ServiceNow expands AI governance, emphasizes ROI

LAS VEGAS -- ServiceNow's AI Control Tower strengthens enterprise governance controls and reflects AI ROI, but industry experts said enterprises still need to see more real-world value demonstrations to get off the sidelines with AI agents.

AI Control Tower will absorb some features that ServiceNow rolled out with its AI Agent Orchestrator earlier this year, such as AI ROI reporting and correlation between AI usage and business performance metrics. Going forward, technical and developer-focused analytics will remain in the previously introduced Now Assist and AI Agent Studio tools to support agent creation and tuning workflows, according to a ServiceNow spokesperson.

AI Control Tower, by contrast, is meant to give business managers a centralized view of all the generative and agentic AI assets in use across an enterprise, including third-party tools and associated data. In addition to ROI data, such as hours saved by using AI tools, AI Control Tower will show AI adoption rates, risk and compliance issues.

"This has been top of mind for all of our customers as they start adopting generative AI and agentic AI," said Dorit Zilbershot, group vice president of AI experiences and innovation at ServiceNow, during a media prebriefing last week. "They want to understand, 'Is it really working? Is it really helping my employees?'"

Industry watchers said this emphasis on AI value reflects enterprises' ongoing struggles to reconcile generative AI and AI agent hype with their specific real-world environments. Concerns are also mounting about the effectiveness and scalability of AI agents driven by general-purpose large language models.

There is a real fear about bringing things into production, because everyone wants to see the guaranteed results.

"There is a real fear about bringing things into production, because everyone wants to see the guaranteed results," said Keith Kirkpatrick, an analyst at The Futurum Group.

However, this reluctance could create a chicken-and-egg scenario because early experimental use cases won't deliver big efficiency gains, he said.

"You're not going to see 10x efficiency gains until you can incorporate multiple workflow agents ... [but] most of the agents can't handle that sort of multithreaded problem," Kirkpatrick said. "We're just not anywhere near the point where we're getting a completely autonomous service experience."

Some AI agent early adopters have begun to measure the value of AI, but caution that it's still early for the technology. ServiceNow's AI Agent Orchestrator has only been generally available since March 12, said Dave Kanter, senior managing director and global team lead for ServiceNow partner Accenture's ServiceNow business group.

"We're starting these with just a couple of reasoning steps and decision points along the way," Kanter said of ServiceNow AI agents, which Accenture is testing for incident response. "We're going after incident types that happen 5% to 7% of the time, which is hundreds or thousands of times per month, and measuring the results closely on one or two decision steps."

However, Kanter said he's encouraged by the results of adding generative AI to two steps of the incident response process for Accenture's IT help desk, where the company runs ServiceNow's Now platform internally for its 801,000 employees. Since it added incident summarization for help desk pros responding to new incidents and for postincident reviews, it has seen a 10% decrease in its mean time to repair issues, Kanter said.

Still, IT pros will need to see more examples like this from industries similar to their own before they expand adoption of AI agents, said Neil Ward-Dutton, an analyst at IDC.

"What all our customers are crying out for is, 'Show us examples of companies that have actually succeeded with this stuff, where we can see not only what they achieved -- did they reduce their risk? Did they improve the quality of their service? Did they maybe make more money? -- but also, what did they do to get there?' Because the journey is not straightforward," he said.

AI Agent Fabric weaves in MCP, Agent2Agent protocol

With AI agent orchestration, governance and data services, ServiceNow is putting pieces in place similar to other large IT vendors that aspire to be the overarching control plane for enterprise agentic AI workloads, Ward-Dutton said. Competitors in this area include other IT and service management vendors such as Appian, Atlassian and UiPath, as well as enterprise application vendors such as Salesforce, Oracle and SAP, and finally, major cloud and productivity suite providers such as Microsoft and Google.

Like many of these competitors, ServiceNow touts connectors to third-party AI agents that it can orchestrate within its platform. This week, it also previewed AI Agent Fabric, which will incorporate emerging industry standards for interagent communication, such as Model Context Protocol (MCP) and Google's Agent2Agent protocol, for which ServiceNow was a launch partner last month. AI Agent Fabric is slated for release in the third quarter, but it will be longer before either open source protocol project is ready for enterprise production, Ward-Dutton said.

"MCP might be enterprise-ready in another year or so," he said. "But right now, any serious vendor needs to have their own [tools] as well."

It will also take more time for a dominant AI orchestration platform to materialize, but ServiceNow has the advantage of a 20-year history of automating IT and business workflows, according to Kirkpatrick.

"Like a lot of other big vendors, ServiceNow is trying to position [itself] as the 'platform of record,' with [comprehensive] access to relevant data," he said. "This is where ServiceNow likes to kind of plant a flag in the ground and say, 'We have all this process data built on years and years of working with organizations, so we understand what workflows are used in IT services.'"

Beth Pariseau, a senior news writer for Informa TechTarget, is an award-winning veteran of IT journalism covering DevOps. Have a tip? Email her or reach out @PariseauTT.