A Salvation Army Australia caseworker is sitting with someone who needs housing, family violence support, and emergency food aid. The work should be human and immediate, yet the screen demands a scavenger hunt through Workday, TechOne, ServiceNow, SharePoint, email, and policy pages before help can move.
For a 145-year-old organisation with 10,000 employees and 20,000 volunteers, that maze taxed workers’ attention. Chief Information Officer Milad Kruze saw the same failure pattern everywhere: the more energy staff spent decoding systems, the less capacity they had for the person in front of them.
Abstracting the noise
The Salvation Army did not treat AI as a shiny layer bolted onto broken process. It started with governance, because a faith-based charity handling vulnerable people cannot afford a system that guesses, leaks, or drifts.
Goldie became the front door. Built on Moveworks and connected through ServiceNow EmployeeWorks and Microsoft Teams, the AI assistant lets staff ask plain-language questions while the machinery behind it routes requests, retrieves approved knowledge, and completes routine work.
ServiceNow remained the operational spine, with more than 100,000 automated workflows each month and 60,000 IT tickets a year. Goldie sat above that stack like a control surface, turning fragmented enterprise software into a single conversation on mobile or desktop.
Successful adoption
Adoption worked because the launch felt social, not imposed.
When we were closing the project off, the project owner coined this term, ‘positive FOMO’. It was just beautiful, because I think that's what we achieved. We created this organic interest around Goldie.
- Milad Kruze, Chief Information Officer, The Salvation Army.
Leaders introduced Goldie as a new puppy joining the workforce, then asked staff to feed it structured knowledge, correct weak answers, and help it learn the rhythm of frontline work. Regional teams tested the questions that mattered before launch, which gave the rollout its uncommon advantage: demand arrived before the mandate.
Understanding hard boundaries, building trust
The governance lesson was simple: Goldie needed a job description before it needed scale. It was framed as an internal support assistant, not a general-purpose chatbot, which gave executives a clean way to decide what belonged inside the system and what stayed out.
The clearest no-go zone was client case notes. Goldie could help staff find policies, forms, and operational answers, but it would not generate sensitive narrative records about the people seeking help.
Access control became the make-or-break layer. An AI assistant can only be as safe as the repositories it searches, so the team classified files, tightened SharePoint permissions, and used tools such as Varonis to stop salary data, board records, and personal information from surfacing to the wrong user.
Measuring impact
Nine months after production release, following four months of testing, Goldie had logged 22,000 conversations and resolved nearly 90 percent of internal IT support queries without human escalation.
Kruze described the operating model as shift-left, meaning specialist knowledge moves closer to the frontline instead of sitting behind escalation queues. In practice, Goldie became level-one support for everyday operational friction and returned more than 5,000 hours of productive capacity to staff, including 3,000 hours for frontline teams.
The deeper win was speed. Standard ticket reassignments had created four-day delays, and manual approval chains added six more. Removing those handoffs turned a support desk into an automation layer and pushed knowledge creation up by 293 per cent as staff improved the material Goldie used.
Goldie supplemented care work. It stripped away the software fog around it. For The Salvation Army Australia, the breakthrough was thanks to an enterprise system quiet enough to let humans do the work.
