SUPPLY CHAIN ​​MANAGEMENT BASED ON MACHINE LEARNING

Description

1 / 25 “MACHINE LEARNING-BASED SUPPLY CHAIN ​​MANAGEMENT” Technical Field

[001] These teachings relate, in general, to the use of artificial intelligence and, more particularly, to the use of machine learning models. Background

[002] Supply chain management sometimes includes attempting to understand (and consequently plan for) future needs and the corresponding supplies to meet them. Key performance indicators, of one type or another, are sometimes used to facilitate and / or measure the functioning of a logistics organization.

[003] Supply chain managers, with or without the aid of computerized support, produce metrics to describe projected future needs and expected corresponding available supplies. For example, when forecasting three weeks for a particular product, a manager might determine that demand will be X for that product in that week, while available inventory will be XY (i.e., a quantity slightly less than the projected demand). While this information is typically useful, the applicant determined that such information may not be entirely reliable and that having an understanding of how reliable (or unreliable) such conclusions might be can be a useful consideration when managing a corresponding supply chain.

[004] Managing a supply chain may involve making changes to the supply chain so that physical elements and aspects of that supply chain can operate more efficiently and physically meet the needs. Petition 870250091752, dated 08 / 10 / 2025, page 12 / 47 2 / 25 future and inventory deliveries. The ability to predict the performance of a supply chain means it can be managed more effectively. Brief Description of the Drawings

[005] The above needs are, at least partially, met by providing the machine learning-based supply chain performance predictions described in the detailed description below, particularly when studied in conjunction with the designs, where:

[006] Figure 1 comprises a timeline as configured according to various modalities of these teachings;

[007] Figure 2 comprises a schematic representation as configured according to various modalities of these teachings;

[008] Figure 3 comprises an index as configured according to various modalities of these teachings;

[009] Figure 4 comprises a timeline as configured according to various modalities of these teachings;

[010] Figure 5 comprises a graph as configured according to various modalities of these teachings;

[011] Figure 6 comprises a timeline and graph as configured according to various modalities of these teachings;

[012] Figure 7 comprises a schematic representation as configured according to various embodiments of the invention; and

[013] Figure 8 comprises a block diagram as configured according to various modalities of these teachings.

[014] The elements in the figures are illustrated for simplicity and clarity and have not necessarily been drawn to scale. For example, the dimensions and / or the Petition 870250091752, dated 08 / 10 / 2025, p. 13 / 47 3 / 25 The relative positioning of some of the elements in the figures may be exaggerated in relation to other elements to help improve the understanding of various embodiments of the present teachings. Furthermore, common but well-understood elements that are useful or necessary in a commercially viable embodiment are often not represented to facilitate a less obstructed view of these various embodiments of the present teachings. Certain actions and / or steps may be described or represented in a specific order of occurrence, while those skilled in the art will understand that such specificity with respect to sequence is not actually necessary. The terms and expressions used in the present invention have the common technical meaning as assigned to such terms and expressions by persons skilled in the art, as set forth above, except where different specific meanings have been established in the present invention.The access word “or,” when used in the present invention, should be interpreted as having a disjunctive construction rather than a conjunctive construction, unless specifically indicated otherwise. Detailed Description

[015] In general, these diverse modalities will accommodate access to historical information resources where each comprises a supply chain performance indicator (such as an order fulfillment rate performance indicator) for each of a plurality of different time windows, weighting at least some of the historical information resources differently for at least some of the historical information resources according to at least one first criterion to provide a training corpus, and then training at least one machine learning model using the corpus of Petition 870250091752, dated 08 / 10 / 2025, page 14 / 47 4 / 25 training to generate a machine learning model (or models) configured to predict supply chain performance.

[016] By one approach, the first criterion mentioned above comprises temporal proximity of each of the different time windows to a target time window (such as, but not limited to, a future time window). By one approach, the aforementioned weighting of at least some of the historical information resources differently for at least some of the historical information resources according to at least one first criterion comprises, at least in part, weighting at least one of the different time windows that is closer to the future time window in a higher position than another of the different time windows that is further away from the future time window.

[017] By one approach, the first criterion mentioned above comprises a stock weeks parameter corresponding to each of the different time windows. In such a case, weighting at least some of the historical information resources differently for at least some of the historical information resources according to at least one first criterion may comprise, at least in part, weighting at least one of the different time windows having a stock weeks parameter that is sufficiently similar to a stock weeks parameter for a larger future time window than another of the different time windows having a stock weeks parameter that is less similar to the stock weeks parameter for the future time window.

