Implementing a pull-based <a target="_blank" rel="noopener noreferrer" href="https://www.vectorconsulting.in/en/research-and-publication/whitepapers/from-threads-to-trends-navigating-supply-chain-complexity-in-apparel-retail/">supply chain</a> solution involves a significant paradigm shift from the conventional push-based supply chain. When most companies, even the big names in many industries, use push-based systems, the common experiential knowledge of managers and even consultants comes from them. As a result, there is a continual threat that a pull-based system will be dismantled, as most managers and consultants rely on their past experience to ‘improve’ systems.  This article was written for stakeholders at the corporate office of a large fashion company (referred to as ‘The Fashion Company’), who, despite getting fantastic results from pilot roll-out of the pull-based system, in one of the divisions were being ‘tempted’ by consulting companies to go back to the push-based system of forecasting.  <h2>Understanding results of pilot implementation</h2>If we compare the actual offtake of MTA (Make-To-Availability)(1) SKUs (end-product entity) for pilot wholesalers with the booking(2) quantities, the results provide an interesting insight. The actual offtake varies significantly from the initial booking quantities provided. Some qualities and shades were sold twice the booking quantity, while for others it was as low as 50% of the booking quantity. This kind of wide variation is observed in almost all wholesalers who have adopted the MTA system, regardless of differences in managerial competency, years of experience in business or IT systems or the systems available to them.  Remember, we are dealing with perennial SKUs here. If the forecasting error is so high for qualities and shades sold year-round, can we even imagine the errors that would occur with new products?  <h2>The revelation</h2>The people who have the most profound knowledge of their local market (the wholesalers) are unable to forecast with any reasonable accuracy, even for products that sell regularly. (In the seasons before the roll-out of the MTA system, they have been suffering from high inventory and stockouts due to the gross errors in the forecast of perennials.)  <h2>Possible hypothesis</h2>We are faced with either of the below two hypotheses to explain the phenomenon (of gross errors of forecasting by the wholesalers): <ul><li>Not all wholesalers use good ‘scientific’ forecasting tools and hence suffer from using ‘gut’ based forecasting decisions.  </li><li>It is futile to expect any significant improvement in accuracy. Forecasting demand for a fashion-related SKU within a wholesaler’s region for a booking season is highly prone to error.   </li></ul>The consulting community and software vendors have led us to believe that poor forecasting is due to the lack of use of ‘scientific tools’.  However, some may assume that hypothesis no. 2 is correct.  <h2>We have a problem!</h2>How do we know which of the two is a valid hypothesis? If something is being promoted as a science, how do we invalidate that?  We will not engage in a philosophical debate about what constitutes ‘science’ and ‘non-science’. However, we should be cautious; the use of mathematical models does not necessarily mean we are being scientific! Math is not always science!  Unless we provide a theoretical explanation of why hypothesis 2 is valid, the perceived sophistication of the tools will overwhelm us into accepting them under the pressure to be ‘scientific’. (Rather, the fear of being labelled ‘unscientific’ if we oppose those tools!)  <h2>Is there a theoretical fallacy behind these tools?</h2>One of the latest branches of science, called Chaos Theory, states that it is almost impossible to predict (with any reasonable accuracy) the output of a complex system due to non-linear feedback loops in reality. The relationships between variables in this system are not predetermined; a small change can have a disproportionate impact, and vice versa. (Remember the butterfly effect? – “A butterfly flapping its wings in China can lead to a tornado in Tokyo!”)  The demand for fashion-related SKUs at selling points exhibits the characteristics of a complex system; therefore, any mathematical model that employs predefined linear or non-linear relationships between variables cannot achieve reasonable accuracy. Even if we put 20 variables in our model, the error can throw us off track. How do we know that market demand for fashion items is a complex system with non-linear feedback loops?  We can relate to sudden surges in demand. A delay in a competitor’s supplies can cause a sudden rise in demand for our material in one location, or a fashion icon wearing a specific style can lead to a surge in demand in another. The news of increased high sales can, in turn, fuel further demand.  