Estimating short life cycle sales curves

Abstract

Embodiments generate a short life cycle sales curve for a short life cycle item. Embodiments generate a plurality of similar sales curves corresponding to at least one similar item that is similar to the short life cycle item. Embodiments parameterize each of the similar sales curves, the parameterization including estimating an innovation coefficient parameter, an imitation coefficient parameter, and an error parameter for each similar sales curve, and embodiments determine a weight for each error parameter. Embodiments combine the innovation coefficient parameter and the imitation coefficient parameter using the weight, and generate the short life cycle sales curve using the combined innovation coefficient parameter and the combined imitation coefficient parameter.

Description

[Technical Field]

[0001] Field One embodiment is directed generally to computer systems, and more particularly to a computer system that forecasts demand for retail items. [Background technology]

[0002] Background information Sales forecasting systems typically face the problem of generating a weekly forecast of sales units for a retail item, referred to as a "sales curve" or "demand curve." Sales of a retail item in a given week are affected by many factors, including seasonality, whether the retail item was discounted during that week, and where that week falls in the product's life cycle. One common approach to forecasting weekly sales units involves building a "causal demand model" for the retail item. This demand model is a mathematical model that explains weekly sales units in terms of factors such as those listed above. These factors are known as "demand variables" or "demand features" that form the demand model.

[0003] A demand model mathematically specifies how a demand variable affects units sold. For example, if the amount of discount is a demand variable, historical data may show that a 50% discount resulted in a four-fold increase in units sold (i.e., associated with price elasticity). In this example, the demand variable is the 50% discount, and the historical sales data is a four-fold increase in sales. For a causal demand model to be useful in predicting units sold, the relationship of the demand variable (the 50% discount) to units sold (the four-fold increase) must be determined. This relationship is referred to as the "demand parameter" associated with the demand variable.

[0004] In this example, the demand parameters may be determined to stipulate that sales of a particular retail item will increase by 2x for every 25% markdown. Once the demand parameters are determined, sales units can be predicted by specifying future values ​​for the demand variables. Continuing with the discount example, a retailer may know that they will be offering a 40% markdown for several weeks next season. The demand model would then predict sales units for those weeks taking that 40% markdown into account.

[0005] The demand parameters are determined by examining historical retail sales data (known as "retail panel data") that includes discounts on the retail item itself or similar retail items. However, as noted above, several demand variables influence the sales of a retail item. Several of these demand variables may be applied simultaneously. For example, a retailer may offer a 50% discount on summer items during the summer, in which case a four-fold increase in sales would be due in part to the seasonal increase in demand for summer retail items during the summer. To isolate the effects of several demand variables on sales, regressions are run on the demand model to determine the values ​​of the demand parameters that best fit the demand model to the retail panel data.

[0006] Furthermore, some goods, such as fashion products, have very short life cycles, and the available previous sales data is usually very sparse. Due to the short sales history and sparse sales cycles, it is very difficult to estimate the sales curves of such short life cycle products. Summary of the Invention [Means for solving the problem]

[0007] overview Embodiments generate a short life cycle sales curve for a short life cycle item. Embodiments generate a plurality of similar sales curves corresponding to at least one similar item that is similar to the short life cycle item. Embodiments parameterize each of the similar sales curves, the parameterization including estimating an innovation coefficient parameter, an imitation coefficient parameter, and an error parameter for each similar sales curve, and embodiments determine a weight for each error parameter. Embodiments combine the innovation coefficient parameter and the imitation coefficient parameter using the weight, and generate the short life cycle sales curve using the combined innovation coefficient parameter and the combined imitation coefficient parameter. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 illustrates a computer system having a computing device configured with a short life cycle product demand forecasting tool, according to an embodiment. [Figure 2] FIG. 1 is a block diagram of a computer server / system according to one embodiment of the present invention. [Figure 3] FIG. 2 is a flow diagram of the functionality of the short life cycle product demand tool of FIG. 1 in estimating promotional effects that can be used in demand forecasting, according to one embodiment. [Figure 4A] 4A to 4C illustrate examples of implementations of embodiments of the present invention and how each example corresponds to the functionality of FIG. 3. [Figure 4B] 4A to 4C illustrate examples of implementations of embodiments of the present invention and how each example corresponds to the functionality of FIG. 3. [Figure 5] 4A to 4C illustrate examples of implementations of embodiments of the present invention and how each example corresponds to the functionality of FIG. 3. [Figure 6]FIG. 2 is a flow diagram of the functionality of the short life cycle product demand forecasting tool of FIG. 1 in generating sales / demand curves for short life cycle items (i.e., a given item / location or SKU), according to one embodiment. [Figure 7] FIG. 1 is a diagram illustrating an example of an implementation of an embodiment of the present invention. [Figure 8] FIG. 1 is a diagram illustrating an example of an implementation of an embodiment of the present invention. [Figure 9] FIG. 1 illustrates an integrated manufacturing, inventory, and logistics system including demand forecasting as disclosed herein, according to one embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0009] Detailed Description Embodiments generate demand forecasts for short life cycle products / items, such as fashion or technology items. Embodiments use a demand forecasting system / method to generate sales curves at various intersection levels for products similar to the short life cycle product. Embodiments then combine the sales curves at the various intersection levels using a parameterization of the sales curves and error and error estimation. The aggregate sales curve is then used as the demand forecast for the short life cycle product.

[0010] Short life cycle products are characterized by demand that occurs for only a short period of time, after which they become obsolete and, in some cases, have very short demand time-series sales curves. Examples of short life cycle products include technology products (e.g., computers, consumer electronics, video games) and fashion products (e.g., toys, clothing, textbooks). The demand period for short life cycle products can vary from years to weeks.

[0011] The dynamics of new product demand are generally characterized by a relatively slow growth during the introduction phase, followed by a rapid growth phase, after which demand stabilizes and the product enters a maturity phase. Eventually, demand declines and the product is replaced by another product that is usually perceived as quite similar to the preceding product.

[0012] Short life cycle products are becoming increasingly common in many industries, especially in the technology sector. Technology companies are driven by the constant introduction of new products as a result of highly competitive markets. A company's competitive advantage is determined primarily by its ability to manage the rapid turnover of products.

[0013] In the fashion industry, fashion products are characterized by short product life cycles and high uncertainty of market success. Unsuccessful products require multiple price discounts to clear inventory. A typical life cycle for a fashion product might be around 10 weeks.

[0014] As mentioned above, in the retail industry, retailers need to forecast future demand to better manage inventory or promotion / markdown plans. To accurately forecast demand, retailers take into account all factors that may affect demand, such as promotions, price fluctuations, seasonality, weather, etc. Known solutions for retailers have used various algorithms to estimate promotion or price effects, but these algorithms are typically not suitable for short life cycle products where historical data about the availability of a particular item may be sparse.

