Accession Number:

AD1098611

Title:

STP 8-A-06 AI Demand Forecasting Final Report

Descriptive Note:

Technical Report,23 Jan 2019,10 Sep 2019

Corporate Author:

Accenture Federal Services Arlington United States

Personal Author(s):

Report Date:

2019-09-10

Pagination or Media Count:

90.0

Abstract:

The Artificial Intelligence AI Demand Forecasting AIDF short-term project STP 8-A-06 explored the potential of leveraging emergent technology for improving DLAs ability to predict customer demand. The project also assessed DLAs technology environment for AI-based forecasting solutions as a critical component to future production enablement. This STP produced a proof-of-concept in collaboration with the WSSP R and D office, the J34 Center of Planning Excellence COPE, the Analytics Center of Excellence ACE, and the Strategic Technology Team J6T over a 9-month period, concluding 9302019.The key accomplishments and findings of the AIDF project can be summarized in five main points.1.Applied AI in Two Ways to Address Challenges 1 Developed new AI-based forecast algorithms not in use in the current JDA solution and 2 Applied AI to select and combine up to nine individual forecast models to create an item-specific model.2.AI Demand Forecasting Shows Improvement This two-fold application of AI demonstrated a 102M annual reduction in over-forecast error for the 48k item sample evaluated, without increasing the risk of an under-forecast error.3.There is no Universal Solution The evaluated AI-powered models were unable to improve accuracy for the sample population of items with extremely sparse demand. This finding supports DLAs current minimum threshold logic for item forecastability.4.Scaling AI Needs a Capable Environment AI models were developed on offline government laptops with modern data science software. Scalable AI development requires more robust software, hardware, and data pipelines.5.There is Value in Alternative Forecast Methods Several methods, including simple, non-AI forecasting methods, demonstrated potential improvement relative to current forecasting methods.

Subject Categories:

Distribution Statement:

APPROVED FOR PUBLIC RELEASE