Artificial intelligence (AI) has moved quickly from experimentation to expectation across many organizations. But a huge challenge is emerging in how to move beyond early pilot projects and embed AI into everyday operations.
That was a key theme during discussions at the ServiceNow AI Summit in London this week, where tech leaders from organizations including Rolls-Royce, Kingfisher, and National Grid shared their experiences of deploying AI in complex enterprise environments.
While enthusiasm for the technology remains high, the speakers highlighted a consistent theme: scaling AI requires far more than deploying new tools. Instead, organizations must tackle issues ranging from governance and culture to data quality, workforce training and the growing risk of “shadow AI”.
Adoption Is a Cultural Challenge
According to Rachel Cameron, VP of performance and improvement for Rolls-Royce’s digital IT services organization, a big problem is that many companies approach AI deployments as if they were conventional software projects.
“I think we treat AI as a software deployment often – and it’s not,” she said. “It’s a cultural shift.”
At Rolls-Royce, the company launched an AI-powered virtual agent called Merlin to support internal IT services. Adoption initially depended on demonstrating clear value to employees and building trust in the system. The tool has since grown rapidly, with usage increasing from around 800 conversations per month to 10,000. But Cameron said that growth required deliberate work around user engagement.
“You have to do adoption right. It has to give value at source,” she said.
That means involving early adopters, maintaining feedback loops, and ensuring AI tools solve genuine problems rather than simply showcasing new technology.
Priorities Compete With Innovation
Another obstacle to scaling AI is the reality of competing business priorities. Jon McKenna, director of service experience at retail giant Kingfisher, said firms often struggle to introduce new technologies while managing day-to-day operational demands.
“It’s just getting it out there. Business priorities are seasonal peak periods, selling to the customer and trying to launch a new thing on top of them,” he said.
That challenge is familiar to many large enterprises, where digital transformation initiatives must coexist with ongoing operational pressures. As a result, many organizations introduce AI gradually rather than attempting large-scale deployments from the outset.
Even when organizations are ready to scale AI, data quality often becomes a limiting factor. Dale Cheeseman, senior director of global IT operations at ServiceNow, said many companies underestimate how much preparation is required before AI tools can deliver meaningful results.
“AI isn’t Harry Potter. It isn’t magic,” he said. “You have to have a good, solid data platform.”
Historically, organizations relied on structured knowledge bases and manually recorded information to support IT service management systems. But in practice, those data sources are often incomplete or inconsistent. But as organizations move toward more autonomous AI systems, ensuring reliable data becomes even more important.
The Rise of Shadow AI
Another issue raised during the Summit was the rapid emergence of “shadow AI” – the unsanctioned use of AI tools by employees outside official governance frameworks.
Cheeseman said organizations sometimes encounter situations where teams deploy AI capabilities independently in order to accelerate innovation. While that can deliver quick wins, it also creates new risks.
“You end up with shadow IT organizations that are just deploying as fast as they can,” he said.
Without oversight, these deployments can create fragmented systems that are difficult to secure or manage.
Lee Massie, head of IT at Oxford University Hospitals NHS Foundation Trust, said his organization deliberately prioritized governance before expanding AI use.
“We started our AI journey a little earlier than many organizations. But it became quite evident quite quickly that we needed structural governance in place to support adoption,” he said,
One of the Trust’s pilot projects involves voice technology that listens to conversations between clinicians and patients and automatically generates clinical documentation. The tech has already demonstrated potential productivity benefits, but ensuring accuracy remains critical.
“There are real concerns around the legitimacy of the transcription in complex medical language,” said Massie.
Scaling Risk Management
In other sectors, the scale of operations is already pushing organizations toward more automated approaches. Jody Elliott, head of IT risk and sustainability at National Grid, described how AI is helping the company manage risk across its complex technology estate.
Large infrastructure organizations often run hundreds of projects simultaneously, each generating changes to systems, code, and operational processes. Maintaining visibility across all of that activity is difficult for human teams.
“In large organizations you’ve got multiple agile projects running. From a risk perspective, how do I have sight of every story and feature in every backlog?” said Elliot.
Generative AI is beginning to help analyse those large volumes of data and highlight potential issues.
“Generative AI gives us the opportunity to analyze all that unstructured data and turn out the top sort of ten highest-risk things every few weeks,” said Elliott.
Training and Trust
As AI becomes more embedded in organizational workflows, training is becoming another key priority. Elliott warned that employees may become overly reliant on AI-generated outputs if they do not understand how the technology works.
“There’s a risk that people become subject matter experts when they’re not subject matter experts,” he said.
To address that risk, National Grid has introduced AI training programmes across the organization, from executives to technical teams. The goal is to ensure employees understand both the capabilities and limitations of AI tools.
Training, added Elliott, must be continuous rather than a one-time exercise. As new tools emerge and capabilities evolve, organizations need to reinforce how they should be used.
Final Thoughts: Five Lessons for Scaling AI Beyond Pilots
The experiences shared during the Summit highlight several practical lessons for organizations looking to move AI from experimentation to production.
- Treat AI as organizational change, not just a software rollout: AI adoption often requires changes to processes, culture, and ways of working – not simply new tools.
- Establish governance before scaling deployments: Clear oversight frameworks help organizations manage risk and build confidence in AI systems.
- Invest in data readiness early: AI systems rely on high-quality operational data. Without it, scaling deployments becomes difficult.
- Focus on practical use cases first: Starting with targeted operational challenges can demonstrate value and build momentum.
- Train employees to use AI responsibly: Training programmes help employees understand both the capabilities and limitations of AI tools.