[0015] Specifically, for products with short life cycles, there is insufficient data early in the process to determine how the product's future sales will change over time. Traditionally, retailers have attempted to find the closest existing item in historical data as a proxy for predicting in-season sales. However, such an approach is prone to error for two reasons: first, customer reactions to newly introduced items are unknown, and second, newly introduced items, by interacting with the rest of the assortment, can fundamentally change the remaining life cycle sales of all items.

[0016] The retail industry, including clothing and fashion retailers, as well as high-end consumer electronics retailers in particular, is currently seeing a proliferation of short life cycle products. New product designs are introduced to the market even as older versions or previous short life cycle products are cleared from inventory or phased out. The time span of the life cycle itself is becoming shorter. For example, the life cycle of a fast fashion product can be as little as 12-13 weeks, leaving little time for retailers to adapt to changing customer preferences over time and in different locations.

[0017] More accurate short life cycle product demand forecasting allows retailers to better allocate and manage short life cycle items in their assortment. However, predicting sales for newly introduced products is challenging due to zero sales history and different store locations releasing on different dates. Short life cycle products typically exhibit a characteristic sales curve in which sales initially increase slowly from the point of market introduction, then increase and reach a peak value, and then gradually decline until all inventory is sold out or the product is withdrawn from the market.

[0018] The terms "item" or "retail item" as used herein refer to merchandise sold, purchased, and / or returned in a sales environment. The terms "particular item" and "single item" are used interchangeably herein and refer to a particular item type (e.g., a particular type of mobile phone, such as an iPhone® 8) rather than a unit item.

[0019] The terms "time," "period," "retail period," or "calendar period" herein refer to unit time increments (e.g., seven days per week) used by sellers to correlate seasonal periods from one year in a calendar to the next for planning and forecasting purposes. These terms may be used interchangeably herein.

[0020] The terms "sales channel" or "location" or "retail location" herein may refer to a physical store where an item is sold or an online store where an item is sold.

[0021] The term "sales data" herein refers to historical sales and promotion information recorded for items sold during past retail periods (e.g., over the past 52 weeks of the past year). The sales data may include, for example, the number of units (or total monetary value) of an item sold during each retail period, along with data characterizing one or more types of promotions for the item. The sales data may be stored, for example, in a database.

[0022] The terms "sales promotion" and "merchandise promotion" are used interchangeably herein and refer to a particular type of promotion for an item. Some examples of promotional elements include price discount promotion elements, television advertising elements, radio advertising elements, newspaper advertising elements, internet advertising elements, email advertising elements, and in-store advertising elements.

[0023] The term "promotion effectiveness" refers to a numerical value that characterizes the effectiveness of promoting an item (e.g., its effect on sales and profitability). For example, an estimated promotion effectiveness of 2.0 may indicate that a promotion or a combination of promotions is estimated to result in double sales (a 100% increase) for an item. Promotion effectiveness (i.e., value) may be used in demand forecasting models to predict demand for an item. Promotion effectiveness may also be used in computerized inventory systems to control various aspects of an item's inventory.

[0024] Embodiments generally utilize the following demand model or function ("Equation (1)") for demand forecasting:

[0025] Demand = Base demand × Seasonality × Sales promotion effect (× Additional feature effect) (1) where "base demand" is the historical demand without taking into account any influences or other factors, seasonality is the effect on demand based on the season (i.e., time of year), and promotion effect is the effect on demand based on one or more promotions offered during a period. Any or all additional features / variables that affect demand can be added to the model as appropriate. However, in some situations, the number of features may exceed 100.

[0026] Many demand models take into account additional influences, such as weather. For example, if the weather this year is significantly different from last year and two years ago, the forecast may need to be revised. For example, if there is a long, hot summer this year, the forecast for steak and ice cream would need to be increased. Another additional influence may be inventory. If some sizes and / or colors of popular fashions are out of stock, the forecast would need to be adjusted downward to account for the out-of-stock items. Yet another influence may be store count. If a retailer plans to expand aggressively and increase the number of stores by 10% next year, the forecast would need to be adjusted accordingly. However, for purposes of embodiments of the present invention, it is assumed that seasonality and promotional effects have the overwhelmingly larger impact on sales forecasts.

[0027] 1 illustrates a computer system 100 having a computing device 105 configured with a short life cycle product demand forecasting tool 110, according to an embodiment. In one embodiment, the short life cycle product demand forecasting tool 110 may be part of a larger computer application (e.g., a computerized inventory management and demand forecasting application) configured to forecast and manage sales, promotions, and inventory for retail items at various retail locations. The short life cycle product demand forecasting tool 110 is configured to computerize the process of forecasting demand for short life cycle products, especially when there is no or limited sales history for the products due to the nature of their short life cycles.

[0028] In one embodiment, system 100 is a computing / data processing system that includes an application or collection of distributed applications for an enterprise organization. These applications and computing system 100 may be configured to operate with or implemented as a cloud-based networking system, a Software-As-A-Service ("SaaS") architecture, or other type of computing solution.

[0029] Demand forecasts in the form of sales or demand curves are a key driver of the supply chain. Inaccurate forecasts lead to poor allocation and replenishment, resulting in financial losses for retailers. The embodiments disclosed herein can provide improved forecast accuracy for promoted items. Furthermore, a better understanding of the impact promotions have on demand can be achieved. This helps retailers plan promotions more effectively, for example, with respect to channels, pricing, and customer demographics.

[0030] Specifically, to plan future actions, many retailers rely on merchandise sales and lifecycle plans for various item categories. These plans typically consist of weekly values ​​that reflect sales patterns during the product's future lifecycle. Because these plans have values ​​spaced at uniform time intervals / periods (e.g., weekly), they are often viewed as time series in the form of a sales curve. For example, this time series may evaluate the expected sales of an item for the next 26 weeks. This time series is typically in the form of a sales versus period graph / curve.

[0031] The length of these plans can vary in time, which makes it difficult to adapt the plan for each product. Often, the shape of the plan / curve is common to multiple item groups, but the products may have different life cycles or sales seasons. It is common for retail planners to create one sales plan and then use tools to scale it up or down to fit seasons of different lengths. For example, an established sales curve may span 13 weeks, but a retailer may want to use this same curve for a high-fashion item that sells for only six weeks. However, known tools tend to distort the rise and fall of the curve.