Because of the non-linear feedback loops, the ‘noise’ of the system is very high. Therefore, any forecast model can behave erratically at times, leading to significant errors that render it impractical for decision-making. To nullify the effect of local variation, many people advocate an ‘aggregate forecast’ or a top-down forecast. Top-down forecasting (which tries to take advantage of aggregation) or bottoms-up forecasting (which takes the advantage of knowledge of many local variables), or even a combination of the two, does not yield better results when one has to forecast for a large number of SKUs for a long horizon. Even if the aggregate forecast looks better, the advantages are lost when the inventory is eventually pushed to the point of maximum errors – the retail point.  <h2>A new approach to the problem</h2>It is a waste of time to even try to model a non-linear market reality. It is not easy to model, but it needs to be done.  Therefore, instead of trying to improve the forecasting model, it is crucial to drastically reduce the reaction time so that we suffer minimally from forecasting errors.  For wholesalers under MTA, we have a system that responds 2-3 times a month to changes in market demand, compared to the earlier scenario of booking for the entire season, where a response occurs only once every six months. The central stocking in the CWH (Central Warehouse) and efforts to reduce production lead time were intended to improve the reaction time to the market.  Due to the supply chain’s faster reaction time, wholesalers have suffered minimal impact from forecasting errors. With low <a target="_blank" rel="noopener noreferrer" href="https://www.vectorconsulting.in/en/research-and-publication/articles/100-availability-with-less-than-half-the-inventory/">replenishment</a> lead times for wholesalers, the company is better equipped to respond to changes in each wholesaler’s market demand.  The rule of Dynamic Buffer Management (DBM)(3) makes any sophisticated forecasting model redundant. If an SKU does not sell quickly, the DBM rules will eliminate it within 3-4 replenishment cycles. Conversely, if another SKU is selling rapidly, the norms increase within 1 or 2 replenishment cycles. This ensures that inventory is always at the optimal level. The replenishment time is about 10 days for wholesalers under MTA, down from six months under the booking cycle. You can estimate the speed of reaction time under the DBM system.   <h2>What about the non-MTA products?</h2>We employ a practical approach to addressing the conundrum of high inventory and stockouts for MTA products. However, a significant volume of business comes from non-MTA, where we expect forecasting errors to be even more critical.  Do we have to rely on a crystal ball, or learn to live with the bigger problem? Since that doesn’t sound interesting, let’s see if there’s a way out.  Let’s switch topics for a moment and approach the problem from a different perspective – the retail selling point.  We have observed forecasting errors at wholesale and retail points when they use a ‘2-season booking’ method. For every season, there is always a high level of inventory for some slow-moving items. This means, over time, we expect the average inventory holding of a retailer to increase – it explains the inventory turns of 1 or 2 at the selling points. If one is the average inventory turnover, we expect some SKUs to be on the shelf for over a year, for sure. Low inventory turns at the selling point mean that a significant portion of the shelf space (and the seller’s attention) is occupied by slow-moving items. Even after discounting schemes are used to clear shelf space, the number of shelves available for new introductions is always limited. At the same time, wholesalers with low ROI also have limited capacity to take on new products.  Combine the above effects with the fact that the company offers a wide range of new products every year, and we have a rather interesting irony. Even when fashion companies introduce a wide variety every year, the actual shop display does not undergo frequent radical churn.  Let’s see if we can redefine the game with MTA in place.  The Fashion Company can have an extensive portfolio of SKUs under MTA, far more than a wholesaler or retailer can manage. With the wholesaler generating higher ROI  under MTA, he will likely shift his business to more MTAs. We are seeing this happen with many pilot wholesalers. He will experiment with new MTA SKUs because the quantities he has to hold for the new ones will be dramatically lower (maximum one month of sale) than under the six-month booking system. If the product sells well, it will be replenished quickly from the CWH. If products do not sell well, the rules of DBM will attempt to reduce inventory rapidly,  thereby reducing the risks to the distributor and improving their portfolio. (Remember, in the MTA scenario, the dealer, at any point in time, needs inventory only to provide for transportation time from the CWH to his location.)  