THE BOSTON CONSULTING GROUP, INC.

A method implemented by a data processing system for calibrating a machine learning model, includes training a machine learning (ML) model, wherein the ML model comprises a plurality of nodes that are connected through edges and are aggregated into a plurality of layers comprising at least an input layer and an output layer, wherein each of the edges is configured to transmit a signal from one node to another node, and wherein an output of each of the plurality of nodes is computed based on inputs of the nodes in accordance with a plurality of weights; generating one or more A/B tests that are related to the area to be improved, determining one or more outcomes from executing the one or more A/B tests; and calibrating the ML model in accordance with the determined one or more outcomes.

Disclosed are techniques that generate a structural representation of a plurality of documents, the structural representation including a plurality of nodes and a plurality of edges, with the plurality of nodes being representations of the plurality of documents and the plurality of edges representing a feature in common between nodes of the plurality of nodes, with each node holding a vector of confidence values for weak models on a current optimization step and a weighted prediction for each of the weak models, generate a local ensemble model from the structural representation of the plurality of documents combined with the weighted prediction of the weak models, with the generated local ensemble model having a higher predictive power than any weak model individually, and generate a label for each node based on the local ensemble model.

A computer system comprising of an orchestration engine that allows for optimizing of investment allocation across plural levers, the orchestration engine configured to: receive input data that includes historic data regarding pricing and sales according to plural levers; prepare the received input data by converting the plural levers in the received input data into a common currency; calculate historical return on investment based on the common currency; simulate return on investment using the calculated historical return on investment and a machine learning model of a given campaign based on historical sales, campaign definitions, and investment metrics; optimize the simulated return on investment from the campaign with respect to one or more optimization goals and constraints; and output from the optimizer an optimized, campaign plan.

A system for inventory management based on transaction predictions can include memory storing transaction records. A processor can be configured to analyze a sequence of the transaction records using a neural network with a long short-term memory (LSTM) layer and at least one dense layer, wherein the LSTM layer extracts temporal purchase patterns and the dense layer generates probabilities of future purchases for multiple items and an estimated time until the next transaction. The processor can optimize the neural network by comparing generated probabilities and time estimates to actual subsequent transactions, computing cross-entropy loss for purchase probabilities and mean squared error for time estimates, and adjusting network weights accordingly. The process can apply the optimized neural network to recent transactions to predict item purchase probabilities and transaction timing, and automatically adjusts inventory levels based on the predictions to optimize inventory management.

In an example implementation, a cloud computing platform receives input data including historical data representing a plurality of first items and a plurality of stores associated with the plurality of first items, and generates a plurality of first data structures. Further, the system uses an artificial intelligence (AI) system to adaptively cluster the first data structures according to item metrics into a plurality of clusters, and classify a plurality of second items into the plurality of clusters based on the centroids. Further, the system determines, using a predictive computer model, a markdown plan for at least one of the first items or the second items associated with that cluster, and optimizes the markdown plan using mix integer programming.

Disclosed are techniques that generate a structural representation of a plurality of documents, the structural representation including a plurality of nodes and a plurality of edges, with the plurality of nodes being representations of the plurality of documents and the plurality of edges representing a feature in common between nodes of the plurality of nodes, with each node holding a vector of confidence values for weak models on a current optimization step and a weighted prediction for each of the weak models, generate a local ensemble model from the structural representation of the plurality of documents combined with the weighted prediction of the weak models, with the generated local ensemble model having a higher predictive power than any weak model individually, and generate a label for each node based on the local ensemble model.

Methods, systems, and computer storage media for providing a multilayer processing engine of a multilayer processing system. The multilayer processing engine supports an event layer, a metadata layer, and a multi-tier processing layer. The metadata layer can refer to a functional layer that operates via a sequential hierarchy of functional layers (i.e., event layer and multi-tier processing layer) to analyze incoming event streams and configure a downstream processing configuration. The metadata layer provides for dynamic metadata-based configuration of downstream processing of data associated with the event layer and the multi-tier processing layer. The multilayer processing system can be a data analytics system—operating via a serverless distributed computing system. The data analytics system implements the multilayer processing engine as a serverless data analytics management engine for processing high frequency data at scale based on dynamically-generated processing code—generated based on a downstream processing configuration—that supports automatically processing the data.

A computer system comprising of an orchestration engine that allows for optimizing of investment allocation across plural levers, the orchestration engine configured to: receive input data that includes historic data regarding pricing and sales according to plural levers; prepare the received input data by converting the plural levers in the received input data into a common currency; calculate historical return on investment based on the common currency; simulate return on investment using the calculated historical return on investment and a machine learning model of a given campaign based on historical sales, campaign definitions, and investment metrics; optimize the simulated return on investment from the campaign with respect to one or more optimization goals and constraints; and output from the optimizer an optimized, campaign plan.

Methods and systems for predicting are disclosed. Records for transactions can be stored, each record comprising an indication identifying items involved in a transaction and an indication of a time elapsed between the transaction and a previous transaction. The records for transactions can be analyzed to produce a probability that the items will be involved in a next transaction.

Methods and systems of using reinforcement learning to optimizing promotions. A promotion can be offered to a user using a reinforcement learning model with a sensitivity parameter, the reinforcement module estimating a time period during which the user will respond to the first information. The user's reaction to the promotion can be observed. The reinforcement learning model can be adapted based on the user's reaction.