US 12,327,214 B1
Triggering a service mitigation action based on a causal-predictive system
Valerie Galluzzi Liptak, Mill Creek, WA (US); Sourav Kumar Agarwal, Hyderabad (IN); Tarun Bhatia, Sunnyvale, CA (US); and Amber Roy Chowdhury, Bellevue, WA (US)
Assigned to Amazon Technologies, Inc., Seattle, WA (US)
Filed by Amazon Technologies, Inc., Seattle, WA (US)
Filed on Dec. 13, 2022, as Appl. No. 18/080,191.
Int. Cl. G06Q 10/0833 (2023.01); G06Q 10/083 (2024.01)
CPC G06Q 10/0833 (2013.01) [G06Q 10/0838 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A computer system comprising:
one or more processors; and
one or more memory storing instructions that, upon execution by the one or more processors, configure the computer system to:
determine, from a datastore, that a past delivery to a location is associated with a past delivery defect;
determine, from the datastore, geographical data associated with the location, the geographical data including a distance derived from a geocode of the location and location data of a device associated with the past delivery, the geographical data further including an identifier of a source of the geocode;
generate a first embedding of first features of a first area that includes the location;
generate a second embedding of second features of a second area that includes the location;
generate a first input to a causal model, the first input comprising the identifier of the source of the geocode, the first embedding, and the second embedding, the causal model trained using predefined defect causes;
determine a first output of the causal model, the first output indicating a first likelihood of a cause of the past delivery defect, a plurality of possible causes of the past defect from the predefined defect causes, and a likelihood distribution of the possible causes;
generate a second input to a predictive model, the second input comprising the geographical data, the first embedding, the second embedding, and the first likelihood;
determine a second output of the predictive model based at least in part on the second input, the second output indicating a second likelihood of a future delivery defect, a plurality of possible future defects, and a likelihood distribution of the possible defects;
store, in the datastore, the second likelihood; and
cause, prior to a start of a delivery to the location, a mitigation action to be performed such that the future delivery defect is prevented.