When Chris Bedi joined ServiceNow as Chief Digital Information Officer in September 2015, he oversaw a small team of data scientists focused on AI and machine learning development.
Fast forward to 2023, and ServiceNow was piloting an application of generative artificial intelligence internally, highlighting 15 ways the technology could automate repetitive tasks for employees. These pilots are no longer seen as just a set of technology products for internal stakeholders, but are designed to pave the way for ServiceNow’s AI services for clients.
Bedi became the company’s chief customer officer in May 2024, and Kelly Lomack, whom the company hired in 2022 as senior vice president of digital technology experience, became CDIO, emphasizing ServiceNow’s strategy to develop AI tools in-house before deploying them to customers. Romac told Business Insider that he is spearheading the company’s efforts to develop and deploy AI tools that automate IT help desk requests and generate code for software developers.
“My team’s job is to create uniformity, an ecosystem of AI, within the company,” Lomac told Business Insider. “We serve the company and work on ourselves first.”
Kate Smaje, a senior partner at management consultancy McKinsey & Company, told Business Insider that taking an internal-first approach to AI development can help companies build confidence and learn, especially when soliciting feedback from employees.
technology
The generative AI tools that ServiceNow piloted and deployed internally influenced external products launched by the company in 2023 and beyond.
ServiceNow’s digital technology engineers, who report to Romack, work closely with product and platform engineers and focus on developing tools for customers, she said. This pipeline helped ServiceNow develop Workflow Data Fabric, a tool that uses machine learning to connect clients’ disparate systems, data, and employees.
A ServiceNow spokesperson said some engineers noticed while building the tool that the system was taking too long to send data. This issue was resolved internally and Workflow Data Fabric will be available to customers in October 2024.
Romac said he also led the development and launch of a governance-focused tool called AI Control Tower in the first quarter of 2024 to track the adoption of internal AI use cases and language models at scale.
Romack said that when ServiceNow launched the product for customers in May 2025, internal development of AI Control Tower led it to focus on three key themes: AI governance, efficiency improvement tracking, and workforce adoption.
result
By December 2025, ServiceNow will have more than 240 internal and external AI use cases, where different AI agents work together within specific workflows to achieve business outcomes, and nearly 3,000 customers using AI tools, according to a ServiceNow spokesperson.
According to Romack, one of the company’s most successful internally generated AI applications is in its IT service desk, where ServiceNow added agent AI capabilities in August 2025. This internal project led to the debut of Autonomous Workforce in February 2026. Autonomous Workforce is a tool that customers can use to resolve common IT issues such as password resets and network issues without human intervention.
Romack said AI adoption is not necessarily a smooth, linear process. For example, when ServiceNow started using generative AI for customer support summaries in 2023, cases were not always summarized accurately in early drafts. ServiceNow had to use employee feedback to identify needed fixes and “refine and adjust” the tool before making it available to customers, Romack said. Well, this is the strategy Romack uses when releasing new features.
“I don’t want to wait two weeks,” Romack said. “We’re talking about looking at it within 24 to 48 hours. With AI, you can get real-time data like this.”
According to Smadje, AI tools that create internal productivity “wins” don’t automatically translate to external audiences. This is because external users have different security protocols and different employee training protocols.
“Experimenting is just the first step,” Smadje said. “The real challenge and value is turning these learnings into robust systems that customers can trust and use at scale.”