For him and his retailers, he has added to his portfolio without waiting for collections. He can now manage his portfolio among the MTA SKUs, both with him and at the retail point. This means we will have new products at retail points without actually having developed them. More and more wholesalers will take advantage of The Fashion Company’s extensive offering of MTAs to add freshness to their portfolios while enjoying very high ROI. We expect the volume of business from MTA to increase significantly, while the stores will also offer more freshness than before.  The Fashion Company will experience a new paradigm – we no longer need to always develop new products to introduce them at the retail counters. <h2>So what about the fashion icon status of The Fashion Company?</h2>With more business coming from MTA SKUs, wholesalers may choose fewer of the new products. Now, on the face of it, it looks like The Fashion Company might lose its ‘fashion’ icon status. (Even though in the MTA system, we are introducing new products much faster without developing them every time.)  However, let us examine whether there is a significant opportunity rather than a problem. If we examine the six-month booking system, in which agents are asked to select from a plethora of new SKUs on a specific day, we will find that many SKUs receive insufficient attention.  It is not only humanly impossible to provide adequate attention to each SKU in the extensive offering, but when wholesalers have low inventory turns (~2), they have limited capital to invest in many new products. Therefore, the fear of locked capital becomes the dominant factor in decision-making. Combine this with the fact that wholesalers do make huge forecasting errors. So, though we might be developing products which are inherently winners, the wholesalers and agents may not pick them up, and at the same time, they could be selecting inherent losers, assuming otherwise.  Remember, the key to a fashion supply chain is a fast reaction, not accurate forecasts. Instead of asking the dealer for forecast requirements for a six-month season, we can ask for shorter horizons to reduce their forecasting errors.  But we have a problem with more seasons. In the current environment, when wholesalers start buying for the season at the start, it forces The Fashion Company to rely on forecasts for the first 1-2 months – so every season starts with a ‘blind’ zone. More seasons mean more blind zones. Each period of blind zone exposes The Fashion Company to the mess of forecasting errors – high inventory and stockouts. It does not seem like a great idea.  But let us try to understand why seasons lead to periods of a blind zone. It stems from a lack of supply reliability, and therefore wholesalers wait until supply is closed before beginning the following season. Thus, if we enhance the reliability of supplies, we can request that wholesalers place orders without waiting for the start of a season, but they would likely prefer a shorter season.  <h2>And now we have a win-win!</h2>We can increase seasons, which will reduce the dealers’ forecast period. At the same time, dealers need not wait until supplies run out to start preparing for the next season – there is no blind zone for The Fashion Company. However, there is one more negative. More seasons might decrease the quantity per SKU and cause more problems in the plant.  When one offers an extensive collection in a single large lot, we have an issue with SKUs not receiving enough attention. With many distributors managing an extensive portfolio, the likelihood of obtaining small quantities per SKU is high. But when we offer a small collection each time, the focus is on a few SKUs. We move to many seasons, but each season has a smaller portfolio. With focused attention on a smaller range, we should see higher quantities per SKU. Yes, we need to find a way to quickly reach out to wholesalers with each collection story.  At this stage of implementation, wholesalers should already be enjoying a high ROI, which will dramatically enhance their ability to invest in new products.  Now, we have a solution where all new products follow a pure MTO (Make-To-Order) route, eliminating the need for forecasting. When a product starts selling more, we can put it in MTA. The DBM rules will kill the MTAs that are not selling well.  We will have a system that does not require a sophisticated forecast model either for MTA or for MTO. As MTA business grows, the quantity per SKU will increase. The plants will have higher quantities per SKU, which will improve output in spinning and weaving.  What does all this do to the financial performance of The Fashion Company? Significant! Yes, the changes are fundamental paradigm shifts in approach to the market, and they are not easy. But then, who wants an easy life?

Key Takeaways

Forecasting in the fashion industry is highly unpredictable. Yet the industry follows the conventional push-based supply chain approach, leading to a lot of wastage and losses. A more agile approach that helps the industry stay in tune with real demand and minimise forecasting errors is needed.

