Patent Application
20240144037
One of the largest causes for shrink in retail is spoilage, i.e., perishable items that have exceeded their expiration dates. Retailers apply price markdowns as a primary technique to move at-risk inventory before spoilage occurs. Current solutions in the industry focus on the operational aspect of markdowns referred to as “reduce to clear,” rather than the decision-making aspect of inventory optimization. For example, retailers depend on store walk-throughs to manually identify inventory at risk.
Spoilage-based shrink accounts for approximately 3.1% of retail shrink overall. Departments most impacted by spoilage typically include produce, deli, bakery, prepared foods, seafood, and meat. Margins are thin for grocery stores, such that any reduction in spoilage-based shrink can substantially improve store profitability. Unfortunately, current approaches rely almost exclusively on predefined markdown policies and the experience/discretion of workers, which is not flexible, sustainable, or optimal.
In various embodiments, a system and methods for optimizing markdowns to reduce loss of perishable items are presented. A machine learning model (MLM) is trained on data relevant to markdowns associated with items subject to spoilage and shrink. The data is collected from a variety of store or retailer systems and input features are derived there from. The input features are provided as input to the MLM and the MLM produces, as output, various guidance associated with the markdown process including, without limitation, an indication as to whether a given item should or should not be marked down, and if a markdown is indicated for an item, a predicted number of markdown level(s) along with a predicted quantity of the item to markdown and a predicted price markdown at each markdown level. The MLM's predicted output is optimized to reduce item spoilage and increase item sales and margins. Item identifiers, the input features corresponding to the item identifiers, and the corresponding output from the MLM may be provided as a network-based service to a retailer. The network-based service can be integrated into existing workflows and interfaces associated with a retailer's markdown processes so as to provide enhanced features to the existing workflows and interfaces, which in turn, increase each item's sales and margins, while at the same time decreasing each item's spoilage and shrink rates.
Current approaches directed to reducing spoilage-based shrink are based on predefined policies and the experience/discretion of workers. Typically, a price markdown is a process by which a store associate manually reduces prices of soon-to-be expired products by printing and attaching a dedicated barcode to each item being marked down. The predefined policy may call for multiple levels of price markdown associated with price markdown amounts. For example, a level-1 markdown could be a 10% discount while a level-4 markdown could be a 40% markdown. Store associates markdown items on a daily basis, and if needed, they may increase the level of markdown for items that were not sold the day before.
Markdowns are an effective way to reduce shrink of perishable items, but there are various disadvantages. For example, existing solutions do not provide an effective way to track markdowns and at-risk inventory. The current approaches focus on facilitating operational implementation of the predefined policies but do not capture sufficient data on the markdown process and the at-risk inventory to provide an optimized, decision-based markdown process. For example, data that is not captured by existing solutions includes, among other things, product expiration date, markdown cycle (e.g., markdown level), total number of units marked down, and total number of marked down items sold. Even assuming availability of the markdown data identified above, the data is dispersed across multiple logs, data stores, and systems of the retailer such that conventional solutions fail to collect, assemble, and/or process the data in an optimal manner.
Additionally, in conventional approaches, the discretion and experience of store associates play a significant role in how markdowns are made. This discretion may include manual, intuition-based decisions relating to 1) the amount of time before an item's expiration date to initiate a markdown, 2) how many items to mark down on any given day, 3) what level of markdown to apply to a given item; and 4) what items recently received from the warehouse have limited shelf lives. While other store employees may have oversight of the store associate's markdowns, these employees also largely base their decisions on intuition and experience. This manual, experience/intuition-based approach, however, fails to consider such things as: 1) what is the impact on sales for each markdown level, 2) how are markdowns impacting, sales, margin, and shrink within a given item category, 3) what is the impact of markdowns on other adjacent items (e.g., similar items or complimentary items), and 4) how are markdown items impacting item returns and/or customer complaints.
These additional considerations during the markdown process are important, not only to reduce shrink but also to manage store-level sales and margins as well as shopper behaviors. For instance, some shoppers may avoid buying items at their regular price and may look only for markdowns before making a purchase. That is, customers may look for markdowns as an incentive to purchase items that they otherwise may not have. In other scenarios, a comprehensive analysis may suggest that retailer will actually lose less if they intentionally avoid marking down certain items, and instead absorbing the cost of spoilage. Thus, because existing solutions rely on human intuition alone and fail to optimize the markdown process based on data-driven, predictive analytics that assesses variables that influence the broader impact of the markdown process, they fail to identify the underlying factors/circumstances that could result in the markdown process negatively impacting the overall sales of a store.
