forecast models make transparent what really drives your demand and prepares you for more automation. Think of statistical forecasting as a way to learn what the core data is that you will need in a more advanced machine learning setup. By focusing on enrichment, the entire organization gains experience in how to structure data, how to prioritize data, and how to connect data. Ultimately, this enrichment data experience will be critical for any setup beyond statistical forecasting with enrichment.    

By automating the routine, repetitive task of identifying historical sales patterns, statistical forecast models create room for demand planners to add their valuable insights and creativity to the mix. They can build stronger relationships with account managers, gain a deeper understanding of customer needs, and refine the forecast as needed. They can shift their focus from the full portfolio to the part of the portfolio that is important but where it is more difficult to forecast. The demand planner of the future should be a business thinker, collaborator, and communicator, not just a number cruncher. Statistical forecasting isn’t about replacing human judgment; it’s about enhancing it. With the right tools and technology, demand planners can make informed decisions based on solid, data-driven insights. Bringing statistical forecast models into the equation is the first step in unlocking this potential.

Who can benefit from statistical forecasting? 

Now that we understand the potential of statistical forecasting, is this potential available to everyone? Let’s talk about the prerequisites for getting started with statistical forecasting and the circumstances in which using statistical forecast models can be beneficial.  Theoretically, there are only two requirements to start using statistical forecasting:  

1. You need some form of time-series sales data. Everyone has sales data that contains valuable information about your demand – there may even be different versions of sales data – sell-in, sell-out, registration, or even customer sales data.   
2. You need some structured master data. Usually this involves product and customer master data.   

Ideally, statistical forecasting is set up with a proper forecasting engine and planning front-end. This should include a relevant set of forecasting models, and it should support the user in working with the best-fit model. But even without a proper statistical forecasting tool, if you have the data, you can already reap the benefits of statistical forecasting. There are enough offerings on the market that can bring a statistical forecast to your doorstep based on your data, such as EyeOn Forecast Service

In addition, we believe that statistical forecast models can be a solid starting point for many companies. When you see that a significant portion of your demand has consistent demand patterns, you should consider statistical forecasting. Even when you see that these patterns are more volatile but still repetitive, statistical forecast models can help you. On top of that, even when you see that the patterns can be quite disruptive without a clear link to the past, starting with statistical forecasting can add value by providing a starting point from which to build your case for enrichment. Ultimately, a machine learning setup will always outperform in this last situation, but it will require more data to work with. So, consider the intermediate step of statistical forecasting.    
 

Ready to kickstart your forecast automation journey through statistical forecasting?

A high-quality statistical forecasting approach will help identify past demand patterns and apply them to future forecasts more accurately and efficiently than manual methods, freeing up planners to enrich the forecast where necessary. But most importantly: it will kickstart your forecast automation journey. 

In our e-book, we guide you on your journey to getting the basics of statistical forecasting right. You will discover: 

• How to get started with statistical forecasting 
• Practical steps on how to get the most out of your statistical setup & effectively tackle challenges 
• How to turn your statistical forecast setup into AI-based forecasting 

In doing so, we will provide you with the steppingstone to optimal “smart-touch” forecasting: a balance of machine and human efforts to deliver the best possible forecast in this ever-changing environment. Get your copy of the e-book here.

The post The value of statistical forecasting in 2024 appeared first on EyeOn.

After reading our previous blogs on demand forecasting and planning techniques and important demand drivers for supply chain forecasting, we hope you’ve gotten a better grip on what smart-touch planning and forecasting entails. Now let’s explore how we can make it a reality. Achieving something starts with understanding your current position and taking the right steps to enhance your capability.

The road toward demand forecasting and planning with machine learning

At EyeOn, we believe that the best forecast results are achieved by working in a set-up that balances human and machine effort. At the core of smart-touch forecasting sits the assumption that a machine will always be better equipped to detect patterns from data, while humans will always be needed to bring intelligence that cannot be found in the available data. This assumption then leads to 3 principles: use the machine to automate the obvious, use the machine to recommend the probable, so the human can decide and use the machine to flag where intervention is needed. 