[018] Consequently, by one approach, these teachings will accommodate a control circuit that is configured as at least one model of Petition 870250091752, dated 08 / 10 / 2025, page 15 / 47 5 / 25 Supply chain performance prediction machine learning that was trained, at least in part, with a training corpus formed by accessing historical information resources where each comprises a supply chain performance indicator for each of a plurality of different time windows and weighting at least some of the historical information resources differently for at least some of the historical information resources according to at least one first criterion.

[019] If desired, such a supply chain performance prediction machine learning model can be further configured to calculate at least one threshold of the supply chain performance indicator (e.g., by analyzing, at least in part, historical relationships between an order fulfillment rate metric and a supply weeks metric and then calculating at least one threshold of the supply chain performance indicator as a threshold of the supply weeks metric that identifies a favorable future order fulfillment rate metric).

[020] One approach provides a computer-implemented method for managing a supply chain, comprising: Access historical information resources, each comprising a supply chain performance indicator for a plurality of different time windows; to weight at least some of the historical information resources differently for at least some of the historical information resources according to at least one first criterion to provide a training corpus; Petition 870250091752, dated 08 / 10 / 2025, page 16 / 47 6 / 25 train a machine learning model using the training corpus to generate a machine learning model configured to predict supply chain performance; Predict supply chain performance using a machine learning model; Varying the supply chain planning data based on predicted supply chain performance and processing inventory within the supply chain (e.g., moving inventory from one location to another, picking components or parts, and / or manufacturing products) allows for more accurate prediction of supply chain performance. If this predicted performance falls short of what is needed (e.g., below a predetermined threshold or thresholds), then the parameters within the planning data used to implement the supply chain can be varied so that the supply chain can operate more effectively. This can lead to more reliable, faster inventory delivery with fewer delivery intervals, for example. This can save fuel and other resources.

[021] By one approach, then, these teachings will additionally accommodate the insertion of information into a supply chain performance prediction machine learning model that has been trained, at least in part, with a training corpus formed by accessing historical information features where each comprises a supply chain performance indicator for each of a plurality of different time windows and weighting at least some of the historical information features differently for at least some of the historical information features according to by Petition 870250091752, dated 08 / 10 / 2025, page 17 / 47 7 / 25 minus a first criterion and then display, from the supply chain performance prediction machine learning model, a forecast of supply chain performance over a future time window.

[022] By one approach, such a supply chain performance prediction machine learning model can be configured to calculate at least one threshold of the supply chain performance indicator.

[023] By another approach, instead of or in combination with the above, such a supply chain performance forecast may comprise, at least in part, a supply weeks metric that results in a favorable order fulfillment rate metric for the future time window.

[024] These and other benefits may become clearer after a thorough review and study of the detailed description that follows. Referring now to the drawings and, in particular, to Figure 1, an example of an illustrative temporal context will be presented first.

[025] As discussed in the present invention, these teachings relate to forecasting supply chain circumstances / performance during different time windows. For the purposes of a useful illustrative example, and without intending to suggest any limitations as to the duration or periodicity of these windows (or even whether they all have similar durations), the following description assumes that the time windows of interest each have a duration of one week (i.e., seven consecutive days). Starting from “now” 101 (i.e., a current time period), future time windows 102 (i.e., in this example, “weeks”) can be consecutively and serially denoted as week 0, week 1, and so on up to week N (where “N” is an integer) or beyond. For many application configurations, this will be Petition 870250091752, dated 08 / 10 / 2025, p. 18 / 47 8 / 25 is useful and beneficial to predict 52 of these weeks, but this number can easily be varied as desired. And, looking back from the “now” 101, the previous 103 time windows (again, in this example, “weeks”) are consecutively denoted as delay -0, delay -1, and so on up to delay -N or beyond, as desired. (“Delay” is described in more detail below, where appropriate).

[026] In general, these teachings concern making supply chain forecasts for each of a plurality of future time windows (in this case, weeks) and assessing the level of confidence that can be placed in such forecasts. This confidence indication can then be leveraged in a variety of ways.