Managing a fashion supply chain

Implementing a pull-based <a target="_blank" rel="noopener noreferrer" href="https://www.vectorconsulting.in/en/research-and-publication/whitepapers/from-threads-to-trends-navigating-supply-chain-complexity-in-apparel-retail/">supply chain</a> solution involves a significant paradigm shift from the conventional push-based supply chain. When most companies, even the big names in many industries, use push-based systems, the common experiential knowledge of managers and even consultants comes from them. As a result, there is a continual threat that a pull-based system will be dismantled, as most managers and consultants rely on their past experience to ‘improve’ systems.  This article was written for stakeholders at the corporate office of a large fashion company (referred to as ‘The Fashion Company’), who, despite getting fantastic results from pilot roll-out of the pull-based system, in one of the divisions were being ‘tempted’ by consulting companies to go back to the push-based system of forecasting.  <h2>Understanding results of pilot implementation</h2>If we compare the actual offtake of MTA (Make-To-Availability)(1) SKUs (end-product entity) for pilot wholesalers with the booking(2) quantities, the results provide an interesting insight. The actual offtake varies significantly from the initial booking quantities provided. Some qualities and shades were sold twice the booking quantity, while for others it was as low as 50% of the booking quantity. This kind of wide variation is observed in almost all wholesalers who have adopted the MTA system, regardless of differences in managerial competency, years of experience in business or IT systems or the systems available to them.  Remember, we are dealing with perennial SKUs here. If the forecasting error is so high for qualities and shades sold year-round, can we even imagine the errors that would occur with new products?  <h2>The revelation</h2>The people who have the most profound knowledge of their local market (the wholesalers) are unable to forecast with any reasonable accuracy, even for products that sell regularly. (In the seasons before the roll-out of the MTA system, they have been suffering from high inventory and stockouts due to the gross errors in the forecast of perennials.)  <h2>Possible hypothesis</h2>We are faced with either of the below two hypotheses to explain the phenomenon (of gross errors of forecasting by the wholesalers): <ul><li>Not all wholesalers use good ‘scientific’ forecasting tools and hence suffer from using ‘gut’ based forecasting decisions.   </li><li>It is futile to expect any significant improvement in accuracy. Forecasting demand for a fashion-related SKU within a wholesaler’s region for a booking season is highly prone to error.   </li></ul>The consulting community and software vendors have led us to believe that poor forecasting is due to the lack of use of ‘scientific tools’.  However, some may assume that hypothesis no. 2 is correct.  <h2>We have a problem!</h2>How do we know which of the two is a valid hypothesis? If something is being promoted as a science, how do we invalidate that?  We will not engage in a philosophical debate about what constitutes ‘science’ and ‘non-science’. However, we should be cautious; the use of mathematical models does not necessarily mean we are being scientific! Math is not always science!  Unless we provide a theoretical explanation of why hypothesis 2 is valid, the perceived sophistication of the tools will overwhelm us into accepting them under the pressure to be ‘scientific’. (Rather, the fear of being labelled ‘unscientific’ if we oppose those tools!)  <h2>Is there a theoretical fallacy behind these tools?</h2>One of the latest branches of science, called Chaos Theory, states that it is almost impossible to predict (with any reasonable accuracy) the output of a complex system due to non-linear feedback loops in reality. The relationships between variables in this system are not predetermined; a small change can have a disproportionate impact, and vice versa. (Remember the butterfly effect? – “A butterfly flapping its wings in China can lead to a tornado in Tokyo!”)  The demand for fashion-related SKUs at selling points exhibits the characteristics of a complex system; therefore, any mathematical model that employs predefined linear or non-linear relationships between variables cannot achieve reasonable accuracy. Even if we put 20 variables in our model, the error can throw us off track. How do we know that market demand for fashion items is a complex system with non-linear feedback loops?  We can relate to sudden surges in demand. A delay in a competitor’s supplies can cause a sudden rise in demand for our material in one location, or a fashion icon wearing a specific style can lead to a surge in demand in another. The news of increased high sales can, in turn, fuel further demand.  