As will be demonstrated herein and below, the above-noted technical problems associated with the existing manual, intuition-driven approaches for determining when and how to perform a markdown are solved by embodiments of the technology disclosed herein according to which historical markdown-related data is captured from multiple disparate data sources and provided as input to a machine learning model (MLM) that predicts various markdown parameters including, without limitation, the timeframe of a markdown (e.g., when to initiate a markdown of an item, how many markdown level(s) to go through, the duration of each markdown level, etc.); the price discount to apply for each markdown level; the impacts of the markdown on item inventory, sales, margins, and customer behavior; and so forth. The MLM is optimized and continuously retrained to improve its F1 predicted values based on actual outcomes associated with marked down items. The predicted values produced by the MLM may be integrated into existing markdown workflows/interfaces and monitored for accuracy from retail systems.
Furthermore, the various components (that are identified in
System 100 is data-driven and provides real-time and dynamic optimized markdown predictions and instructions for marking down items of a store. The predictions and instructions are integrated into existing workflows, existing services, existing interfaces, and existing systems as data-driven enhancements to markdown processes.
System 100 includes a cloud 110 or a server 110 (hereinafter simply “cloud 110”), transaction terminals 120, retail servers 130, and user-operated devices 140. Cloud 110 includes a processor 111 and a non-transitory computer-readable storage medium 112, which includes executable instructions for a trainer 113, one or more MLMs 114, and a shrink optimizer 115. Processor 111 obtains or is provided the executable instructions from medium 112 causing processor 111 to perform operations discussed herein and below with respect to 113-115.
Each transaction terminal 120 includes a processor 121 and a non-transitory computer-readable storage medium 122, which includes executable instructions for a transaction manager 123 and a shrink application 124. Processor 121 obtains or is provided the executable instructions from medium 122 causing processor 121 to perform operations discussed herein and below with respect to 123-124.
Each retailer server 130 includes a processor 131 and a non-transitory computer-readable storage medium 132, which includes executable instructions for a store manager 133, a shrink optimizer interface 134, and a variety of systems 135. Processor 131 obtains or is provided the executable instructions from medium 132 causing processor 131 to perform operations discussed herein and below with respect to 133-135.
Each user-operated device 140 includes a processor 141 and a non-transitory computer-readable storage medium 142, which includes executable instructions for a shrink application (app) 143. Processor 141 obtains or is provided the executable instructions from medium 142 causing processor 141 to perform operations discussed herein and below with respect to 143.
Trainer 113 trains MLM 114 on input features to produce fine-grained predictions. The input features may be based on actual observed information associated with previous markdowns on previous items within a given store of the retailer. The input features can be obtained by trainer from systems 135 and/or store manager 133 of a given retailer associated with the given store.
For example, trainer 113 can obtain a listing of perishable items for a given store that is subject to spoilage (and thus shrink) when the items fail to sell before their expiration date. This listing can be obtained from an item catalogue system 135 of retail server 130 for a given retail store. Each item may be identified by an item type within the list; for example, chicken breasts, pre-peeled shrimp, T-bone steak, etc. Trainer 113 assembles training data sets for each item type using historical data captured by systems 135.
Each training data set for each item type includes historical data gathered by trainer 113 over a given interval of time (e.g., a week, a day, a few days, two weeks, etc.). For each interval of time, trainer 113 obtains, from the historical data, a total number of items of the given item type (hereinafter referred to a “units of the item”) available for sale at the store. The available units for the current interval of time can be obtained from historical data associated with the retailer's inventory system 135 and transaction system 135. Trainer 113 retains the available units in the corresponding training data set for the current interval of time as a first labeled input feature to MLM 114.
Next, for the current interval of time, trainer 113 obtains the expiration date for each unit, which can be obtained from an inventory system 135 when the item unit was scanned into inventory at the store. The expiration date data for the current interval of time is retained within the corresponding training data set as a second labeled input feature to MLM 114.
Trainer 113 also identifies, for each day within the current interval of time, the number of days until the expiration date for a given unit. The number of days until expiration for each unit of item within the current interval of time is retained as a third labeled input feature to MLM 114.
For the current interval of time, trainer 113 identifies how many units of a given item are already marked down from their original listed price using historical data associated with the store's inventory and/or transaction systems 135 where a price change would appear for the item. The total number of units already marked down in the current interval of time is retained as a fourth labeled input feature to MLM 114. The total number of markdown cycles active for the units is also determined based on the markdown levels identified for the units of the given item type present in the current interval of time, and this may be retained, along with the current interval of time, as a fifth labeled input feature to MLM 114. The original price of each unit may also be identified from the historical data and retained, along with the current interval of time, as a sixth labeled input feature to MLM 114.
Next, for each current interval of time, sales data associated with each unit, and broken down by that unit's markdown levels (e.g., markdown cycles) for a given day and a given item, may be obtained from the transaction system 135 and retained, along with the current interval of time, within the corresponding training data set as a seventh labeled input feature to MLM 114.