[027] Referring now to Figure 2, an approach to estimating / predicting a key supply chain performance indicator will be described. In this example, and for illustrative purposes, the key performance indicator is the order fulfillment rate (CFR). Order fulfillment rate represents the difference between the quantity of product a customer ordered and the quantity actually shipped. For example, if a customer orders 100 boxes of an item, but only 90 boxes are actually shipped, the order fulfillment rate for that item in this situation is 90%. If it is possible to accurately predict the future risk of CFR (i.e., that an undesirable mismatch between order and fulfillment is anticipated), a proactive organization will be better able to prevent the risk from materializing.

[028] The approach presented in Figure 2 includes access data 201. This 201 data may comprise previous planning data for each week called a snapshot and each week called a fiscal week during some prior period of interest (such as the last year). A snapshot week is a week during which planners developed a corresponding plan in relation to Petition 870250091752, dated 08 / 10 / 2025, page 19 / 47 9 / 25 advance order fulfillment, and fiscal weeks refer to future weeks (i.e., in the future, compared to the corresponding snapshot week) to which the aforementioned plan applies.

[029] The difference between a fiscal week and a corresponding week (such as, but not limited to, an instant week) is called a lag.

[030] In this example, it is assumed that planners develop / adjust plans weekly. These plans are formed based on demand, production, and inventory. As will be described in detail, this approach compares the planner's previous projections with the numbers actually achieved (i.e., results) and facilitates the correction of current plans.

[031] Through one approach, and as illustrated, some or all of the data 201 can be subjected to preprocessing 202. This preprocessing 202 can vary according to the needs and / or opportunities presented by a given application configuration. In general, the data 201 can be cleaned to, for example, remove extraneous or duplicate content, correct syntax and / or formatting, and so on.

[032] One approach, this 202 preprocessing can also encompass the generation of leading indicators. Leading indicators capture inertia in the corresponding supply chain. For example, if newly received stock is not available for shipment in the same week and can only be shipped a week after arrival, stock from the previous week might be a better indicator than stock from the current week. During 202 preprocessing, this approach will accommodate the creation of features, for example, from the previous four weeks. These features can allow the downstream machine learning model to learn the upward or downward trend leading to the current week. As Petition 870250091752, dated 08 / 10 / 2025, page 20 / 47 10 / 25 As an illustrative example in this regard, consider weeks of supply (WOS) (described in more detail in the present invention), demand, production, and fixed interest rate (CFR) as four variables of interest. Useful corresponding machine learning features could be: WOSwk-4, WOSwk-3, WOSwk-2, WOSwk-1, WOSwk, Demandwk-4, Demandwk-3, Demandwk-2, DemandWk-1, Demandwk, Productionwk-4, Productionwk-3, Productionwk-2, Productionwk-1, Productionwk, CFRwk-4, CFRwk-3, CFRwk-2, CFRwk-1

[033] Segregated data subsets 203 for a single product (as corresponds, for example, to a single stock keeping unit (SKU) number) for each of the relevant delay windows can then be drawn and formed from the previous data 201 (optionally pre-processed). In particular, a data subset 203 can be formed for each of the following: delay-0, delay-1, delay-2 and so on, sequentially up to delay-N.

[034] Training datasets (i.e., a training corpus for a corresponding machine learning model) 204 are then formed using the data subsets 203 mentioned earlier. In general, in this illustrative example, each training dataset 204 is formed from two or three of the data subsets 203. For example, the training dataset named “Training Data Delay-1” is formed by compiling the data subsets for Delay-0, Delay-1, and Delay-2, and the training dataset named “Training Data Delay-3” is formed by compiling the data subsets for Delay-2, Delay-3, and Delay-4. Figure 2 illustrates that, in this example, the sequentially first and last training datasets are formed using only two of the data subsets 203. Petition 870250091752, dated 08 / 10 / 2025, page 21 / 47 11 / 25

[035] As indicated by reference number 205, this approach can then provide data prioritization. In general, this activity serves to assign higher priority to certain data points (such as, for example, more recent data points in time compared to data points that are further away in time).

[036] And, as indicated by reference number 206, this approach can provide resource prioritization. In this illustrative example, this prioritization uses a weeks of supply (WOS) metric. The weeks of supply metric measures the relationship between inventory and demand, in particular, how many weeks of demand a given inventory will cover.

[037] For example, if a current week ends with 100 boxes in stock and the demand for the following week is 40 boxes, and the demand for the following week is 60 boxes, the WOS would be 2. If, however, the future demand for the same 100 boxes in stock is 20 boxes for each of the following consecutive weeks, those same 100 boxes would equate to a WOS value of 5. (It is worth noting that the applicant determined that leveraging the relationship between historical WOS and historical CFR when making future CFR forecasts creates a more sophisticated and accurate model than if we simply use the relationship between historical CFR and stock).