However, the step up to full usage of machine learning in demand forecasting is often a bridge too far. Many companies in practice still apply a ‘full-touch’ forecasting approach, meaning that the forecast is mainly the result of a lot of manual editing. So why not bring in statistical forecasting as a first step-change? Statistical forecasting is the technique that leverages historical demand/sales data to predict future demand. Wherever you have structured historical sales data, there is the opportunity to plug in statistics to do part of the forecasting work. By getting the basics right with statistical forecasting, it can be the steppingstone towards optimal ‘smart-touch’ forecasting: balanced machine and human efforts to provide the best-possible forecast in this ever-changing environment. 

While every company has its unique context, we’ll outline three typical forecast maturity stages on the road towards smart-touch forecasting that might resonate with your own. In describing these stages, we focus on how forecasting is organized as part of the demand planning set-up.   

Maturity stage 1: Demand planning with full-touch forecasting 

Firstly, there’s situation 1 – what we refer to as ‘full-touch forecasting’. In this scenario, planning processes, roles, data, and tools are still in the early stages of development. “Full-Touch” then indicates that there is little automation in how the forecast is created or in other words, without human effort there most likely would not even be a forecast. Here, the forecast often serves to secure stocks rather than steer business decisions. As a result, the forecast undergoes detailed editing almost daily. Excel is usually the forecast weapon of choice, not allowing to track why the forecast was edited and learn from the work that was done.  

Evolving out of this stage typically requires a process-people-data-tool approach with high attention to awareness and change management. Start by experiencing the value of more organized demand forecasting and planning for your organization and build an organization that is used to thinking in demand drivers.  

Maturity stage 2: Demand planning with statistical forecasting & enrichment 

Next up is situation 2, characterized by an organization embracing ‘statistical forecasting with enrichment’. To streamline forecasting efforts, the company has introduced statistical forecasting models to uncover hidden insights in sales history. As this does not deliver a good enough forecast for all products, there will be processes that bring the relevant enrichment information to the surface so it can be decided how to take this into the forecast. In general, there is more awareness and understanding on how to plan & forecast, and on the value of forecasting.   

However, it is also voiced that demand forecasting and planning can be labor-intensive in the early stages of this set-up, even with the objective of delivering one forecast version only. We would argue that you need to experience these challenges to become aware that a set-up of statistical forecasting with enrichment requires tuning as you go.  

In the best set-ups, you will already see several smart elements appearing and we could consider those set-ups as ‘smart-touch with statistical forecasting’. For example, applying portfolio segmentation brings focus and distinction to the demand forecasting and planning approach. Organized enrichment processes that connect automatically in the forecast, reduce manual efforts, creating room to focus on specific situations. Connecting forecast enrichment to formal assumption logging makes the forecast learning process smarter. Tracking value-add for all manual enriching of the forecast, brings in awareness and supports conscious enrichment. 

Maturity stage 3: Demand planning with smart-touch forecasting & machine learning 

Finally, we arrive at Situation 3 – ‘smart-touch planning & forecasting with machine learning’. Here, the organization is adept at working with data and recognizes the potential of qualitative and timely data to automate obvious tasks. Trust is placed in a machine learning engine to generate an unbiased and accurate forecast based on the information provided. Decision processes are aligned to leverage the forecast as a starting point for gap-closing decisions instead of merely managing targets.  

In this setup, the machine smartly covers the obvious aspects, while humans focus their smart efforts where the forecast shows potential for additional value. Roles and responsibilities are structured to emphasize how each element contributes value to the organization. 

Want to dive deeper into these maturity stages and the essential aspects of a smart-touch set-up? Our smart-touch forecasting ebook, provides you with a detailed explanation of each maturity stage, helps you decide your ideal future outcome and how to plot the journey towards it.

Define the roadmap that will bring your demand forecasting and planning lightyears ahead 

The above-mentioned maturity stages might give you a first idea of what stage your current forecasting set-up is in. However, there are many different factors that we should consider. 

Before you start your smart-touch journey, it’s important to identify your company’s maturity level in aspects essential for a successful smart-touch setup, namely: intrinsic motivation to work with data, belief in the feasibility of automation, an organized approach to data, and a clear direction for sustainably developing smart-touch planning and forecasting.  

Finding out beforehand the status of these elements and which ones require more attention in your journey will not only prevent unpleasant surprises later but also speed up your forecast transformation journey significantly.  

The forecast self-assessment will help you discover:  

• Your current position on the road to future-proof forecasting  
• The most suitable roadmap forward based on your current position  
• Practical steps to accelerate progress in your forecast maturity level  

Discover your best next step to speed up your forecast optimization journey – in just 10 minutes. Get started here.