[0032] Furthermore, the shape of the sales curve towards the end of the life cycle directly impacts the need for price reductions. The shape of the curve drives the forecast for an item. If the value is higher than it should be, the forecast for the item will be high. As a result, retailers risk having unsold items by not aggressively reducing prices. If the shape is lower than expected, the forecast will be low and retailers will try to increase demand by reducing prices. The result is lower profit margins due to unnecessary price reductions.

[0033] In one embodiment, short life cycle product demand forecasting tool 110 is implemented on computing device 105 and includes logic or modules for implementing various functional aspects of short life cycle product demand forecasting tool 110. In one embodiment, short life cycle product demand forecasting tool 110 includes visual user interface logic / module 120, similar item sales curve generation logic / module 130, sales curve parameterization logic / module 140, and demand forecast generation logic / module 150.

[0034] Other embodiments may provide different logic or combinations of logic that provide the same or similar functionality as the seasonal forecasting model tool 110 of FIG. 1. In one embodiment, the short life cycle product demand forecasting tool 110 is an executable application that includes algorithms and / or program modules configured to perform the functions of the logic. The application is stored on a non-transitory computer storage medium. In one embodiment, the logic of the short life cycle product demand forecasting tool 110 is implemented as modules of instructions stored on a computer-readable medium.

[0035] Computer system 100 also includes a display screen 24 operatively connected to computing device 105. According to one embodiment, display screen 24 is implemented to display views of and facilitate user interaction with a graphical user interface ("GUI") generated by visual user interface logic 120 for viewing and updating information associated with generating short life cycle product demand forecasts (e.g., seasonality curves, sales data, etc.). The graphical user interface may be associated with a demand forecasting application, and visual user interface logic 120 may be configured to generate the graphical user interface.

[0036] In one embodiment, computer system 100 provides at least the functionality disclosed herein and is a centralized server-side application accessed by many users via computing devices / terminals that communicate with computer system 100 (which functions as a server) over a computer network. Thus, display screen 24 may represent multiple computing devices / terminals that allow users to access and receive services from short life cycle product demand forecasting tool 110 via networked computer communications.

[0037] In one embodiment, computer system 100 further includes at least one database 17 operably connected to computing device 105 and / or a network interface for accessing database 17 via a network connection. For example, in one embodiment, database 17 is operably connected to visual user interface logic 120. According to one embodiment, database 17 is configured to store and manage data structures (e.g., records of sales data) associated with short life cycle product demand forecasting tool 110 in a database system (e.g., a computerized inventory management and demand forecasting application).

[0038] In one embodiment, visual user interface logic 120 is configured to generate a graphical user interface (“GUI”) to facilitate user interaction with seasonality forecasting model tool 110. For example, visual user interface logic 120 includes program code that generates the graphical user interface and causes the interface to be displayed based on an implemented graphical design of the interface. In response to user actions and selections via the GUI, associated aspects of the generation of sales curves and forecasting models for short life cycle retail items can be manipulated.

[0039] For example, in one embodiment, visual user interface logic 120 is configured to facilitate receiving input and reading data in response to user actions. For example, visual user interface logic 120 may facilitate the selection, reading, and input of sales data (e.g., seasonality information and sales quantity or monetary data) associated with retail items sold at a retail location. The sales data may be in at least one data structure (e.g., in database 17) associated with (and accessible by) a demand forecasting application (e.g., short life cycle product demand forecasting tool 110) via the graphical user interface.

[0040] The sales data may include, for example, data representing past sales and promotions of items similar to short life cycle items for which demand is determined over multiple past retail periods. The sales data may be segmented into past retail weeks, with each past week having a numerical value assigned to it to indicate the number of items sold (or the amount of money earned for the item) that week. According to one embodiment, the sales data may also include numerical values ​​representing price discounts and values ​​of other promotional elements over the retail period, as well as seasonality information (which may be separate from the sales data). According to one embodiment, the sales data for items may be accessed via network communications.

[0041] In one embodiment, the similar item sales curve generation logic / module 130 is configured to generate sales curves at various intersection levels for items that are similar to the short life cycle product for which a demand forecast is being determined, as disclosed below.

[0042] In one embodiment, sales curve parameterization logic / module 140 is configured to parameterize each sales curve generated by logic 130. In one embodiment, demand forecast generation logic / module 150 generates a forecast of demand for short life cycle products by aggregating the parameters generated in 140 using an error metric to arrive at a final sales curve for the short life cycle products for which it provides a demand forecast.

[0043] In one embodiment, the generated demand forecast predicts the amount of inventory needed (e.g., for an item at a single store), which is then used to instruct a computerized inventory system (e.g., by a computerized inventory control and demand forecasting system). The demand forecast may also control the amount of inventory (e.g., for an item at a single store) allocated by the computerized inventory system. Furthermore, the demand forecast may control adjustments to the amount of inventory (e.g., for an item at a single store) by the computerized inventory system.

[0044] FIG. 2 is a block diagram of a computer server / system 100 according to an embodiment of the present invention. FIG. 2 shows further hardware / software details of system 100. Although shown as a single system, the functionality of system 100 may be implemented as a distributed system. Furthermore, the functionality disclosed herein may be implemented on separate servers or devices that may be coupled via a network. Furthermore, one or more components of system 100 may not be included. For example, for server functionality, system 100 may need to include a processor and memory, but may not include one or more of the other components shown in FIG. 2, such as a keyboard or display.

[0045] System 100 includes a bus 12 or other communication mechanism for communicating information and a processor 22 coupled to bus 12 for processing information. Processor 22 may be any type of general-purpose or special-purpose processor. System 100 further includes memory 14 for storing information and instructions executed by processor 22. Memory 14 may be comprised of any combination of random access memory ("RAM"), read-only memory ("ROM"), static storage such as a magnetic or optical disk, or other types of computer-readable media. System 100 also includes a communication device 20, such as a network interface card, for providing access to a network. Thus, a user may connect with system 100 directly, remotely over a network, or otherwise. Some or all of the components of system 100 may implement the entire system.

[0046] Computer-readable media can be any available media that can be accessed by processor 22 and includes both volatile and nonvolatile media, removable and non-removable media, and communication media. Communication media may include computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media.

[0047] Processor 22 is further coupled to a display 24, such as a liquid crystal display ("LCD"), via bus 12. A keyboard 26 and a cursor control device 28, such as a computer mouse, are further coupled to bus 12 to enable a user to interact with system 100.