[038] A WOS threshold, to achieve, for example, an order fulfillment rate of 95%, can be calculated based on historical data. By one approach, a WOS closer to this threshold should also be closer to an order fulfillment rate of 95%, and vice versa. It is possible to calculate the order fulfillment rate for future weeks based on the proximity to this threshold value. An order fulfillment rate calculated with this method shows greater accuracy when Petition 870250091752, dated 08 / 10 / 2025, page 22 / 47 12 / 25 the WOS is close to the threshold value and, conversely, the accuracy decreases as the WOS moves away from this threshold value.

[039] As explained above, data from more recent weeks can be given greater weight than data from more distant weeks. Threshold values ​​such as 95%, 80%, and 65% can be calculated. Three order fulfillment rates can be calculated based on the proximity to these threshold values. A weighting can then be assigned to these order fulfillment rates based on their relative proximity to a corresponding threshold value. For example, if a predicted order fulfillment rate is 72% based on the 95% threshold, this order fulfillment rate appears unreliable, and the weighting assigned to this order fulfillment rate can be zero or close to zero. However, when the order fulfillment rate is equal to 81% based on the 80% threshold, the results appear more reliable, and the order fulfillment rate can be given greater weighting.

[040] In the reference number, the (optionally prioritized) 204 training datasets for each delay window are then used to train a machine learning model. To clarify, and as mentioned earlier, this occurs product by product, so there may be a separate machine learning model trained for each product in a total inventory comprising, for example, hundreds of thousands of individual products.

[041] Experts in the field understand that machine learning comprises a branch of artificial intelligence. Machine learning typically employs learning algorithms such as Bayesian networks, decision trees, nearest neighbor approaches, and so on, and the process can operate in a supervised or unsupervised manner, as desired. Learning Petition 870250091752, dated 08 / 10 / 2025, page 23 / 47 Deep learning (also referred to as hierarchical learning, deep neural learning, or deep structured learning) is a subset of machine learning that employs networks capable of learning (often supervised, where the data consists of pairs (such as input_data and labels) and the goal is to learn a mapping between the input_data and the associated labels) from data that may, at least initially, be unstructured and / or unlabeled. Deep learning architectures include deep neural networks, deep belief networks, recurrent neural networks, and convolutional neural networks. Many machine learning algorithms build a so-called "model" based on sample data, known as training data or training corpus, in order to make predictions or decisions without being explicitly programmed to do so.

[042] In the present example, it will be assumed that the machine learning model comprises an N-regression type machine learning model. An N-regression type machine learning model is a type of regression model that involves predicting a numerical target variable based on multiple input features. The term N refers to the number of input features used to make the prediction. In N-regression, the model learns a mathematical relationship between the input features and the target variable by minimizing the difference between the predicted and actual values ​​of the target variable. The goal is to develop a model that can accurately predict the target variable for new instances based on their input features.

[043] As indicated by reference number 208, each trained machine learning model generates a predicted order fulfillment rate for a corresponding future week (such as week 0, week 1, week 2, and so on). The above approach can be performed with frequency or infrequency (and with Petition 870250091752, dated 08 / 10 / 2025, page 24 / 47 14 / 25 the periodicity (such as weekly) or with the randomization that is appropriate for a given application scenario.

[044] Therefore, it will be understood and appreciated that these teachings provide an effective way to utilize and significantly leverage the results of previous forecasts to better evaluate / establish future forecasts.

[045] A more detailed example in this regard will now be presented. It is understood that the details of this example are intended to illustrate and not to suggest any limitations with respect to the practice of these teachings.

[046] With reference to Figure 3, three geometric icons serve to indicate a past time window for which a corresponding forecast for the order fulfillment rate was accurate to a given percentage level. Consequently, in the following example, a square 301 indicates a resulting actual order fulfillment rate with an accuracy of 97% or more. A circle 302 indicates a resulting actual order fulfillment rate with an accuracy within a range of 90% to 97%. And a triangle 303 indicates a resulting actual order fulfillment rate with an accuracy of less than 90%.