Trainer 113 may also calculate spoilage per day for a given item type as an eighth labeled input feature to MLM 114. This can be calculated based on units that expired before a sale of the units occurred, from the historical data of the transaction system 135. This spoilage per day per item type data may be retained as a ninth labeled input feature to MLM 114. Trainer 113 also identifies a shrink reduction for the given item type based on items that sold in a markdown cycle before expiration and may retain this data as a tenth labeled input feature to MLM 114.
Trainer 113 may then obtain data relating to similar or complementary items to a given item of a given item type from existing analytics associated with analytic services of the retailer. These analytic services may leverage multiple data sources including the item catalogue system 135, transaction system 135, and/or inventory system 135. The trainer 113 may retain item identifiers for the similar or complementary items as eleventh labeled input features to MLM 114.
Next, trainer 113 obtains historical forecasted demand for the item. Forecasting systems 135 or services for the item may generate the historical forecasted demand data. This is retained as a twelfth labeled input feature.
The training data sets for each item may include the aforementioned twelve labeled input features across multiple intervals of time. In example embodiments, the expected output from the MLM 114 when provided an item's training data set may include a set of markdown parameters including a type of markdown (markdown level and/or markdown amount) and a quantity of units of the item to markdown. The markdown parameters may be optimized to reduce item shrink, increase item margins, and/or increase item sales. In some embodiments, when a threshold F1 score is attained by MLM 114—which may be reflected in decreased item shrink, increased item margins, and/or increase item sales—trainer 113 releases MLM 114 for production use by shrink optimizer 115. In an example embodiment, output from MLM 114 includes a predicted markdown level or markdown discount for a given item and a total quantity of the item to markdown on a given day within a store.
During operation of system 100, a store associate at the beginning of a business day for the store operates either terminal 120 or device 140 for purposes of identifying the items that are within a predefined time of expiring. This can be done in a variety of manners such as by utilizing an existing workflow associated with an existing markdown process that is enhanced to include shrink app 124 and/or shrink app 143. When a barcode for an item that is within the predefined time of expiring is scanned using a portable scanner or other user-operated device 140, the enhanced workflow sends the item code or item identifier to shrink optimizer 115.
Shrink optimizer 115 obtains input features for the item and passes the input features as input to MLM 114. The input features for the item may relate to some historical period of time (e.g., the prior day). MLM 114 returns an indication whether the given item associated with the item identifier should or should not be marked down, and if a markdown is provided, the MLM 114 also includes a markdown level for the item and a total quantity of units of the item to which to apply the markdown level. Shrink optimizer 115 may obtain the sales and demand forecast for the item from a same analytics system 135 or service used by the store. The sales and demand forecast calculated for the item may be provided as one of the input features to MLM 114.
The item codes or identifiers can be entered through a user-interface associated with app 124. For example, a user may scan the items that are within a predefined time of expiring and/or a store associate can walk the department of the store using a laptop 140, phone 140, or handheld scanner 140 to scan the barcodes on the items. Shrink applications 124 and 143 provide each item code to shrink optimizer 115. Shrink optimizer 115 maintains current input features for each of the store's perishable items, including sales and demand forecasts, and provides each item's input features as input to MLM 114. MLM 114 returns the indication as to whether the given item should or should not be marked down, and if the item is designated for markdown, the markdown level and the quantity of units for the item to which to apply the markdown. Shrink optimizer 115 may return this information back to user interfaces associated with shrink application 124 and shrink application 143.
The above-discussed approach reduces item shrink, increases item margin, and increases item sales for each item through the predictive capabilities of MLM 114. MLM 114 may be trained on data associated with the above-discussed input features, which as noted earlier, may be collected from a variety of systems 135 and services of a retailer. In example embodiments, MLM 114 not only weighs the negative impact of spoilage or waste but also a evaluates the tradeoff between the negative impact of spoilage and the negative impacts of markdowns on broader, long-term shopper behavior. Because MLM 114 is continuously re-trained through trainer 113 with actual results to create a feedback loop that improves the F1 values and optimizes the markdown process to minimize item spoilage and increase item margins and item sales, scenarios that result in negative shopping behavior influenced by the markdown process would manifest as decreased sales, decreased margins, increased spoilage, and/or decreased item sales forecasts, and thus, would be identifiable and addressable by the MLM 114. System 100 replaces static store markdown policies and discretionary managerial oversight with an objective and data-driven approach to item markdowns on a micro level and to sales and margins of a given store on a macro level. System 100 can be fully integrated and provided to retailers as a cloud-based service that enhances the retailers' existing systems 135, workflows, and services.
In an embodiment, an existing markdown service/user interface of a given retailer is enhanced with a workflow that calls shrink optimizer 115 for each scanned or entered item code. The user interface is further enhanced to provide the markdown indication, the markdown level, and the quantity of item units to be subjected to any markdown.