[0048] In one embodiment, memory 14 stores software modules that provide functionality when executed by processor 22. These modules include operating system 15, which provides operating system functionality for system 100. These modules further include short-lifecycle demand forecasting module 16, which implements one or more of modules 120, 130, 140, and 150 and all other functionality disclosed herein. System 100 may be part of a larger system. Thus, system 100 may include one or more additional functional modules 18 for including additional functionality, such as a retail management system (e.g., the Oracle Retail Demand Forecasting System or the Oracle Retail Advanced Science Engine (ORASE) from Oracle Corporation) or enterprise resource planning (ERP) or other types of inventory management systems. Database 17 is coupled to bus 12 and provides centralized storage for modules 16 and 18 to store customer data, product data, transaction data, and the like. In one embodiment, database 17 is a relational database management system ("RDBMS") capable of managing stored data using structured query language ("SQL"). In one embodiment, dedicated point of sale ("POS") terminals 99 generate transactional and historical sales data (e.g., data regarding transactions for each item / SKU (stock keeping unit) at each retail store) that is used to forecast demand. The POS terminals 99 themselves may include additional processing capabilities for forecasting demand according to one embodiment and can operate alone or in conjunction with other components of FIG. 2 as a dedicated demand forecasting system.

[0049] In one embodiment, especially with a large number of retail stores, a large number of items, and a large amount of historical data, database 17 is implemented as an in-memory database ("IMDB"). An IMDB is a database management system that relies primarily on main memory for computer data storage. It contrasts with database management systems that utilize disk storage mechanisms. Main-memory databases are faster than disk-optimized databases because disk access is slower than memory access, the internal optimization algorithms are simpler, and fewer CPU instructions are executed. Accessing data in memory provides faster and more predictable performance than disk because it eliminates seek times when querying data.

[0050] In one embodiment, database 17, when implemented as an IMDB, is implemented based on a distributed data grid. A distributed data grid is a system in which a collection of computer servers cooperate in one or more clusters to manage information and related operations, such as computation, in a distributed or clustered environment. Distributed data grids can be used to manage application objects and data shared across servers. Distributed data grids provide low response times, high throughput, predictable scalability, continuous availability, and information reliability. In a particular example, a distributed data grid, such as the "Oracle Coherence" data grid from Oracle Corporation, stores information in memory to achieve higher performance and uses redundancy in keeping copies of that information synchronized across multiple servers, ensuring system resilience in the event of a server failure and continuous availability of data.

[0051] In one embodiment, system 100 is a computing / data processing system that includes an application or collection of distributed applications for an enterprise organization and may also implement logistics, manufacturing, and inventory management functions. Application and computing system 100 may be configured to operate with or implemented as a cloud-based networking system, a software-as-a-service ("SaaS") architecture, or other type of computing solution.

[0052] Embodiments are disclosed in terms of an item (i.e., a class of items such as yogurt or men's shirts or individual SKUs) sold at a location (e.g., a retail location), where the item may be promoted in different ways at different times (i.e., predefined retail periods such as a day, a week, a month, or a year). A retail calendar has many retail periods (e.g., weeks), which are organized in a particular manner (e.g., four 13-week quarters) over a typical calendar year. Retail periods may occur in the past or the future. Historical sales / performance data may include, for example, the number of units of an item sold in each of multiple past retail periods and associated promotional data (i.e., for each retail period, which promotions were effective in that period), as well as other relevant demand characteristics / variables.

[0053] The embodiments generate short life cycle curve estimates by combining sales patterns at various nodes and using a bus diffusion model. The embodiments generate more accurate and robust short life cycle curves compared to known approaches.

[0054] To request a prediction for a short life cycle item, which is generally only available for a few months and therefore is newly introduced and therefore often has no historical sales history available, historical sales data for similar items must be received and used. For example, if the short life cycle item is a laptop computer (which will be on sale for approximately 10 weeks before being replaced by a similar laptop computer, perhaps with an improved processor or a larger solid-state drive (“SSD”)), sales history for previous “versions” of that laptop would be received. Similarly, for fashion items such as sports-related t-shirts, similar previously sold t-shirts can be taken into account.

[0055] Embodiments receive sales histories for "similar" sales items at various intersection levels, such as SKU / store, subclass / store, department store / region, etc., and then generate various sales / demand curves for the similar items at each of the various intersection levels. Similar sales items may be similar, for example, in that they were sold over approximately the same period in the previous year, or may be similar simply based on attributes (e.g., T-shirts, computers, etc.), even if they are not considered short life cycle items because the time frame in which they were sold spans multiple months or years. The latter scenario can be used when there is no other similar item previously sold that is also considered a short life cycle item. In general, retailers typically select "similar" products based on product characteristics such as type, size, color, shape, etc.

[0056] In some embodiments, any known solution can be used to determine the sales / demand curves for similar items at each of the various intersection levels. In one embodiment, instead of generating sales curves at the various intersection levels, an automated clustering predictive model is used to generate sales curves for clusters of similar items, and these sales curves can be combined with the sales curves of similar items. In this embodiment, promotion effects / features are estimated using pooled sales data points at the aggregate level by automated clustering of these data points. Each of the multiple cluster models is trained using regression, and then each saved cluster-trained model is used to forecast demand for retail products / items.

[0057] As mentioned above, in the retail industry, retailers need to forecast future demand in order to better manage inventory or promotion / markdown plans. To accurately forecast demand, retailers take into account all factors that may affect demand, such as promotions, price fluctuations, seasonality, weather, etc. Known solutions for retailers have used various algorithms to estimate promotion or price effects.

[0058] Furthermore, a single product / location combination typically does not have enough sales observations to generate a robust effect estimate. To take this into account, effects are typically estimated using a large number of observations pooled across different products / locations / periods, which are then used to forecast demand for the pooled products / locations during the estimation phase. Known solutions typically pool data based on information related to product and location hierarchies (e.g., pooling data within the same subclass and region). Then, during forecasting, the promotion effect is the same across the entire forecast period and for all products and locations that participated in the estimation.

[0059] However, these known solutions ignore at least two facts: (1) each product / location is affected differently by the same promotion. For example, a product with slow sales will be affected differently from a product with strong sales, or different store styles will be affected differently; and (2) customers will respond differently to the same event depending on the time of year. In other words, the effectiveness of a promotion has an expiration date. For example, an ice cream promotion will be more effective in hot summer than in cold winter.

[0060] In contrast, embodiments use machine learning and automated clustering to estimate the impact of promotions on demand by taking into account the timing of events and details of each product / location during estimation using pooled data. These details and timing are collectively referred to as "features." Examples of features include base sales, price, seasonality, brand, promotion, size, color, pack size, supplier, length, etc. While features such as price and seasonality may be relevant to all types of products, other features are item-specific. For example, pack size influences the demand for yogurt, but pack length is not important. Conversely, brand is very important for fashion items but much less important for hardware items such as nails or hammers.