[047] Figure 4 presents a series of time periods, each beginning with a specific instant 401 week. For example, the first 402 row begins with an instant 401 week for the 4th week of 2021, while the third 403 row begins with an instant 401 week for the 12th week of 2021. In each row, and extending to the right from the corresponding instant 401 week, are the lag weeks. One of the geometric icons described above serves, in this example, to indicate the accuracy of the predicted order fulfillment rate for each specific week. For example, in the first 402 pair, the week corresponding to Lag=6 (i.e., the 10th week of 2021) has an icon. Petition 870250091752, dated 08 / 10 / 2025, page 25 / 47 15 / 25 geometric square representing a result with precision greater than or equal to the 97th percentile.

[048] In this illustrative example, the current focus is on the weeks corresponding to Delay=4 (collectively represented by the bounding box indicated by the reference digit 405). Figure 5 graphically represents the historical order fulfillment rates corresponding to the Delay=4 weeks in a 500 graph defined by a first variable for the X-axis and a second variable for the Y-axis. Only two variables / two dimensions are shown in the present invention for the sake of clarity and simplicity. In a typical application environment, considerably more variables are considered by the machine learning model, which can also be used.

[049] Figure 6 presents a timeline 601 that begins with an instantaneous week 401, which is the current week (in this illustrative example, the 45th week of 2021; it can be noted that the timelines represented in Figure 4 conclude with the 44th week of the same year, immediately preceding the current week. Consequently, in this example, the history under consideration begins with the 4th week of the year 2021 and ends with the 44th week of the same year, immediately preceding the current week, which serves as the instantaneous week to predict future order fulfillment rates on a forward-looking weekly basis.

[050] As noted above, this example focuses on a history specifically related to Delay=4. This is because, in this example, the objective is to establish a forecast of the order fulfillment rate for a future week that is 4 weeks beyond the current week (this 4th week being denoted by the reference number 602). As illustrated by graph 500 (presented above in Petition 870250091752, dated 08 / 10 / 2025, page 26 / 47 16 / 25 Figure 5), the objective is to determine where the predicted order fulfillment rate can be placed in terms of probable accuracy.

[051] Figure 7 serves to graphically illustrate the estimation of a confidence level for a predicted order fulfillment rate using unsupervised learning, as described in the present invention. In this example, the features for the data points used can be things like demand, production, inventory, transportation, and so on. In this example, one can divide the historical weeks (see Figure 4) into n clusters based on the similarity of the features. The probability of service capacity disruption for a future week can be calculated based on which cluster that future week resembles. (Note that, in this example, Delay is not used as a feature).

[052] In Figure 7, each geometric shape represents a historical week, along with an indication of the accuracy of the corresponding order fulfillment rate forecast for that week. A historical week may consist of several input features and one output (i.e., order fulfillment rate). Weeks are divided into n clusters (usually denoted by reference number 701) using an unsupervised learning method as described in the present invention. The cluster membership of a future week is calculated, and the probability of disruption can then be calculated based on how many times the service capacity was interrupted (the disruptions being represented by the triangle-shaped geometric icons) in any given cluster.

[053] For example, if the next week (represented as a star-shaped icon, indicated by reference number 702) is within cluster 2 (indicated by reference number 703), it can be safely predicted that there is a 100% chance of service failure. If, however, the next week (represented as an icon in Petition 870250091752, dated 08 / 10 / 2025, page 27 / 47 If the 17 / 25 X-shape, indicated by reference number 704), is within cluster 1 (indicated by reference number 705), then it can be said that there is only a 22% chance of service failure.

[054] Several illustrative examples of an 800 enabling device to support the foregoing teachings will now be described with reference to Figure 8.

[055] For illustrative purposes, it will be assumed in the present invention that one or more selectable control circuits 801 perform the actions, steps and / or functions described in the present invention. Being a circuit, the control circuit 801 therefore comprises a structure that includes at least one (and typically many) electrically conductive paths (such as paths composed of a conductive metal, such as copper or silver) that carry electricity in an orderly fashion, these paths also typically including corresponding electrical components (both passive (such as resistors and capacitors) and active (such as any of a variety of semiconductor-based devices), as appropriate) to enable the circuit to effect the control aspect of these teachings.

[056] Such an 801 control circuit may comprise a fixed-purpose, fixed-wire hardware platform (including, but not limited to, an application-specific integrated circuit (ASIC) (which is an integrated circuit customized by design for a specific use, rather than intended for general use), a field-programmable gate array (FPGA), and the like) or may comprise a partially or fully programmable hardware platform (including, but not limited to, microcontrollers, microprocessors, and the like). These architectural options for such structures are well known and understood in the art and require no further description in the present invention. This 801 control circuit is configured (e.g., using corresponding programming, as will be well understood by those versed in Petition 870250091752, dated 08 / 10 / 2025, page 28 / 47 18 / 25 technique) to perform one or more of the steps, actions and / or functions described in the present invention.