Because of the non-linear feedback loops, the ‘noise’ of the system is very high. Therefore, any forecast model can behave erratically at times, leading to significant errors that render it impractical for decision-making. To nullify the effect of local variation, many people advocate an ‘aggregate forecast’ or a top-down forecast. Top-down forecasting (which tries to take advantage of aggregation) or bottoms-up forecasting (which takes the advantage of knowledge of many local variables), or even a combination of the two, does not yield better results when one has to forecast for a large number of SKUs for a long horizon. Even if the aggregate forecast looks better, the advantages are lost when the inventory is eventually pushed to the point of maximum errors – the retail point.  <h2>A new approach to the problem</h2>It is a waste of time to even try to model a non-linear market reality. It is not easy to model, but it needs to be done.  Therefore, instead of trying to improve the forecasting model, it is crucial to drastically reduce the reaction time so that we suffer minimally from forecasting errors.  For wholesalers under MTA, we have a system that responds 2-3 times a month to changes in market demand, compared to the earlier scenario of booking for the entire season, where a response occurs only once every six months. The central stocking in the CWH (Central Warehouse) and efforts to reduce production lead time were intended to improve the reaction time to the market.  Due to the supply chain’s faster reaction time, wholesalers have suffered minimal impact from forecasting errors. With low <a target="_blank" rel="noopener noreferrer" href="https://www.vectorconsulting.in/en/research-and-publication/articles/100-availability-with-less-than-half-the-inventory/">replenishment</a> lead times for wholesalers, the company is better equipped to respond to changes in each wholesaler’s market demand.  The rule of Dynamic Buffer Management (DBM)(3) makes any sophisticated forecasting model redundant. If an SKU does not sell quickly, the DBM rules will eliminate it within 3-4 replenishment cycles. Conversely, if another SKU is selling rapidly, the norms increase within 1 or 2 replenishment cycles. This ensures that inventory is always at the optimal level. The replenishment time is about 10 days for wholesalers under MTA, down from six months under the booking cycle. You can estimate the speed of reaction time under the DBM system.   <h2>What about the non-MTA products?</h2>We employ a practical approach to addressing the conundrum of high inventory and stockouts for MTA products. However, a significant volume of business comes from non-MTA, where we expect forecasting errors to be even more critical.  Do we have to rely on a crystal ball, or learn to live with the bigger problem? Since that doesn’t sound interesting, let’s see if there’s a way out.  Let’s switch topics for a moment and approach the problem from a different perspective – the retail selling point.  We have observed forecasting errors at wholesale and retail points when they use a ‘2-season booking’ method. For every season, there is always a high level of inventory for some slow-moving items. This means, over time, we expect the average inventory holding of a retailer to increase – it explains the inventory turns of 1 or 2 at the selling points. If one is the average inventory turnover, we expect some SKUs to be on the shelf for over a year, for sure. Low inventory turns at the selling point mean that a significant portion of the shelf space (and the seller’s attention) is occupied by slow-moving items. Even after discounting schemes are used to clear shelf space, the number of shelves available for new introductions is always limited. At the same time, wholesalers with low ROI also have limited capacity to take on new products.  Combine the above effects with the fact that the company offers a wide range of new products every year, and we have a rather interesting irony. Even when fashion companies introduce a wide variety every year, the actual shop display does not undergo frequent radical churn.  Let’s see if we can redefine the game with MTA in place.  The Fashion Company can have an extensive portfolio of SKUs under MTA, far more than a wholesaler or retailer can manage. With the wholesaler generating higher ROI  under MTA, he will likely shift his business to more MTAs. We are seeing this happen with many pilot wholesalers. He will experiment with new MTA SKUs because the quantities he has to hold for the new ones will be dramatically lower (maximum one month of sale) than under the six-month booking system. If the product sells well, it will be replenished quickly from the CWH. If products do not sell well, the rules of DBM will attempt to reduce inventory rapidly,  thereby reducing the risks to the distributor and improving their portfolio. (Remember, in the MTA scenario, the dealer, at any point in time, needs inventory only to provide for transportation time from the CWH to his location.)  