The post The 3 maturity stages to advanced demand forecasting and planning appeared first on EyeOn.

In one of our previous blogs, we mentioned various types of data integral to supply chain forecasting: sales data, factual data, plan data, and market indicators. To shed more light on the smart-touch planning and forecasting approach, let’s dive deeper into some examples, introducing the concept of demand drivers. These are elements that drive change or explain shifts in your demand. Since they help to increase forecast accuracy and reduce bias, both play essential roles in supply chain forecasting.

1. Promotions 

Take promotions, for instance – they’re catalysts for changes in demand. If you organize promotions, it’s reasonable to expect a shift in demand for your products.  

Now, imagine knowing when you conducted past promotions and when you plan future ones. Armed with marker information during promotions, a machine could forecast potential demand changes and incorporate them into your forecast. By enriching promotion details – promo mechanisms, discount percentages, marketing investments, or indicators like second placements and folder advertisements – a machine can learn to link this information to sales data effects (peaks and dips) for a more accurate forecast. 

Consider the alternative – planning promotions without machine learning. It often involves copy-pasting from previous efforts, assuming a slight sales increase even with the same promotions, expecting a uniform uplift on all products, and assuming consistent phasing of promotional effects for all customers. Due to data complexity and the lower priority given to supply chain forecasting promotions compared to other tasks, these shortcuts become the only way to deliver a forecast.

2. Order book data

Now, what about elements providing insights into likely changes in demand? Enter the order book, a typical example readily available in most ERP systems. Depending on your business, your order book horizon may differ. What if we could leverage this data to automate processes like allocation and short-term forecast tuning? By collecting data on when orders are received and their requested delivery dates, you can predict your order book’s future shape. Offset against your latest forecast, organized with supply and inventory considerations, you can identify service risks and overstock situations earlier, even asking the machine to distribute the remaining inventory based on likely demand.

3. Market data

Moving to another example – the increasing focus on using market data. Your demand isn’t solely influenced by your actions but also by the broader market context. Incorporating information on how your market evolves and will evolve, can aid in driving part of your supply chain forecasting effort.  

Market growth indicators can help to tune the overall trend you take into your forecast. Information on past and future external events not under your control can be used to predict event-based disruptive demand e.g. during a World Cup. Information on how the renovation and housing market will evolve can help you to predict the longer-term demand for construction materials.

Granted, these examples focus more on trends and mid-term effects. Yet, in the dynamic landscape of continuous information and opinions, you’ll also encounter market effects impacting demand on much shorter notice. For instance, weather outlook information can predict short-term swings in demand, enabling you to steer logistical execution capacity effectively. Or think about social media product trends that can suddenly boost or destroy sales.  

In relation to market data, we would expect to recognize 2 elements in a well-structured smart-touch planning and forecasting setup: 

• Firstly, fast insights and a decision-making framework that uses available information to detect the disruption in the market quickly and manage it.  
• Secondly, the ability to translate market insights & disruptions into assumptions for your mid-term forecast, allowing for proper steering of the supply chain.

Effective supply chain forecasting with demand drivers 

So, bringing smart-touch to life essentially involves understanding your demand drivers – those influencing demand and those providing valuable information. Once you grasp your drivers, maintain clarity on what they contribute to your forecast. Using the order book finetunes your shorter-term horizon but doesn’t impact your longer-term forecast. Promotions bring one-off uplift effects but also introduce dips. Sales history captures regular and seasonal sales effects. Market indicators tune trends or prepare for disruptions. Expecting the right consequence of a demand driver ensures you set the right things in motion. And that is the essence of smart-touch planning and forecasting.

Ready to elevate supply chain forecasting and unlock the full potential of demand planning?

Despite some huge steps we have taken in demand planning, many planning teams remain stuck in the labor-intensive full-touch forecasting phase. It’s time demand planning teams unlock the true value of demand planning through smart-touch forecasting. 

That’s why we’ve created the smart-touch forecasting roadmap. 

In this roadmap, we’ll dig deeper into the hype surrounding complete forecast automation, advocating the crucial role of human expertise when applying statistical and machine-learning models. And we’ll provide you with practical steps to accelerate progress in your supply chain forecasting maturity level. Get your copy.

The post Essential demand drivers to consider in supply chain forecasting appeared first on EyeOn.