[057] In this example, the control circuit 801 is operatively coupled to a memory 802. This memory 802 may be an integral part of the control circuit 801 or may be physically discrete (wholly or partially) from the control circuit 801, as desired. This memory 802 may also be local with respect to the control circuit 801 (where, for example, both share a common circuit board, chassis, power supply and / or enclosure) or may be partially or wholly remote with respect to the control circuit 801 (where, for example, the memory 802 is physically located in another facility, metropolitan area or even country compared to the control circuit 801). It should also be understood that this memory 802 may comprise a plurality of physically discrete memories that, together, store the pertinent information corresponding to these teachings.

[058] In addition to the historical data mentioned above, this memory 802 can serve, for example, to store non-transiently computer instructions and machine learning models which, when executed by the control circuit 801, cause the control circuit 801 to behave as described in the present invention. (As used in the present invention, this reference to non-transiently will be understood as referring to a non-ephemeral state for the stored content (and therefore excludes when the stored content constitutes merely signals or waves) rather than the volatility of the storage medium itself and therefore includes both non-volatile memory (such as read-only memory (ROM)) and volatile memory (such as dynamic random-access memory (DRAM)).

[059] In this example, the control circuit 801 is also operatively coupled to a user interface 803. This user interface 803 can Petition 870250091752, dated 08 / 10 / 2025, page 29 / 47 19 / 25 comprise any of a variety of user input mechanisms (such as, but not limited to, keyboards and numeric keypads, cursor control devices, touch screens, speech recognition interfaces, gesture recognition interfaces, and so forth) and / or user output mechanisms (such as, but not limited to, visual displays, audio transducers, printers, and so forth) to facilitate the receipt of information and / or instructions from a user and / or the provision of information to a user. Thus configured, the output of information by the described process can be presented to a user to facilitate, for example, the evaluation of the likely accuracy of a given forecast of the order fulfillment rate for a given future time period.

[060] By means of an optional approach, the control circuit 801 can also be operationally coupled to a network interface 804. Thus configured, the control circuit 801 can communicate with other elements (both within the device 800 and externally) by means of the network interface 804. In particular, the control circuit 801 can communicate by means of one or more intervening networks 805 (such as, but not limited to, the well-known Internet) with one or more remote resources 806 (such as, for example, servers that provide some or all of the aforementioned historical data) and / or one or more remote user interfaces 807 (to thus allow, for example, the communication of the prediction results described in the present invention to one or more remote users). Network interfaces, including wireless and wireless platforms, are well understood in the art and do not require any specific elaboration in the present invention.

[061] Experts in the technique will recognize that a wide variety of modifications, alterations, and combinations can be made with respect to the modalities described above without departing from the scope of these teachings. For example, these Petition 870250091752, dated 08 / 10 / 2025, page 30 / 47 20 / 25 teachings can also be beneficially employed with other key performance indicators. Consequently, such modifications, alterations, and combinations should be seen as falling within the scope of the inventive concept.

