Systems and methods for machine-learning optimization for estimated time of arrival analysis
Techniques for determining an estimated time of arrival associated with a pickup location are described herein. For example, a customer can order a good or service for pickup at a pickup location. In some examples, order data can include a pickup location intent, such as a drive-through location, a pickup window, or a parking location. A location of a user computing device can be determined and can be input to a machine-learning model trained to determine estimated times of arrival for a particular location. In some examples, the machine-learning model can be trained with ground truth data specific to the location for which the model is to determine estimated times of arrival. The machine-learning model can output the estimated time of arrival and can update the time at any regular or irregular intervals, which can be used to prioritize an order in an order queue.
1 . A system comprising:
one or more processors; and
one or more non-transitory computer-readable media storing computer executable instructions that, when executed, cause the one or more processors to perform operations comprising:
receiving map data representing a plurality of polygonal areas proximate a store;
receiving a pickup order associated with a requesting entity, the pickup order indicating at least one good or service and a pickup location intent;
associating the pickup order with a user computing device associated with a pickup entity;
receiving location data from the user computing device;
determining additional data associated with at least one of the user computing device or the store;
refraining from inputting the pickup order, the location data, the pickup location intent, and the additional data into a machine-learning model in response to determining that a distance between the user computing device and the pickup location intent is above a threshold distance;
inputting the pickup order, the location data, the pickup location intent, and the additional data into the machine-learning model in response to determining that the distance between the user computing device and the pickup location intent is below the threshold distance, wherein the machine-learning model is trained to determine a particular estimated time of arrival of a particular entity to a pickup location of the store, wherein the machine-learning model is further trained based at least in part on ground truth data representing actual travel times between a first location of a training user computing device and a second location associated with the store and a ground truth intent representing a declared region to arrive at the store, wherein the declared region corresponds to an area of the plurality of polygonal areas associated with the store;
reducing a rate at which the particular estimated time of arrival is determined based at least in part on determining an accuracy with which the machine-learning model determines the particular estimated time of arrival, wherein reducing the rate at which the particular estimated time of arrival is determined causes an overall processing of the one or more processors to be reduced;
receiving, from the machine-learning model and based at least in part on the pickup order, the location data, and the additional data, an estimated time of arrival of the pickup entity to a polygonal area of the plurality of polygonal areas associated with the pickup location intent; and
sending, to a computing device associated with the store, the estimated time of arrival of the pickup entity to arrive at the polygonal area of the plurality of polygonal areas.
2 . The system of claim 1 , wherein the polygonal area of the plurality of polygonal areas represents at least one of:
a premises boundary associated with the store;
a location of a drive-through associated with the store;
a location of a curbside-pickup location associated with the store;
a location of a parking lot associated with the store; or
a physical location within the store.
3 . The system of claim 1 , the operations further comprising receiving, from the machine-learning model, confidence data associated with the estimated time of arrival.
4 . The system of claim 1 , wherein the additional data includes at least one of:
a time of day the estimated time of arrival is determined;
a day of a month the estimated time of arrival is determined;
a day of a year the estimated time of arrival is determined;
a minute of an hour the estimated time of arrival is determined;
a weather pattern;
order volume associated with the store;
a number of customers associated with a location of the store;
wait time associated with the store;
order throughput data associated with the store; or
a direction the requesting entity is relative to the store.
5 . The system of claim 1 , wherein the location data is first location data at a first time and the estimated time of arrival is a first estimated time of arrival determined at the first time, the operations further comprising:
receiving, at a second time after the first time, second location data from the user computing device;
inputting, based on receiving the second location data from the user computing device, the pickup order, the second location data, and the additional data into the machine-learning model;
receiving, from the machine-learning model and based at least in part on the pickup order, the second location data, and the additional data, a second estimated time of arrival of the pickup entity to the polygonal area of the plurality of polygonal areas associated with the pickup location intent; and
sending, to the computing device associated with the store, the second estimated time of arrival of the pickup entity to arrive at the polygonal area of the plurality of polygonal areas.
6 . The system of claim 1 , wherein the machine-learning model includes at least one of a random decision forest model or a deep neural network.
7 . A method, implemented at least in part by one or more computing devices, the method comprising:
receiving, from a merchant associated with a store, map data representing a plurality of polygonal areas proximate the store;
receiving a pickup order indicating at least one good or service and a pickup location intent;
associating the pickup order with a user computing device associated with a customer;
receiving location data from the user computing device;
determining additional data associated with the user computing device;
refraining from inputting the pickup order, the location data, the pickup location intent, and the additional data into a machine-learning model in response to determining that a distance between the user computing device and the pickup location intent is above a threshold distance;
inputting the pickup order, the location data, the pickup location intent, and the additional data into the machine-learning model in response to determining that the distance between the user computing device and the pickup location intent is below the threshold distance, wherein the machine-learning model is trained to determine a particular estimated time of arrival of a particular customer to a location of the store, wherein the machine-learning model is further trained based at least in part on ground truth data representing actual travel times between a first location of a training user computing device and a second location associated with the store and a ground truth intent representing a declared region to arrive at the store, wherein the declared region corresponds to an area of the plurality of polygonal areas associated with the store;
reducing a rate at which the particular estimated time of arrival is determined based at least in part on determining an accuracy with which the machine-learning model determines the particular estimated timed of arrival, wherein reducing the rate at which the particular estimated time of arrival is determined causes an overall processing of at least one processor of the one or more computers to be reduced;
receiving, from the machine-learning model and based at least in part on the pickup order, the location data, the pickup location intent, and the additional data, an estimated time of arrival of the customer to a polygonal area of the plurality of polygonal areas associated with the pickup location intent; and
sending, to a computing device associated with the store, the estimated time of arrival of the customer to arrive at the polygonal area of the plurality of polygonal areas.
