Using a trained model of an online system to generate action recommendations by predicting future demand
A trained model of an online system is used to generate action recommendations by predicting future demands. The online system gathers in-store data by receiving, from a device of a picker and/or a computing system of an in-store physical receptacle, data with information about an inventory of an item. The online system estimates, based on conversion data for the item, a level of inventory for the item. The trained model is then applied to predict, based on the in-store data and the estimated level of inventory, a demand prediction score indicative of a future demand for the item. The online system generates, based on the estimated level of inventory and the demand prediction score, a depletion metric indicative of a time period until the inventory of the item is depleted. Based on the depletion metric, the online system triggers an action in relation to the inventory of the item.
1 . A method, performed at a computer system comprising a processor and a computer-readable medium, comprising:
gathering, via cameras of a smart shopping cart at a location of a retailer, in-store data including visual data that represent images of shelves and aisles at the location of the retailer indicative of a current inventory of an item at the location of the retailer;
receiving, via a network and from the smart shopping cart utilized for shopping at the location of the retailer, the in-store data with information about the current inventory of the item at the location of the retailer;
obtaining conversion data by receiving, by an order management module of the computer system via the network and from a plurality of devices associated with a plurality of users of an online system, real time information about conversions of the item associated with a plurality of orders placed by the plurality of users;
estimating, by an inventory prediction module of the computer system and based at least in part on the conversion data, a level of inventory for the item at the location of the retailer;
accessing, by a demand prediction module of the computer system, a demand prediction model, wherein the demand prediction model is a machine-learning model trained to predict a future demand for the item;
applying the demand prediction model to the in-store data, the conversion data and the level of inventory to generate a demand prediction score for the item indicative of the future demand for the item for a defined future time period;
generating, using the level of inventory and the demand prediction score, a depletion metric for the item indicative of a time period until the inventory of the item is depleted at the location of the retailer;
generating, using the depletion metric, action data associated with one or more actions in relation to the inventory of the item; and
communicating, from an action application module of the computer system via the network and to a computing system associated with the retailer, the action data prompting the one or more actions in relation to the inventory of the item.
2 . The method of claim 1 , wherein gathering the in-store data further comprises:
receiving, via the network and from a device associated with a picker, scanning data with information about purchases of the item over a defined time period.
3 . The method of claim 1 , wherein estimating the level of inventory for the item further comprises:
receiving, via the network and from a device associated with a picker, picker data with information about at least one of one or more replacements of the item or one or more refunds in relation to the item; and
estimating, further based on the picker data, the level of inventory for the item.
4 . The method of claim 1 , wherein obtaining the conversion data further comprises:
retrieving, from a database of the online system, data with information about conversion of the item over a defined time period.
5 . The method of claim 1 , wherein obtaining the conversion data comprises:
receiving, via the network and from one or more devices associated with one or more users of the online system, data with information about conversions of the item over a defined time period.
6 . The method of claim 1 , wherein estimating the level of inventory comprises:
accessing an inventory prediction model, wherein the inventory prediction model is a machine-learning model trained to predict the level of inventory of the item; and
applying the inventory prediction model to the conversion data to predict the level of inventory of the item.
7 . The method of claim 1 , wherein estimating the level of inventory comprises:
accessing an inventory prediction model, wherein the inventory prediction model is a machine-learning model trained to predict the level of inventory of the item; and
applying the inventory prediction model to the conversion data and the in-store data to predict the level of inventory of the item.
8 . The method of claim 7 , further comprising:
re-training the inventory prediction model by updating, using the depletion metric, a set of parameters of the inventory prediction model.
9 . The method of claim 1 , further comprising:
generating training data by collecting information about a set of demands for the item achieved by a first collection of users of the online system over a defined time period given a set of inventories for the item;
training the demand prediction model using the training data to generate a set of initial values for a set of parameters of the demand prediction model;
collecting feedback data with information about conversions of the item by a second collection of users of the online system upon the retailer performed an action in relation to the inventory of the item in response to the action data; and
re-training the demand prediction model by updating, using the feedback data, the set of parameters of the demand prediction model.
10 . The method of claim 1 , wherein communicating the action data comprises:
communicating, via the network and to the computing system, a command that triggers an auto-order at the computing system for resupplying the inventory of the item.
11 . The method of claim 1 , wherein communicating the action data comprises:
communicating, via the network and to the computing system, an alert message about a depletion of the inventory of the item within the time period prompting the retailer to resupply the inventory of the item.