[062] The following numbered clauses provide additional illustrative examples: 1. A computer-implemented method for managing a supply chain, comprising: Access historical information resources, each comprising a supply chain performance indicator for a plurality of different time windows; to weight at least some of the historical information resources differently for at least some of the historical information resources according to at least one first criterion to provide a training corpus; Train a machine learning model using the training corpus to generate a machine learning model configured to predict supply chain performance; Predict supply chain performance using a machine learning model; To vary the supply chain planning data of a supply chain based on the predicted supply chain performance; and to process inventory within the supply chain. 2. The computer-implemented method of clause 1, in which the supply chain performance indicator comprises an order fulfillment rate performance indicator. Petition 870250091752, dated 08 / 10 / 2025, page 31 / 47 21 / 25 3. The computer-implemented method of clause 1 or clause 2, where the first criterion comprises the temporal proximity of each of the different time windows to a target time window. 4. The method implemented by computer in accordance with any of the preceding clauses, wherein the target time window comprises a future time window. 5. The computer-implemented method of clause 4, in which weighting at least some of the historical information resources differently for at least some of the historical information resources according to at least one first criterion comprises, at least in part, weighting at least one of the different time windows that is closer to the future time window in a higher position than another of the different time windows that is further away from the future time window. 6. The method implemented by computer according to any previous clause, wherein the first criterion comprises a parameter of weeks of stock corresponding to each of the different time windows. 7. The computer-implemented method of clause 6, wherein weighting at least some of the historical information resources differently for at least some of the historical information resources according to at least one first criterion comprises, at least in part, weighting at least one of the different time windows having a stock weeks parameter that is sufficiently similar to a stock weeks parameter for a larger future time window than another of the different time windows having a stock weeks parameter that is less similar to the stock weeks parameter for the future time window. 8. An apparatus comprising: Petition 870250091752, dated 08 / 10 / 2025, page 32 / 47 22 / 25 a memory; A memory-operably coupled control circuit configured as a supply chain performance prediction machine learning model that has been trained, at least in part, on a training corpus formed by accessing historical information features where each comprises a supply chain performance indicator for each of a plurality of different time windows and weighting at least some of the historical information features differently according to at least one first criterion, wherein the control circuit is additionally configured to: Predict supply chain performance using a machine learning model; To vary the supply chain planning data of a supply chain based on the predicted supply chain performance; and to process inventory within the supply chain. 9. The apparatus of clause 8, in which the supply chain performance indicator comprises an order fulfillment rate performance indicator. 10. The apparatus of clause 8 or clause 9, in which the first criterion comprises the temporal proximity of each of the different time windows to a destination time window. 11. The apparatus of clause 10, in which the target time window comprises a future time window. 12. The apparatus of clause 11, in which the weighting of at least some of the historical information resources is different for at least some of the Petition 870250091752, dated 08 / 10 / 2025, pp. 33 / 47 23 / 25 historical information resources according to at least one first criterion comprise, at least in part, weighting at least one of the different time windows that is closer to the future time window in a higher position than another of the different time windows that is further away from the future time window. 13. The device meets any of clauses 8 to 12, where the first criterion comprises a parameter of weeks of stock corresponding to each of the different time windows. 14. The apparatus of clause 13, whereby weighting at least some of the historical information resources differently for at least some of the historical information resources according to at least one first criterion comprises, at least in part, weighting at least one of the different time windows having a stock weeks parameter that is sufficiently similar to a stock weeks parameter for a larger future time window than another of the different time windows having a stock weeks parameter that is less similar to the stock weeks parameter for the future time window. 15. The device in accordance with any of clauses 8 to 14, in which the supply chain performance prediction machine learning model is configured to calculate at least one threshold of the supply chain performance indicator. 16. The apparatus of clause 15, in which the supply chain performance prediction machine learning model is configured to calculate at least one threshold of the supply chain performance indicator by analyzing, at least in part, historical relationships between an order fulfillment rate metric and a supply weeks metric. Petition 870250091752, dated 08 / 10 / 2025, pp. 34 / 47 24 / 25 17. The apparatus of clause 16, in which the supply chain performance prediction machine learning model is configured to calculate at least one threshold of the supply chain performance indicator as a threshold of the supply weeks metric that identifies a favorable future order fulfillment rate metric. 18. A method comprising: Insert information into a supply chain performance prediction machine learning model that has been trained, at least in part, with a training corpus formed by accessing historical information features where each comprises a supply chain performance indicator for each of a plurality of different time windows and weight at least some of the historical information features differently for at least some of the historical information features according to at least one first criterion; To display, using a machine learning model for supply chain performance prediction, a forecast of supply chain performance over a future time window; To vary the supply chain planning data of a supply chain based on forecasting supply chain performance; and to process inventory within the supply chain. 19. The method in clause 18, where the supply chain performance prediction machine learning model is configured to calculate at least one threshold of the supply chain performance indicator. 20. The Clause 19 method, in which the forecast of supply chain performance comprises, at least in part, a supply weeks metric. Petition 870250091752, dated 08 / 10 / 2025, pp. 35 / 47 25 / 25, which results in a favorable order fulfillment rate metric for the future time window. Petition 870250091752, dated 08 / 10 / 2025, pp. 36 / 47

Claims

1 / 5 CLAIMS 1. A computer-implemented method for training a machine learning model to predict supply chain performance, CHARACTERIZED in that it comprises: accessing historical information resources where each comprises a supply chain performance indicator for each of a plurality of different time windows; weighting at least some of the historical information resources differently for at least some of the historical information resources according to at least one first criterion to provide a training corpus; training a machine learning model using the training corpus to generate a machine learning model configured to predict supply chain performance.

