Enterprise AI resolved hundreds of IT support requests. It estimated savings of 406 hours in three months

ShipStation Global presented the results of an internal automated IT-support project. The system, called Deckhand, operates in the corporate Slack environment and uses Amazon Bedrock AgentCore technology. Before deployment, employees asked recurring technical questions through a shared communication channel. Answers often existed in internal documentation, older discussions or service guides, but finding them required work by IT specialists. During the twelve months before the project launched, more than 1,500 discussion threads were created. More than one fifth of requests arrived outside the normal working hours of the US team. The company also operates in the United Kingdom, Spain, Poland, Australia and the Philippines. Deckhand connects an internal knowledge base, historical resolved requests and Jira, the service-management system. It can answer common questions, analyze images of error messages and prepare a service ticket when necessary. An interesting part of the project was processing older discussions. Using language models, the developers identified threads in which the problem had actually been solved. They then converted these into structured guides. Quality control was important. Incomplete or unverified procedures were not supposed to enter the knowledge base. Otherwise, the system could repeat earlier incorrect recommendations. Another rule is the ability not to answer. The agent does not join ordinary social conversations, does not interrupt communication between employees and remains inactive if a human support worker has already taken over the issue. Results for the first three months show 1,935 processed messages and 1,221 conversations with the agent. Only 409 led to creation of a service ticket. The remaining 812 were resolved within the conversation. The company estimates that creating, processing and closing a routine service request takes about thirty minutes. On this basis it calculated savings of 406 hours over three months, or roughly 1,600 hours annually if the same volume of work is maintained. Positive ratings of responses rose from 82 percent in April to 96 percent in June. About one third of questions came outside IT working hours, and the agent answered 94 percent of them. The figures come from a supplier-and-customer case study. The hours saved are an estimate based on the assumed time needed to handle a service request, not an independent measurement of actual labor capacity released. For other companies, the process itself is more important. The project did not begin with an attempt to automate the entire IT department, but with the specific problem of repeated questions. The developers used existing documentation, defined the system’s authority and continuously evaluated feedback. The economic benefit came mainly from reducing unnecessary administration and making existing knowledge available faster. The model could also be applicable in HR departments, internal services or enterprise-application administration. KEY TERMS Knowledge base: An organized collection of guides, procedures and answers to recurring questions. Service-ticket deflection: Resolving a problem without the need to create and manually process a separate service ticket. Autonomous assistant: An AI system that can retrieve information and perform selected actions under defined rules.

Enterprise AI resolved hundreds of IT support requests. It estimated savings of 406 hours in three months

The full article is available to users registered at Hard Skills