Enabling Selective Automation of Human Decision-Making Using Rules as Preferences in a Service-industry Application

Rules, in their various forms, has been a prominent AI tool in automating
the processing of real-world applications. In this paper, we explore how rules
can be effectively used as preferences to expedite processing on behalf of
humans in the labor-intensive service industry. The new challenges we handle
stem from human behavior and audit/ quality considerations that impose design
considerations for the rules and the rule system. We consider the domain of
Information Technology (IT) change management which seeks to control and
reduce the risk of any alteration made to an IT infrastructure in its
hardware, software or attached network. We apply rules to selectively
automate the decision-making to proceed with a proposed change (called approval)
while respecting the considerations of multiple (human) role players. We piloted
our system, called Approval Accelerator, in a live environment with about 150
servers for a month and obtained up to 93\% reduction in effort and 44\% reduction
in end-to-end duration. We identify hither-to unknown issues about the impact of
the rules on the application process and how rules can be better managed
(e.g., storage, elicitation) to address them.