12 . A computer program product comprising a non-transitory computer readable storage medium having instructions encoded thereon that, when executed by a processor, cause the processor to perform steps comprising:
gathering, via cameras of a smart shopping cart at a location of a retailer, in-store data including visual data that represent images of shelves and aisles at the location of the retailer indicative of a current inventory of an item at the location of the retailer;
receiving, via a network and from the smart shopping cart utilized for shopping at the location of the retailer, the in-store data with information about the current inventory of the item at the location of the retailer;
obtaining conversion data by receiving, by an order management module of a computer system via the network and from a plurality of devices associated with a plurality of users of an online system, real time information about conversions of the item associated with a plurality of orders placed by the plurality of users;
estimating, by an inventory prediction module of the computer system and based at least in part on the conversion data, a level of inventory for the item at the location of the retailer;
accessing, by a demand prediction module of the computer system, a demand prediction model, wherein the demand prediction model is a machine-learning model trained to predict a future demand for the item;
applying the demand prediction model to the in-store data, the conversion data and the level of inventory to generate a demand prediction score for the item indicative of the future demand for the item for a defined future time period;
generating, using the level of inventory and the demand prediction score, a depletion metric for the item indicative of a time period until the inventory of the item is depleted at the location of the retailer;
generating, using the depletion metric, action data associated with one or more actions in relation to the inventory of the item; and
communicating, from an action application module of the computer system via the network and to a computing system associated with the retailer, the action data prompting the one or more actions in relation to the inventory of the item.
13 . The computer program product of claim 12 , wherein the instructions further cause the processor to perform steps comprising:
gathering the in-store data further by receiving, via the network and from a device associated with a picker, data with information about at least one of one or more replacements of the item or one or more refunds in relation to the item; and
gathering the in-store data further by receiving, via the network and from the device associated with the picker, scanning data with information about purchases of the item over a defined time period.
14 . The computer program product of claim 12 , wherein the instructions further cause the processor to perform steps comprising:
obtaining the conversion data by further retrieving, from a database of the online system, data with information about conversions of the item over a defined time period.
15 . The computer program product of claim 12 , wherein the instructions further cause the processor to perform steps comprising:
accessing an inventory prediction model, wherein the inventory prediction model is a machine-learning model trained to predict the level of inventory of the item; and
applying the inventory prediction model to the conversion data to predict the level of inventory of the item.
16 . The computer program product of claim 12 , wherein the instructions further cause the processor to perform steps comprising:
accessing an inventory prediction model, wherein the inventory prediction model is a machine-learning model trained to predict the level of inventory of the item; and
applying the inventory prediction model to the conversion data and the in-store data to predict the level of inventory of the item.
17 . The computer program product of claim 12 , wherein the instructions further cause the processor to perform steps comprising:
generating training data by collecting information about a set of demands for the item achieved by a first collection of users of the online system over a defined time period given a set of inventories for the item;
training the demand prediction model using the training data to generate a set of initial values for a set of parameters of the demand prediction model;
collecting feedback data with information about conversions of the item by a second collection of users of the online system upon the retailer performed an action in relation to the inventory of the item in response to the action data; and
re-training the demand prediction model by updating, using the feedback data, the set of parameters of the demand prediction model.
18 . The computer program product of claim 12 , wherein the instructions further cause the processor to perform steps comprising:
communicating the action data by communicating, via the network and to the computing system, an alert message about a depletion of the inventory of the item within the time period prompting the retailer to resupply the inventory of the item.
19 . A computer system comprising:
a processor; and
a non-transitory computer-readable storage medium having instructions that, when executed by the processor, cause the computer system to perform steps comprising:
gathering, via cameras of a smart shopping cart at a location of a retailer, in-store data including visual data that represent images of shelves and aisles at the location of the retailer indicative of a current inventory of an item at the location of the retailer;
receiving, via a network and from the smart shopping cart utilized for shopping at the location of the retailer, the in-store data with information about the current inventory of the item at the location of the retailer;
obtaining conversion data by receiving, by an order management module of the computer system via the network and from a plurality of devices associated with a plurality of users of an online system, real time information about conversions of the item associated with a plurality of orders placed by the plurality of users;
estimating, by an inventory prediction module of the computer system and based at least in part on the conversion data, a level of inventory for the item at the location of the retailer;
accessing, by a demand prediction module of the computer system, a demand prediction model, wherein the demand prediction model is a machine-learning model trained to predict a future demand for the item;
applying the demand prediction model to the in-store data, the conversion data and the level of inventory to generate a demand prediction score for the item indicative of the future demand for the item for a defined future time period;
generating, using the level of inventory and the demand prediction score, a depletion metric for the item indicative of a time period until the inventory of the item is depleted at the location of the retailer;
generating, using the depletion metric, action data associated with one or more actions in relation to the inventory of the item; and
communicating, from an action application module of the computer system via the network and to a computing system associated with the retailer, the action data prompting the one or more actions in relation to the inventory of the item.