[0061] Embodiments estimate promotion effects at an aggregate level by pooling all data together. The resulting effects are product / location specific and have expiration dates, meaning they can change depending on the time period in which the event occurs. In general, in embodiments, promotion effects are estimated by pooling data at an aggregate level, and different products / locations / times are automatically grouped into clusters. Promotion effects are dynamically fetched during the forecasting phase based on product / location and time-related features. Even for new product / location combinations that do not have historical demand, embodiments can generate promotion effects with expiration dates.

[0062] Figure 3 is a flow diagram of the functionality of the short life cycle product demand forecasting tool 110 of Figure 1 in estimating promotional effectiveness that can be used in demand forecasting, according to one embodiment. In one embodiment, the functionality of the flow diagram of Figure 3 (and Figure 6 below) is implemented by software stored in memory or other computer-readable or tangible medium and executed by a processor. In other embodiments, the functionality may be performed by hardware (e.g., by using an Application Specific Integrated Circuit ("ASIC"), a Programmable Gate Array ("PGA"), a Field Programmable Gate Array ("FPGA"), etc.), or any combination of hardware and software.

[0063] At 302, historical item sales data is received for all items / SKUs in all stores, for a particular class / category of products, or for just one item of interest. For example, a class / category may be "yogurt," "coffee," or "milk." Each class has one or more subclasses down to the SKU or Universal Product Code ("UPC") level, which would be each individual item offered for sale. For example, in the yogurt class, subclasses may be each brand of yogurt, and further subclasses may be flavors, sizes, types (e.g., Greek yogurt or regular yogurt) down to the SKU, which would correspond to each individual different type of yogurt item sold.

[0064] The historical sales and performance data may include, for example, data representing past sales and promotions for each item over multiple past retail sales periods. The historical performance data may be segmented into past retail weeks, with each past week having a numerical value assigned to it to indicate the number of items sold that week. According to one embodiment, the historical performance data may also include numerical values ​​representing price discounts over the retail period and values ​​of other promotional elements. According to one embodiment, the historical performance data for an item may be accessed via network communications, including from each POS terminal 99 at each retail store and / or from database 17.

[0065] The historical performance data includes sales data associated with multiple promotional elements over multiple time periods (e.g., weeks). Examples of promotional elements include, but are not limited to, price discount elements, television advertising elements, radio advertising elements, newspaper advertising elements, email advertising elements, internet advertising elements, and in-store advertising elements. For each item, the historical data includes a list of features / variables / attributes for that item, such as price, promotion, seasonality, brand, color, style, etc.

[0066] The historical sales data is received as multiple data points or "data sets," a single data point for each sale of an item per store (i.e., at the product / store / week level in embodiments where "week" is the desired time period). In embodiments, these data points may be received by electronically parsing data generated by all POS 99s at all relevant retail stores.

[0067] At 304, various types of features are extracted at the aggregate product / location (store) level. Many of these features are extracted from non-sales related data such as ERP, merchandising and inventory control systems associated with the relevant store. Some of these features (e.g., sales data, promotions running during a sales period) are extracted from the sales data at 302. In an embodiment, the types of features extracted include:

[0068] a.Product-related features p1..p n (e.g., national brands vs. private labels, package size, low sales vs. high volume, units of measure, etc.) b. Store Features s1..s m (e.g. convenience stores vs. supermarkets, densely populated areas vs. non-densely populated areas, different store layouts, etc.) c. Timing-related features t1..t k (e.g., holiday season vs. non-holiday season, hot days vs. cold days, low traffic vs. high traffic, football season, etc.) d. Promotions and Prices x1..x j At 306, embodiments may select the product features (p1..p n ), location features (s1..s m ) and timing features (t1..t k ) to generate "N" clusters of data points from 302, and then create the N clusters (c1..c N) where the number of clusters N can be a user-defined number. Clustering is a type of data mining in which a set of objects (i.e., features) are grouped such that objects within the same group (cluster) are more similar to each other in some sense than objects in other groups (clusters). In one embodiment, a k-means clustering algorithm is used. K-means clustering aims to divide n observations into k clusters, with each observation belonging to the cluster with the closest mean value, which serves as the prototype for the cluster. However, other clustering algorithms, such as hierarchical clustering, can also be used.

[0069] In one embodiment, the following k-means clustering algorithm is used: k-mean value m1 (1) ,...,m k (1) Given an initial set of, the algorithm proceeds by alternating between two steps.

[0070] (1) Assignment step: Assign each observation to the cluster with the closest mean, i.e., the least squared Euclidean distance. (Mathematically, this means dividing the observations according to the Voronoi diagram generated by the means.)

[0071]

number

[0072] In the formula, each x p is assigned to exactly one S, even if it can be assigned to two or more of them. (t) is assigned to.

[0073] (2) Update step: Recalculate the mean (centre of mass) for the observations assigned to each cluster.

[0074]

number

[0075] When the allocation no longer changes, the algorithm has converged. At 308, each cluster generated at 306 is trained to generate a trained cluster model (i.e., promotion effects per cluster generated at 306). In one embodiment, training is performed by running a regression on each cluster i of the N clusters to generate N sets of promotion / price effects (e1..e k )1...(e1..e k ) N In one embodiment, linear regression is used for training or effect estimation, although other regression algorithms can be used. Each set of promotion / price effects is an optimized set of effects for the features of the corresponding cluster. The features in each cluster can be decomposed into products / locations / time periods. Each product / location / time period is mapped to one of N sets of effects.

[0076] At 310, the clustering models are saved in binary files that can be easily loaded on demand. For example, in embodiments, these models are trained using features of a language specific to machine learning, such as Python or C++. These models may be saved in binary files with a specific serialized format. For example, a model trained in Python may be saved in a Python pickle binary file.

[0077] At 312, if a demand forecast for retail products is desired, for each product / location / week, the saved cluster trained model is used to map it to one of N clusters by identifying features for a particular week. For example, for cluster i, the effect set (e1..e k ) i Promotion / price effects from are fetched.

[0078] At 314, the promotion / price effect from 312 is used to forecast final demand. In one embodiment, the following demand forecasting algorithm is used to forecast demand: Demand = Base Demand x Promotion Lift x Price Lift. This is a relatively simple demand algorithm; other embodiments may use much more complex demand forecasting algorithms as long as the promotion effect is incorporated into the demand forecast in some way. Embodiments may forecast demand on a per SKU / store basis, or may forecast demand for all SKUs in all stores simultaneously using a large matrix (i.e., each row is one SKU / store).