In an embodiment, shrink optimizer interface 134 includes a user interface associated with shrink optimizer 115 for purposes of requesting and generating reports on a daily basis for each store of a retailer. The reports may list an item identifier for each item of a given store to be marked down, its markdown level, and the quantity of units of the item to mark down. Shrink optimizer 115 may maintain up-to-date inventory levels for each item of the store using the retailer's store manager 133 and inventory system 135 and the store's transaction system 135, which obtains real-time sales data from transaction manager 123 of terminals 120. Shrink optimizer interface 134 and/or shrink optimizer 115 can further push the daily reports directly to terminals 120 and devices 140 for displaying to store associates via apps 124 and 143. In this way, the conventional approach associated with scanning item identifiers for items that are within a predefined time of expiring can be eliminated or supplemented with the daily store reports.
In an embodiment, shrink optimizer 115 maintains metrics for each item based on its current spoilage, sales, and margin, such that when spoilage is increasing and/or sales and margins decreasing, optimizer 115 may initiate a feedback re-training session through trainer 113. In example embodiments, trainer 113 maintains past input features associated with item data for a given store that were identified subsequent to a last training session with MLM 114 and initiates a new training session with MLM 114. A feedback loop is created to maintain the proper acceptable F1 values for MLM 114 to ensure optimization with respect to reducing spoilage and increasing item margins and sales.
In an embodiment, shrink optimizer 115 is provided as a software-as-a-service (SaaS) to systems 135 and existing enhanced services of a retailer. In this way, system 100 can be fully integrated into an existing retailer's markdown process via a call (e.g., an application programming interface (API) call) to shrink optimizer 115 from these existing systems 135 and services to established enhanced systems 135 and enhanced services for the retailer.
The above-referenced embodiments and other embodiments will now be discussed with reference to
In an embodiment, the device that executes item markdown optimizer is cloud 110. In an embodiment, the device that executes item markdown optimizer is server 110. In an embodiment, the device that executes item markdown optimizer is a retail server 130. In an embodiment, the item markdown optimizer is all of, or some combination of 113, 114, and/or 115. In an embodiment, the item markdown optimizer is provided to a retail server 130, retail terminal 120, retail system 135, retail service, and/or a user-operated device 140 as a SaaS.
At 210 (shown in
In an embodiment, at 211 (shown in
In another embodiment, at 212 (shown in
At 220 (shown in
In an embodiment, at 221 (shown in
At 230 (shown in
At 240 (shown in
In an embodiment, and at 241 (shown in
At 242 (shown in
At 243 (shown in
In an embodiment, at 250 (shown in
In an embodiment, at 260, the item markdown optimizer (210-240) executes as a SaaS to the retail systems 135 and/or to retail services. That is the systems 135 and services can be enhanced to provide the item identifier to the item markdown optimizer and receive the predictions for the item identifier generated by the MLM 114 through item markdown optimizer.
In an embodiment, the device that executes the item markdown manager is cloud 110. In an embodiment, the device that executes the item markdown manager is server 110. In an embodiment, the device that executes the item markdown manager is retail server 130. In an embodiment, the item markdown manager is provided to a retail server 130, a retail terminal 120, a retail system 135, a retail service, and/or a user-operated device 140 as a SaaS.
In an embodiment, the item markdown manager is all of, or some combination of 113, 114, 115, and/or method 200. The item markdown manager presents another and, in some ways, enhanced processing perspective from that which was discussed above with the method 200 and system 100.
At 310 (shown in
In an embodiment, at 311 (shown in
At 320, (shown in
In an embodiment, at 321 (shown in
In an embodiment, at 322 (shown in
At 330 (shown in
In an embodiment, at 331 (shown in
In an embodiment of 331, and at 332 (shown in
At 340 (shown in
At 350 (shown in
In an embodiment, at 360 (shown in
It should be appreciated that where software is described in a particular form (such as a component or module) this is merely to aid understanding and is not intended to limit how software that implements those functions may be architected or structured. For example, modules are illustrated as separate modules, but may be implemented as homogenous code, as individual components, some, but not all of these modules may be combined, or the functions may be implemented in software structured in any other convenient manner.
Furthermore, although the software modules are illustrated as executing on one piece of hardware, the software may be distributed over multiple processors or in any other convenient manner.
The above description is illustrative, and not restrictive. Many other embodiments will be apparent to those of skill in the art upon reviewing the above description. The scope of embodiments should therefore be determined with reference to the appended claims, along with the full scope of equivalents to which such claims are entitled.
In the foregoing description of the embodiments, various features are grouped together in a single embodiment for the purpose of streamlining the disclosure. This method of disclosure is not to be interpreted as reflecting that the claimed embodiments have more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive subject matter lies in less than all features of a single disclosed embodiment. Thus, the following claims are hereby incorporated into the Description of the Embodiments, with each claim standing on its own as a separate exemplary embodiment.
