Organisations are moving beyond the question of whether AI works. That’s easy to understand. The harder challenge is turning successful pilots into measurable business outcomes when AI is deployed across the wider enterprise.
In conversation with iTNews Asia, Melissa Ries, Group Vice President & Managing Director Asia at ServiceNow, explained that the barriers to scaling AI are increasingly operational rather than technical. She said while many organisations have invested heavily in AI pilots, they continue to struggle to translate those investments into broader business returns.
“Everyone’s moving budget to AI and has invested in AI, and there have been a lot of pilots. But the challenge has been that we’ve lots of incremental improvements, but no one’s really getting a full business outcome and an ROI,” Ries said.
The problem often lies beneath the AI itself.
If you put AI on top of the same disconnected systems that were already limiting productivity before, then it’s actually difficult to scale. You’re just automating broken processes.
-Melissa Ries, Group Vice President & Managing Director Asia at ServiceNow
Ries said the focus is shifting from proving that AI can work to making it work across the enterprise. “A year ago, we were asking, ‘Can AI work?’ Now it’s like, ‘Why isn’t it working across my entire business yet?’.
That becomes harder when AI has to operate across legacy systems, fragmented data, multiple functions and different regulatory environments. The organisations making progress, she said, are those connecting AI to their data, people and workflows.
Customer experience is revealing the gaps
Customer experience is one of the clearest indicators of whether those connections exist.
“A customer doesn’t see your organisational chart; they experience the outcome of it. They don’t know that billing sits in one organisation or one team and services with another. They just experience whether the organisation is connected or not,” Ries said.
She pointed to Griffith University in Australia, which first consolidated fragmented systems into an AI platform. Within six months, the university achieved an 87 percent first-contact resolution rate and a 43 percent improvement in first-contact resolution..
“None of that came from a smarter chatbot. It came from AI operating on workflows that are finally talking to each other,” she said.
The same problem affects employees. Customer service representatives can still spend significant time navigating systems, finding information, switching screens and waiting for other teams.
“If you put an AI assistant in but leave the five systems underneath untouched, then you haven’t really fixed anything,” Ries said. “You’ve just given them a faster way to switch between those five systems.”
Redesign the workflow
Ries distinguishes between automating individual tasks and transforming the processes around them. “Task automation gives you efficiency. Workflow transformation gives you leverage or an outcome,” she said.
She said organisations should redesign workflows so AI can surface information and execute actions within the flow of work rather than simply adding another layer on top of existing systems.
“If you automate a broken process, you don’t get transformation. You get a broken process moving faster with the bigger AI bill attached,” she said.
This becomes more important as enterprises adopt AI agents that can take actions rather than simply recommend or draft. “What’s proprietary is your business, your data, your policies, your processes and your institutional knowledge on how work flows through the organisation,” Ries said.
She pointed to semiconductor company Micron, which streamlined 32 processes across sales, quoting and service delivery and created greenfield platforms with AI-first foundations in eight months.
Agentic AI raises the governance stakes
As AI systems become more autonomous, governance also has to extend beyond the model.
“AI governance frameworks matter,” Ries said. “But a governance framework sitting on top of a fragmented environment gives you visibility into the fragments, not into how work actually flows across your enterprise.”
“You can’t govern what you can’t see. And I would take that a step further: you can’t govern what you have not connected.”
As AI agents gain the ability to act, she said governance needs to extend beyond models to identity, permissions, data and workflows.
“Identity, permissions, data and workflow governance now matter as much as model governance,” Ries said.
She cited Singapore’s Infocomm Media Development Authority approach to agentic AI governance, which she said addresses autonomy boundaries, access to tools and data, and human accountability.
“AI governance gives you the rules. Workflow governance is what makes those rules executable,” she said.
Measure the business outcome
Ries also argued that organisations need to change how they measure AI investments.
“The number of agents you’ve deployed or the prompts generated doesn’t necessarily tell you that your business is performing better,” she said.
Instead, organisations should track outcomes such as conversion, retention, customer resolution, risk reduction, time to market and employee onboarding.
She also highlighted the end-to-end cycle time. “If your AI is genuinely transforming work, then the whole workflow should move faster, not just one task within it,” Ries said.
Hours saved remain relevant, she said, but organisations should ask what employees do with that additional capacity. “AI adoption is an input. Productivity is an intermediate outcome. We should ultimately be aiming for business performance, and that’s what matters.”
A four-part test for AI readiness
For organisations deciding whether they are ready to scale AI, Ries said the starting point should be the work itself.
“Sit with the frontline employee and watch how they work. How many systems do they open? How often do they re-enter information? How many handoffs are involved? Where are they waiting for people?,” she explained.
From there, organisations should examine whether critical workflows are connected, whether data is accessible and trusted, and whether there is visibility into the AI agents operating across the business.
An immediate warning sign, she said, is “fragmentation without visibility”.
“If you’ve got different business units independently deploying agents, then nobody has an enterprise view of what’s running,” Ries said. “That also means there’s a risk in terms of permissions and accountability.”
Her readiness test is straightforward: “Can you see it? Can you connect it? Can you govern it? Can you measure it? If you can’t do those four things, you’re not ready to scale.”
The next phase is operational
Scaling AI may also require organisations to rethink accountability and, in some cases, organisational structures. Ries said she has seen companies combine functions such as HR and IT, while others are beginning to manage AI agents alongside human employees.
The next phase of AI adoption, she said, will depend less on adding more tools and more on changing how work is structured.
“The barrier to AI is operational.People do need to redesign work to get to their outcome, and then lastly, we’ve got to govern,
