Out-of-stock prediction model and intervention method
Systems and methods for determining out-of-stock predictions for merchant items and performing intervention actions based on the out-of-stock predictions using machine-learned models. The method includes obtaining feature data including at least (i) aggregations of historical found rate data, (ii) item level metadata, and (iii) merchant signal data into an out-of-stock model. The method includes determining a number of output scores for a number of items. The output scores can be indicative of a likelihood of an item being out-of-stock. The method includes determining a first output score for a first item of the plurality of items satisfies an intervention criterion. The method includes responsive to determining the first output score satisfies the intervention criterion, performing an intervention action.
1 . A computing system, comprising:
one or more processors; and
one or more non-transitory computer-readable media storing instructions that are executable by the one or more processors to cause the computing system to perform operations, the operations comprising:
obtaining feature data associated with one or more third-party merchants comprising at least (i) aggregations of historical found rate data, (ii) item level metadata, and (iii) merchant signal data into an out-of-stock model;
determining, by the out-of-stock model, a plurality of output scores for a plurality of items associated with the one or more third-party merchants, wherein the output scores are indicative of a likelihood of an item being out-of-stock, wherein the out-of-stock model comprises a machine learning model trained on training data generated based on at least one of sensor data or user input data obtained via one or more shopper devices located within a merchant location performing an item check;
determining a first output score for a first item of the plurality of items satisfies an intervention criterion; and
responsive to determining the first output score satisfies the intervention criterion, performing an intervention action.
2 . The computing system of claim 1 , wherein the operations comprise:
determining the intervention action based on user session data and the first output score.
3 . The computing system of claim 2 , wherein determining the intervention action is based on one or more output scores of the plurality of output scores.
4 . The computing system of claim 2 , wherein the intervention action comprises at least one of: (i) removing the first item from a catalog associated with a merchant associated with the first item or (ii) providing for display a notification via the user interface.
5 . The computing system of claim 4 , wherein the notification comprises at least one of: (i) a message indicative of the first item likely being out-of-stock or (ii one or more substitution recommendations.
6 . The computing system of claim 1 , wherein the training data comprises at least: (i) the one or more items and (ii) an out-of-stock or in-stock indication for each respective item of the one or more items.
7 . The computing system of claim 1 , wherein the training data comprises at least: (i) the one or more items and (ii) an out-of-stock or in-stock indication for each respective item of the one or more items.
8 . The computing system of claim 1 , wherein the output score satisfying the intervention criterion comprises the output score exceeding a predetermined probability that the item is likely out-of-stock.
9 . The computing system of claim 8 , wherein each item of the plurality of items has a unique predetermined probability.
10 . The computing system of claim 9 , wherein the intervention action is determined based on a subset of the plurality of output scores.
11 . A computer-implemented method comprising:
accessing a plurality of inventory listings associated with a plurality of third-party merchants, the inventory listings comprising a plurality of items and plurality of quantities of the respective items;
obtaining data comprising one or more features associated with the plurality of items associated with the plurality of third-party merchants;
inputting the obtained data as input into an out-of-stock prediction model;
generating, by the out-of-stock prediction model, a probability that a first item of the plurality of items is out-of-stock, wherein the out-of-stock model comprises a machine learning model trained on training data generated based on at least one of sensor data or user input data obtained via one or more shopper devices located within a merchant location performing an item check;
determining, based on the probability that the first item of the plurality of items is out-of-stock, an intervention action; and
transmitting data comprising instructions that when read by a processor cause a computing device to perform operations, the operations comprising performing, the intervention action.
12 . The computer-implemented method of claim 11 , comprising:
determining the intervention action based on: (i) user session data, (ii) a first output score, and (iii) historical user data.
13 . The computer-implemented method of claim 12 , wherein the historical user data comprises at least one of: (i) one or more user preferences, (ii) one or more prior orders, or (iii) one or more frequently ordered items.
14 . The computer-implemented method of claim 11 , wherein the intervention action comprises at least one of: (i) removing the first item from a catalog associated with a merchant associated with the first item or (ii) providing for display a notification via a user interface.
15 . The computer-implemented method of claim 14 , wherein removing the first item from a catalog associated with the merchant associated with the first item results in the item being inaccessible via an in-application query.
16 . The computer-implemented method of claim 14 , wherein providing for display a notification via the user interface comprises at least one of: (i) a message indicative of the first item likely being out-of-stock, (ii) one or more substitution recommendations, or (iii) a message indicative of the first item being out-of-stock.
17 . One or more non-transitory computer readable media storing instructions that are executable by one or more processors to perform operations comprising:
processing the merchant data via a feature engineering pipeline to normalize the merchant data obtained from at least one of the plurality of third-party merchants;
accessing a plurality of inventory listings associated with a plurality of third-party merchants, the inventory listings comprising a plurality of items and plurality of quantities of the respective items;
obtaining data comprising one or more features associated with the plurality of items associated with the plurality of third-party merchants;
inputting the obtained data as input into an out-of-stock prediction model;
generating, by the out-of-stock prediction model, a probability that a first item of the plurality of items is out-of-stock, wherein the out-of-stock model comprises a machine learning model trained on training data generated based on at least one of sensor data or user input data obtained via one or more shopper devices located within a merchant location performing an item check;
determining, based on the probability that the first item of the plurality of items is out-of-stock, an intervention action; and
transmitting data comprising instructions that when read by a processor cause a computing device to perform operations, the operations comprising performing, the intervention action.
18 . The computing system of claim 1 , wherein the merchant signal data is obtained at inconsistent cadences, wherein the operations comprise:
processing, by a feature engineering pipeline, the merchant data to normalize the data by at least one of: (i) performing extrapolation of the merchant data or (ii) pulling the merchant data until new data for a respective merchant is obtained.