2. A computer-implemented method according to claim 1, CHARACTERIZED in that the supply chain performance indicator comprises an order fulfillment rate performance indicator.

3. A computer-implemented method according to claim 1, characterized in that the first criterion comprises the temporal proximity of each of the different time windows to a target time window.

4. A computer-implemented method according to claim 3, characterized in that the target time window comprises a future time window.

5. A computer-implemented method according to claim 4, CHARACTERIZED in that weighting at least some of the historical information features differently for at least some of the historical information features according to at least one first criterion comprises, at least in part, weighting at least one of the different time windows that is closer to the future time window in a higher position than another of the different time windows that is further away from the future time window.

6. A computer-implemented method according to claim 1, characterized in that the first criterion comprises a parameter of weeks of inventory corresponding to each of the different time windows.

7. A computer-implemented method according to claim 6, CHARACTERIZED in that weighting at least some of the historical information features differently for at least some of the historical information features according to at least one first criterion comprises, at least in part, weighting at least one of the different time windows having a stock weeks parameter that is sufficiently similar to a stock weeks parameter for a larger future time window than another of the different time windows having a stock weeks parameter that is less similar to the stock weeks parameter for the future time window.

8. Device, CHARACTERIZED by the fact that it comprises: a memory; a control circuit operably coupled to the memory and configured as a supply chain performance prediction machine learning model that has been trained, at least in part, with a training corpus formed by accessing historical information resources, each comprising a supply chain performance indicator for each of a plurality of different time windows and weighting at least some of the historical information resources differently for at least some of the historical information resources according to at least one first criterion.

9. Apparatus, according to claim 8, CHARACTERIZED in that the supply chain performance indicator comprises an order fulfillment rate performance indicator.

10. Apparatus, according to claim 8, CHARACTERIZED in that the first criterion comprises temporal proximity of each of the different time windows to a target time window.

11. Apparatus, according to claim 10, CHARACTERIZED in that the target time window comprises a future time window.

12. Apparatus, according to claim 11, CHARACTERIZED in that the weighting of at least some of the historical information features differently for at least some of the historical information features according to at least one first criterion comprises, at least in part, weighting at least one of the different time windows that is closer to the future time window in a higher position than another of the different time windows that is further away from the future time window.

13. Apparatus, according to claim 8, CHARACTERIZED in that the first criterion comprises a parameter of weeks of stock corresponding to each of the different time windows.

14. Apparatus, according to claim 13, CHARACTERIZED in that weighting at least some of the historical information features differently for at least some of the historical information features according to at least one first criterion comprises, at least in part, weighting at least one of the different time windows having a stock weeks parameter that is sufficiently similar to a stock weeks parameter for a larger future time window than another of the different time windows having a stock weeks parameter that is less similar to the stock weeks parameter for the future time window.

15. Device according to claim 8, CHARACTERIZED in that the supply chain performance prediction machine learning model is configured to calculate at least one threshold of the supply chain performance indicator.

16. Apparatus, according to claim 15, CHARACTERIZED in that the supply chain performance prediction machine learning model is configured to calculate at least one threshold of the supply chain performance indicator by analyzing, at least in part, historical relationships between an order fulfillment rate metric and a supply weeks metric.

17. Device, according to claim 16, CHARACTERIZED in that the supply chain performance prediction machine learning model is configured to calculate at least one threshold of the supply chain performance indicator as a threshold of the supply weeks metric that identifies a favorable future order fulfillment rate metric.

18. Method, CHARACTERIZED by the fact that it comprises: inputting information into a supply chain performance prediction machine learning model that has been trained, at least in part, with a training corpus formed by accessing historical information features, each comprising a supply chain performance indicator for each of a plurality of different time windows and weighting at least some of the historical information features differently for at least some of the historical information features according to at least one first criterion; displaying, from the supply chain performance prediction machine learning model, a forecast of supply chain performance with respect to a future time window.

19. Method, according to claim 18, CHARACTERIZED in that the supply chain performance prediction machine learning model is configured to calculate at least one threshold of the supply chain performance indicator.

20. Method, according to claim 19, CHARACTERIZED in that the forecast of supply chain performance comprises, at least in part, a supply weeks metric that results in a favorable order fulfillment rate metric for the future time window. Petition 870250091752, dated 10 / 08 / 2025, pp. 41 / 47

Images