[0079] The functionality of Figure 3 can be used to generate multiple sales curves for SKUs at various intersections by clustering products / locations into different groups based on product attributes such as color, size, store format, and sales length for each given intersection, and a short life cycle sales curve (i.e., a demand forecast for each intersection) as disclosed below is then generated, as used in connection with Figure 6.

[0080] Figures 4A-4B and 5 illustrate examples of implementations of embodiments of the present invention and how each corresponds to the functionality of Figure 3. In a relatively simple example, a retailer might model demand using the following equation:

[0081] Demand = Base demand x Promotional lift x Price lift For simplicity, let's assume that the base demand is 5 for all product / store combinations.

[0082] Historical sales data points are received at 302. In a simple example, suppose there is information about two stores and six products from week1_2018 to week52_2019 (i.e., every week of 2018 and 2019).

[0083] At 304, an embodiment associates each item / store / week sales data received at 302 with product, location, and time period related extracted features. These extracted features include product brand, package size, store type, weather for the corresponding week, what sporting events are taking place in the corresponding week, price, promotions 1, 2, 3, 4, etc.

[0084] At 306, a clustering algorithm is run to cluster the sales into N clusters (i.e., three clusters in this example). In Figure 4A, each cluster (402, 403, 404) is shown with a different cross-hatching scheme. The table shown in Figure 4A assigns one of the three clusters to each product in each store each week as a result of the clustering.

[0085] At 308, for each cluster 402-404, a regression model is trained using the sales data and its features by using the data in each cluster to generate promotion effects per cluster 402-404. FIG. 4B shows the trained promotion effects of price and promotions 1-4 for each cluster. The effects generated at 308 are based on the effects that make up the demand model (in this simple example, only price and promotions). In more complex demand models, the generated effects could also include holidays, back-to-school events, weather, etc. Each effect number in FIG. 4B acts as a multiplier.

[0086] At 310, the trained clustering model and three trained regression models are saved to a binary file. The binary file may include both the clustering model and the regression model. Figure 4A shows a clustering model, and Figure 4B shows a regression model. In an embodiment, the clustering and regression may be performed separately and then saved to different binary files or a single file.

[0087] At 312, to create forecasts for week1_2020 through week4_2020 (i.e., the first four weeks of 2020), the characteristics of those weeks are first identified. The saved cluster model is then used to forecast each item / store / week into one of three clusters. Figure 5 illustrates how an embodiment maps each product / location / week to a cluster. For example, store 1, product B, week 2 is mapped to cluster 404, and store 2, product C, week 3 is mapped to cluster 403.

[0088] At 314, for each item / store / week, the base demand plus the effects (price, promotion) in the corresponding cluster are applied to create a forecast using the demand formula: Demand = Base Demand x Promotion Lift x Price Lift. In this example, the base demand for product A at store 1 is 2.0, and if promotions 1 and 3 are implemented in week 2_2020, the demand for product A at store 1 in week 2_2020 is Demand = 2 x 1.05 x 3.2 = 6.72.

[0089] FIG. 6 is a flow diagram of the functionality of the short life cycle product demand forecasting tool 110 of FIG. 1 in generating sales / demand curves for short life cycle items (i.e., a given item / location or SKU), according to one embodiment.

[0090] At 602, multiple sales curves associated with an item or items that are similar to the short life cycle item are generated. In general, these multiple sales curves are then "blended" using the remaining functionality of FIG.

[0091] In one embodiment, for items similar to short life cycle items, sales curves are generated at various ("n") intersection levels by using historical sales data for similar items (because there is no or sparse sales history available for short life cycle items). For example, intersection levels may be SKU / store, subclass / store, department store / region, etc. Historical sales from SKU / store are aggregated to higher intersection levels to determine the sales curve for that intersection level. In some embodiments, any known function can be used to determine the sales curve (i.e., predicted demand).

[0092] In one embodiment, the functionality disclosed above in conjunction with FIG. 3 is used to generate multiple sales curves for similar items at 602. Specifically, the end result of the functionality of FIG. 3 may be a first SKU and a corresponding cluster of similar items, a second SKU and a different corresponding cluster of similar items, etc. The sales curve for each SKU and the sales curve for each corresponding cluster may be generated using known methods. Assuming the first and second SKUs are also similar to short life cycle items, as disclosed above, these sales curves may be used as input to the functionality of FIG. 6 below and then “blended” to form the short life cycle item sales curve. In one embodiment, only two sales curves may be generated, i.e., a single similar SKU and corresponding cluster. In other embodiments, if multiple similar SKUs are available, many sales curves may be generated for each similar SKU and for each corresponding cluster.

[0093] When the functionality of FIG. 3 is used, at 304, the extracted features may, in one embodiment, include extracting multiple different types of features related to sales for each product, including product characteristics such as color and size, location characteristics such as price range, store format, calendar-related characteristics such as total weeks on shelf, first week on shelf, etc.

[0094] In an embodiment, the sales curve time frame in 602 would approximately match the desired sales curve time frame for the short life cycle item. In other words, if the short life cycle item is expected to have a 10-week sales season during the summer of 2021, the sales curves for similar items should also be oriented toward those same 10 weeks or a close equivalent (e.g., 12 weeks).

[0095] At 604, the sales curve at each intersection point "i" is parameterized using a "Bass diffusion" model. The Bass diffusion model contains differential equations that describe the process of how new products are adopted by people. The model provides a basis for how current and potential adopters of a new product interact. The basic premise of the model is that adopters can be classified as innovators or imitators, and the speed and timing of adoption depend on the degree of innovation and imitation among adopters. The Bass diffusion model has been widely used in forecasting, particularly in new product sales forecasting and technology forecasting. Mathematically, the basic Bass diffusion model is a Riccati equation with constant coefficients.

[0096] In an embodiment, a Bass diffusion model is used to parameterize each sales curve from 602 as follows:

[0097] For each intersection point (or, in other embodiments, each sales curve) i, By using a least squares optimizer to minimize the objection MSE(i), m is calculated based on the following formula: i ,p i ,q i Estimate parameters / coefficients.

[0098]

number

[0099] In the formula, F i(t) is the estimated sales for week t at intersection point i, and S i (t) is the actual sales history of intersection point i in week t. E i is the error for the current intersection point i.

[0100] At 606, p from 604 is calculated based on the weighted error. i and q i Parameters can be combined. i is also called the innovation coefficient, external influence, or advertising effect. i The parameter is also called the imitation coefficient, internal influence, or word-of-mouth effect. i No parameters are needed or used. i and q i The parameters are combined as follows: W i is the determined weight for each intersection point i.

