Toast, Inc.

A system and method are provided for dynamically updating a digital menu using deep-learning-based preparation-time prediction. A first deep learning neural network generates item-level embeddings for menu items based on historical preparation time records. Actual and estimated item-level preparation time vectors are generated using cosine-similarity weighting across subsets of the historical data. A second deep learning neural network is trained using the item-level vectors, ground-truth preparation times, and normalized non-categorical metadata processed through dense vector layers and concatenation. The trained network is executed to generate predicted item-level preparation times for menu items currently available for ordering. An updated digital menu including the predicted item-level preparation times is generated and transmitted to a client device, and the digital menu is automatically updated in real time on the client device in response to changes in the predicted preparation times.

A computer-implemented method for displaying a digital menu on a point-of-sale device, the method including, on a touchscreen display, displaying selectable icons as a single virtual page, where the selectable icons correspond to menu tier elements in menu tiers according to a menu tier hierarchy; via a processor, periodically analyzing historical fulfilled orders corresponding to the digital menu to determine a ranking of the menu tier elements from most ordered to least ordered; and, responsive to the ranking, rearranging order of the first selectable icons on the single virtual page so that the order of displayed first selectable icons is from most ordered to least ordered.

A computer-implemented method for displaying a digital menu on a point-of-sale device, the method including: on a touchscreen display, displaying selectable icons as a single virtual page, where: the selectable icons correspond to menu tier elements in menu tiers according to a menu tier hierarchy; and the touchscreen display includes scrolling mechanisms to display one or more of the selectable icons that are not initially visible on the single virtual page; selecting one of the selectable icons in a menu tier for an order, where the menu tier is not a highest-level tier in the menu tier hierarchy; and responsive to selection of the one of the selectable icons, modifying the single virtual page by replacing one or more of the selectable icons in higher-level and lower-level menu tiers relative to the menu tier with new replacement icons that correspond to the one of the selectable icons in the menu tier hierarchy.

A method for predicting preparation times includes: retrieving set of item-level records from a database for preparation menu items for all subscriber restaurants; training and executing a first neural network to generate embeddings for each of the menu items; for a first subset of the set, calculating actual item-level preparation time vectors; for a second subset of the historical set, generating estimated item-level preparation time vectors; retrieving a set of order-level records for preparation of orders from the database; training a second neural network to predict the order-level preparation times, wherein inputs to the second neural network comprise one or more of the item-level preparation time vectors and metadata taken from the order-level records; and following training, executing the second neural network to generate predicted order-level preparation times for current orders within a restaurant.

A method for handoff includes: scanning an indirect token displayed on a guest device, where a guest has previously registered a payment instrument with a server, and associated the payment instrument with a plurality of indirect tokens, and the server has stored a payment token that associates the plurality of indirect tokens with the payment instrument in a database record along with a notification identifier; transmitting the ephemeral token to the server; simultaneously displaying details for a transaction, where the server has accessed the notification identifier to transmit a push notification to the guest device, and where the guest device is executing a proprietary application that allows for simultaneously displaying details for the transaction; and receiving notification from the server that the transaction is complete, where transaction completion data was entered via the guest device and transmitted to the server.

A computer-implemented method for predicting interchange charges includes: retrieving a historical transactions set, where each completed transaction in the set includes transaction features, a bank identification number (BIN), and a corresponding true interchange code; transforming all BINs in the set into a corresponding plurality of BIN features that comprise probabilities; creating a first training set including all transaction features, all pluralities of BIN features, and all true interchange codes associated with the historical transactions set; training a random forest model using the first training set and generating a second training set including rounded BIN features, rounded transaction features, discrete ones of the transaction features, and the true interchange codes; training the random forest model using the second training set to generate a trained random forest model for prediction of the interchange codes; and executing the trained random forest model for new transactions to generate corresponding predicted interchange codes.

A computer-implemented method for predicting interchange charges includes: retrieving a historical transactions set, where each completed transaction in the set includes transaction features, a bank identification number (BIN), and a corresponding true interchange code; transforming all BINs in the set into a corresponding plurality of BIN features that comprise probabilities; creating a first training set including all transaction features, all pluralities of BIN features, and all true interchange codes associated with the historical transactions set; training a random forest model using the first training set and generating a second training set including rounded BIN features, rounded transaction features, discrete ones of the transaction features, and the true interchange codes; training the random forest model using the second training set to generate a trained random forest model for prediction of the interchange codes; and executing the trained random forest model for new transactions to generate corresponding predicted interchange codes.

An apparatus for dynamic credit card reader mode management is provided, the apparatus including: a point-of sale (POS) terminal, configured to accept employee login credentials via a magnetic stripe card, to display menu items for selection, to add selected menu items to an order, and to accept payment for the order, where the payment may be obtained via swipe, dip, or tap, the POS terminal including: a first card reader, configured to read first provided cards when enabled; a second card reader, configured to read second provided cards when enabled; and a dual reader controller, coupled to the first and second card readers, configured to determine one or more workflow states of the POS terminal, and configured to dynamically control modes of the first and second card readers according to the one or more workflow states.

A computer-implemented method for predicting interchange charges includes: retrieving a historical transactions set, where each completed transaction in the set includes transaction features, a bank identification number (BIN), and a corresponding true interchange code; transforming all BINs in the set into a corresponding plurality of BIN features that comprise probabilities; creating a first training set including all transaction features, all pluralities of BIN features, and all true interchange codes associated with the historical transactions set; training a random forest model using the first training set and generating a second training set including rounded BIN features, rounded transaction features, discrete ones of the transaction features, and the true interchange codes; training the random forest model using the second training set to generate a trained random forest model for prediction of the interchange codes; and executing the trained random forest model for new transactions to generate corresponding predicted interchange codes.