For him and his retailers, he has added to his portfolio without waiting for collections. He can now manage his portfolio among the MTA SKUs, both with him and at the retail point. This means we will have new products at retail points without actually having developed them. More and more wholesalers will take advantage of The Fashion Company’s extensive offering of MTAs to add freshness to their portfolios while enjoying very high ROI. We expect the volume of business from MTA to increase significantly, while the stores will also offer more freshness than before.  The Fashion Company will experience a new paradigm – we no longer need to always develop new products to introduce them at the retail counters. <h2>So what about the fashion icon status of The Fashion Company?</h2>With more business coming from MTA SKUs, wholesalers may choose fewer of the new products. Now, on the face of it, it looks like The Fashion Company might lose its ‘fashion’ icon status. (Even though in the MTA system, we are introducing new products much faster without developing them every time.)  However, let us examine whether there is a significant opportunity rather than a problem. If we examine the six-month booking system, in which agents are asked to select from a plethora of new SKUs on a specific day, we will find that many SKUs receive insufficient attention.  It is not only humanly impossible to provide adequate attention to each SKU in the extensive offering, but when wholesalers have low inventory turns (~2), they have limited capital to invest in many new products. Therefore, the fear of locked capital becomes the dominant factor in decision-making. Combine this with the fact that wholesalers do make huge forecasting errors. So, though we might be developing products which are inherently winners, the wholesalers and agents may not pick them up, and at the same time, they could be selecting inherent losers, assuming otherwise.  Remember, the key to a fashion supply chain is a fast reaction, not accurate forecasts. Instead of asking the dealer for forecast requirements for a six-month season, we can ask for shorter horizons to reduce their forecasting errors.  But we have a problem with more seasons. In the current environment, when wholesalers start buying for the season at the start, it forces The Fashion Company to rely on forecasts for the first 1-2 months – so every season starts with a ‘blind’ zone. More seasons mean more blind zones. Each period of blind zone exposes The Fashion Company to the mess of forecasting errors – high inventory and stockouts. It does not seem like a great idea.  But let us try to understand why seasons lead to periods of a blind zone. It stems from a lack of supply reliability, and therefore wholesalers wait until supply is closed before beginning the following season. Thus, if we enhance the reliability of supplies, we can request that wholesalers place orders without waiting for the start of a season, but they would likely prefer a shorter season.  <h2>And now we have a win-win!</h2>We can increase seasons, which will reduce the dealers’ forecast period. At the same time, dealers need not wait until supplies run out to start preparing for the next season – there is no blind zone for The Fashion Company. However, there is one more negative. More seasons might decrease the quantity per SKU and cause more problems in the plant.  When one offers an extensive collection in a single large lot, we have an issue with SKUs not receiving enough attention. With many distributors managing an extensive portfolio, the likelihood of obtaining small quantities per SKU is high. But when we offer a small collection each time, the focus is on a few SKUs. We move to many seasons, but each season has a smaller portfolio. With focused attention on a smaller range, we should see higher quantities per SKU. Yes, we need to find a way to quickly reach out to wholesalers with each collection story.  At this stage of implementation, wholesalers should already be enjoying a high ROI, which will dramatically enhance their ability to invest in new products.  Now, we have a solution where all new products follow a pure MTO (Make-To-Order) route, eliminating the need for forecasting. When a product starts selling more, we can put it in MTA. The DBM rules will kill the MTAs that are not selling well.  We will have a system that does not require a sophisticated forecast model either for MTA or for MTO. As MTA business grows, the quantity per SKU will increase. The plants will have higher quantities per SKU, which will improve output in spinning and weaving.  What does all this do to the financial performance of The Fashion Company? Significant! Yes, the changes are fundamental paradigm shifts in approach to the market, and they are not easy. But then, who wants an easy life?

Download PDF

Discover how shifting to a pull-based supply chain cuts replenishment cycles, eliminates forecast errors, and boosts fashion retail sales.