[0101]

number

[0102] At 608, for a given length of short life cycle sales length l, an embodiment calculates the curve value at the τth week as follows:

[0103]

number

[0104] The final curve for a given product / location at τ is calculated in 606 by normalizing l as follows:

[0105]

number

[0106] This generates a final sales / demand curve that provides a forecast of sales for the short lifecycle item for each week (or other pre-defined period) over the sales period (e.g., weeks 20-30 of 2021).

[0107] At 610, the sales curve or demand forecast from 608 is used in logistics, including being the basis for determining how many items to manufacture, inventory levels, and levels of shipment to specific stores. The output at 608 can be in the form of specialized data structures that can be used in conjunction with fully automated manufacturing, inventory, and logistics systems, as described more fully below.

[0108] 7 and 8 illustrate examples of implementations of embodiments of the present invention and how each corresponds to the functionality of FIG. 6. In a relatively simple example, at 602, sales curves for similar items are generated for only two intersection levels: SKU / store and SKU / region. FIG. 7 shows the SKU / store sales curve at 701 and the SKU / region sales curve at 702. In FIG. 7, the X-axis is the calendar (e.g., week 1, week 2, etc.) and the Y-axis is the shape / value of the curve. Sales curves are typically normalized to the number of weeks. For example, for a 10-week sales curve, the curve values ​​could be 0.76, 0.88, 1.02, 1.15, 1.2, 1.28, 1.11, 1.05, 0.8, and 0.75. As noted, these values ​​sum to 10, which is equal to the length of the curve in weeks.

[0109] In 604 and 606, m i ,p i ,q i ,E i and W i The parameters are calculated (i.e., the sales curve is "parameterized") 801 in Figure 8 shows the calculation results in this example.

[0110] At 606, the p and q parameters are combined based on the weighted errors. 802 in FIG. 8 shows the calculation result in this example. After the p and q parameters are combined, a combined sales curve is calculated at 608. Sales curve 703 in FIG. 3 is the combined sales curve and represents the demand forecast for short life cycle items. The lengths of the curves in FIG. 7 do not necessarily have to be equal. The formula includes a factor "l," which is the length of the curve. This allows curves of different lengths to be combined. The final sales curve does not indicate the number of items predicted to be sold. Instead, it indicates the shape of the sales curve during its life cycle.

[0111] FIG. 9 illustrates an integrated manufacturing, inventory, and logistics system 900 including the demand forecasting disclosed herein, according to one embodiment. An embodiment can fully automate the process using a dedicated demand forecast data structure for short lifecycle items generated in 608. As shown in FIG. 9 , the system 900 can include a product demand forecasting system 970 that forecasts future product demand and, in some applications, predicts and / or accounts for future demand for hundreds of thousands of products, or in some applications, tens of millions of products or more, at one or more retail stores 901-904. The forecasting system 970 communicates with one or more inventory systems 920 and one or more manufacturing systems 980 via a cloud network 950 or other type of communication network.

[0112] Forecasting system 970 generates demand forecasts by performing the functions disclosed in conjunction with Figure 3 and / or Figure 6 above. Inventory system 920 stores inventory and provides transportation logistics for delivering items to stores 901-904 using trucks 910-913 or other transportation mechanisms. In one embodiment, inventory system 920 implements an enterprise resource planning ("ERP") dedicated computer system or dedicated inventory control system that uses input from demand forecasting system 970 to determine inventory levels and the quantity and timing of item deliveries to stores 901-904. In some embodiments, the functionality of Figure 9 can be fully automated using automated loading mechanisms and autonomous vehicles and dedicated data structures.

[0113] Manufacturing system 980 manufactures items sent to inventory system 920 and provides transportation logistics for delivering these items to inventory system 920 using trucks 981 or other transportation mechanisms. In one embodiment, manufacturing system 980 implements an ERP-specific computer system or a specific manufacturing system that uses input from forecasting system 970 to determine the quantity of items to manufacture, the inventory of resources used in manufacturing, and the quantity and timing of delivery of items to inventory system 920.

[0114] The forecasting system 970 may utilize information from the inventory system 920, sales tracking systems (not shown), and / or databases in forecasting demand for products. In forecasting demand, the forecasting system 970 attempts to anticipate uncharacteristic demand for one or more products due to events, weather, societal demand, economic factors, and other factors. Dozens, hundreds, or even thousands of different variables may be tracked that can affect demand for one or more products. Changes in these variables can result in uncharacteristic demand. For example, predicted weather changes can be tracked, and one or more variables associated with the predicted weather can be used to determine whether such changes are weather that may affect demand, and further predict changes in demand.

[0115] Generally, the elements in FIG. 9 perform sales, manufacturing, or inventory consumption. Direct-to-consumer retail locations / stores 901-904 exhibit the most volatile inventory patterns due to the random nature and external factors affecting sales. However, manufacturing facilities and sites that consume inventory (e.g., product integrators for products used in local facilities, internet shippers, etc.) also benefit from the demand forecasting disclosed herein. As disclosed, each retail location 901-904 sends sales data and historical forecast data to a forecasting system 970. The sales data includes inventory shrink statistics for each item or SKU / UPC for each sales period, typically several days, over the previous sales cycle (i.e., week), typically a 4-7 week inventory cycle.

[0116] Forecasting system 970 stores sales data in repository 972 and uses this sales data to generate orders to replenish inventory. These orders include a set of items and the quantity of each item to maintain inventory levels at stores 901-904.

[0117] Many retail ordering schemes rely on sales periods and days of the week in a sales cycle. In one configuration, in an inventory management environment with inventory statistics specific to each day of the week, the inventory system 920 determines target inventory levels by collecting inventory level statistics from previous sales for each day of the week. Based on the inventory level statistics, an embodiment calculates inventory levels for each day of the week such that safety stocks absorb inventory fluctuations for each day of the week. For each of a plurality of items, an embodiment renders a stock level indicating the target inventory level, including safety stock, for each day of the week. An embodiment calculates order quantities based on lead times so that the quantity ordered meets the rendered stock level for a given day of the week. Identifying actual stock levels involves identifying stock levels for a particular day of the week from past weeks from historical data, thereby focusing on the same day of the week over time rather than an average for all days of the week.

[0118] In certain configurations, the disclosed embodiments may be utilized with dedicated and / or particularly high-volume retail environments. In large-scale logistics and distribution operations, it is beneficial to fill trucks as fully as possible, and if it is necessary to defer an item to the next trip, it is beneficial to select items that are least likely to interrupt sales activity. Thus, embodiments are operable with POS systems 99 to identify high-velocity or high-turnover items that tend to be sold and replenished faster than other items. An item's UPC barcode symbology or radio-frequency identification ("RFID") includes a field, name, or value that, alone or in conjunction with a database lookup, designates the item as a high-velocity item suitable for safety stock processing as defined herein.