8 . The method of claim 7 , wherein the polygonal area of the plurality of polygonal areas represents at least one of:
a premises boundary associated with the store;
a location of a drive-through associated with the store;
a location of a curbside-pickup location associated with the store;
a location of a parking lot associated with the store; or
a physical location within the store.
9 . The method of claim 7 , further comprising receiving, from the machine-learning model, confidence data associated with the estimated time of arrival.
10 . The method of claim 7 , wherein the additional data includes at least one of:
a time of day the estimated time of arrival is determined;
a day of a month the estimated time of arrival is determined;
a day of a year the estimated time of arrival is determined;
a minute of an hour the estimated time of arrival is determined;
a weather pattern;
order volume associated with the store;
a number of customers associated with the location of the store;
wait time associated with the store;
order throughput data associated with the store; or
a direction the customer is relative to the store.
11 . The method of claim 7 , wherein the location data is first location data at a first time and the estimated time of arrival is a first estimated time of arrival determined at the first time, the method further comprising:
receiving, at a second time after the first time, second location data from the user computing device;
inputting, based on receiving the second location data from the user computing device, the pickup order, the second location data, and the additional data into the machine-learning model;
receiving, from the machine-learning model and based at least in part on the pickup order, the second location data, and the additional data, a second estimated time of arrival of the customer to the polygonal area of the plurality of polygonal areas associated with the pickup location intent; and
sending, to a computing device associated with the store, the second estimated time of arrival of the customer to arrive at the polygonal area of the plurality of polygonal areas.
12 . The method of claim 7 , wherein the machine-learning model includes at least one of a random decision forest model or a deep neural network.
13 . One or more non-transitory computer-readable media storing computer executable instructions that, when executed, cause one or more processors to perform operations comprising:
receiving, from a merchant associated with a store, map data representing a plurality of polygonal areas proximate the store;
receiving a pickup order indicating at least one good or service and a pickup location intent;
associating the pickup order with a user computing device associated with a customer;
receiving location data from the user computing device;
determining additional data associated with the user computing device;
refraining from inputting the pickup order, the location data, the pickup location intent, and the additional data into a machine-learning model in response to determining that a distance between the user computing device and the pickup location intent is above a threshold distance;
inputting the pickup order, the location data, the pickup location intent, and the additional data into the machine-learning model in response to determining that the distance between the user computing device and the pickup location intent is below the threshold distance, wherein the machine-learning model is trained to determine a particular estimated time of arrival of a particular customer to a location of the store, wherein the machine-learning model is further trained based at least in part on ground truth data representing actual travel times between a first location of a training user computing device and a second location associated with the store and a ground truth intent representing a declared region to arrive at the store, wherein the declared region corresponds to an area of the plurality of polygonal areas associated with the store;
reducing a rate at which the particular estimated time of arrival is determined based at least in part on determining an accuracy with which the machine-learning model determines the particular estimated timed of arrival, wherein reducing the rate at which the particular estimated time of arrival is determined causes an overall processing of the one or more processors to be reduced;
receiving, from the machine-learning model and based at least in part on the pickup order, the location data, and the additional data, an estimated time of arrival of the customer to a polygonal area of the plurality of polygonal areas associated with the pickup location intent; and
sending, to a computing device associated with the store, the estimated time of arrival of the customer to arrive at the polygonal area of the plurality of polygonal areas.
14 . The one or more non-transitory computer-readable media of claim 13 , wherein the polygonal area of the plurality of polygonal areas represents at least one of:
a premises boundary associated with the store;
a location of a drive-through associated with the store;
a location of a curbside-pickup location associated with the store;
a location of a parking lot associated with the store; or
a physical location within the store.
15 . The one or more non-transitory computer-readable media of claim 13 , the operations further comprising receiving, from the machine-learning model, confidence data associated with the estimated time of arrival.
16 . The one or more non-transitory computer-readable media of claim 13 , wherein the additional data includes at least one of:
a time of day the estimated time of arrival is determined;
a day of a month the estimated time of arrival is determined;
a day of a year the estimated time of arrival is determined;
a minute of an hour the estimated time of arrival is determined;
a weather pattern;
order volume associated with the store;
a number of customers associated with the location of the store;
wait time associated with the store;
order throughput data associated with the store; or
a direction the customer is relative to the store.
17 . The one or more non-transitory computer-readable media of claim 13 , wherein the location data is first location data at a first time and the estimated time of arrival is a first estimated time of arrival determined at the first time, the operations further comprising:
receiving, at a second time after the first time, second location data from the user computing device;
inputting, based on receiving the second location data from the user computing device, the pickup order, the second location data, and the additional data into the machine-learning model;
receiving, from the machine-learning model and based at least in part on the pickup order, the second location data, and the additional data, a second estimated time of arrival of the customer to the polygonal area of the plurality of polygonal areas associated with the pickup location intent; and
sending, to the computing device associated with the store, the second estimated time of arrival of the customer to arrive at the polygonal area of the plurality of polygonal areas.
18 . The system of claim 1 , wherein the declared region is received from the training user computing device.
19 . The system of claim 1 , wherein the machine-learning model is retrained based at least in part on at least one of:
a frequency of orders placed at the store;
a number of customers at the store;
a number of requests made to the machine-learning model; or
a difference between outputs from the machine-learning model and actual times of arrival.
20 . The system of claim 1 , the operations further comprising:
updating an order processing queue based at least in part on the estimated time of arrival received from the machine-learning model.