[0119] High velocity items may be accommodated by identifying, for each of a plurality of items represented in the inventory database, a field for a product identifier and a field indicating the safety stock for that item, and for each product identifier, determining a product segmentation field indicating an increase in product replenishment demand due to sales based on product velocity. The disclosed embodiments determine whether to calculate safety stock, i.e., whether the overhead and burden of replenishing according to the safety stock is worthwhile given product throughput, based on the velocity field.

[0120] In other embodiments, supply logistics may call for a higher delivery frequency than one truck per day, and therefore operate replenishment windows with greater granularity. In such cases, safety stocks may be more specific than individual days, such as Monday morning and Monday afternoon, or may specify multiple deliveries or time windows within a particular day of the week, such as 7:00 AM, 11:00 AM, and 4:00 PM.

[0121] Embodiments including the generated demand forecasts can be utilized in implementing supply logistics to assign flights (i.e., trucks) and manifests (i.e., items to be included) according to demand and profit margins for the items transported. Express items may be considered to have priority space on a particular flight, but may further be selected based on profit margins or markups for included items, with items having the greatest revenue generating potential being selected for inclusion.

[0122] In such a product inventory shipping environment using demand forecasting as disclosed herein and having multiple transport vehicles, each vehicle (e.g., truck) is configured to receive a fixed payload of items for delivery to a sales location for inventory replenishment. Embodiments can provide guidance in loading the delivery vehicle by calculating a safety stock for each of a plurality of items, including a first item and a second item, and determining an amount of each of the first item and the second item to be loaded onto the delivery vehicle based on the calculated safety stock for the first item and the second item. Embodiments recalculate the truckload based on the safety stock if the available space within the delivery vehicle is insufficient for the determined amount of the first item and the second item, meaning that certain items need to be excluded and postponed to a subsequent delivery trip.

[0123] As disclosed, embodiments combine sales patterns of items that are similar to short-lifecycle items at various intersections, such as SKU / store, subclass / store, department store / region, etc. This makes the curve for short-lifecycle items more robust. Embodiments then parameterize the curve using two parameters and then regenerate the curve by applying the expected number of weeks the product will be on the shelf. This approach makes recalculating the curve very convenient, even if the sales pattern changes year-over-year. Given the parameterization of the curve, combining sales patterns is independent of the length of the curve.

[0124] Although several embodiments have been specifically shown and / or described herein, it will be understood that variations and modifications of the disclosed embodiments are encompassed within the scope of the appended claims in accordance with the above teachings without departing from the spirit and intended scope of the invention.

Claims

1. 1. A method for generating a short life cycle sales curve for a short life cycle item, comprising: one or more processors receiving historical sales data for at least one similar item at various intersection levels that is similar to the short life cycle item; the one or more processors using the historical sales data to generate a plurality of similar sales curves corresponding to the at least one similar item; and parameterizing each of the similar sales curves, the parameterizing step including estimating, for each similar sales curve i, a parameter m i , an innovation coefficient parameter p i , an imitation coefficient parameter q i , and an error parameter E i using the following equation: [Equation 1] F i (t) is the estimated sales for week t for each similar sales curve i, and S i (t) is the actual sales history for week t for each similar sales curve i, and the method further comprises: the one or more processors determining a weight W i for each error parameter; and the one or more processors combine the innovation coefficient parameter p i and the imitation coefficient parameter q i based on the weights W i using the following equation: [Equation 2] The method further comprises the one or more processors generating the short life cycle sales curve using the combined innovation factor parameter and the combined imitation factor parameter.

2. The method of claim 1, further comprising the step of the one or more processors generating a demand forecast for the short life cycle item using the short life cycle sales curve.

3. 3. The method of claim 1, wherein the plurality of similar sales curves span a first period of time and the short life cycle sales curve spans a second period of time that is different from the first period of time.

4. The one or more processors generating a dedicated data structure representing the demand forecast; the one or more processors transmitting the dedicated data structure to an automated logistics system; 3. The method of claim 2, wherein in response to receiving the specialized data structure, the automated logistics system automatically transports one or more of the short life cycle items to at least one retailer.

5. The step of generating a plurality of similar sales curves comprises: receiving historical sales data for an aggregate product / location level, the historical sales data including a plurality of sales data points, the plurality of sales data points including a sales data point for a first item at each of a plurality of locations, and generating a plurality of like sales curves further comprising: extracting a plurality of different types of features associated with the sale of each of the similar items; generating a plurality of clusters of sales data points based on the plurality of different types of features; and training each of the clusters to generate a plurality of trained cluster models, the plurality of trained cluster models including a promotion effect per cluster, wherein generating a plurality of similar sales curves further comprises: For a specific time period, a specific location and the first item, identifying the features for the specific time period and mapping them to one of the trained cluster models to fetch the promotion effect for the specific time period; generating a first sales curve corresponding to the first item and generating a second sales curve corresponding to the mapped trained cluster model.

6. The step of generating the short life cycle sales curve comprises: [Equation 3] Calculate as follows: [Equation 4] The method of any one of claims 1 to 5, comprising the step of normalizing l so that:

7. A computer readable program for causing one or more processors to carry out the method of any one of claims 1 to 6.

8. A system for generating a short life cycle sales curve for a short life cycle item, comprising: one or more processors coupled to one or more point of sale systems, the one or more processors comprising: receiving historical sales data for at least one similar item similar to the short life cycle item at various intersection levels; generating a plurality of similar sales curves corresponding to the at least one similar item using the historical sales data; parameterizing each of the similar sales curves, the parameterization including estimating, for each similar sales curve i, a parameter m i , an innovation coefficient parameter p i , an imitation coefficient parameter q i , and an error parameter E i using the following equations: [Equation 5] F i (t) is the estimated sales for week t for each similar sales curve i, and S i (t) is the actual sales history for week t for each similar sales curve i, and the one or more processors further: Determine a weight W i for each error parameter; The innovation coefficient parameter p i and the imitation coefficient parameter q i are combined based on the weight W i using the following formula [6]: [Equation 6] The one or more processors further generate a short life cycle sales curve using the combined innovation factor parameter and the combined imitation factor parameter.

9. The system of claim 8 , wherein the one or more processors further generate a demand forecast for the short life cycle item using the short life cycle sales curve.

10. The one or more processors further comprise: generating a specialized data structure representing said demand forecast; transmitting said dedicated data structure to an automated logistics system; 10. The system of claim 9, wherein in response to receiving the specialized data structure, the automated logistics system automatically transports one or more of the short life cycle items to at least one retailer.

Images