A technically functional AI system isn’t necessarily a successful product.
Enterprise AI pilots can deliver promising technical results without establishing a viable path to production. A solution may perform well in a controlled environment while leaving questions about operational requirements, organizational ownership, and long-term investment unresolved. Those uncertainties can prevent successful experiments from becoming services that deliver sustained business value.
Marco Fahmi, former Director of Artificial Intelligence at the Queensland Government's Department of Customer Services, has extensive experience leading enterprise AI strategy and implementation. He helped establish Queensland Government's AI capabilities, including QChat, a generative AI assistant deployed to more than 50,000 users, while also encountering promising initiatives that never progressed beyond experimentation.
"I've discovered that technology is actually the simplest part of the problem," Fahmi says. Technical success is only an early milestone in enterprise AI adoption, with much of the difficulty emerging as organizations attempt to turn experiments into operational services.
Production environments expose limitations that AI pilots overlook
Enterprise deployment introduces exceptions, different user needs, and operational requirements that the initial experiment may never have considered. "A solution might solve 80% of the problems, but the hard work is in the other 20%," Fahmi says. Those remaining edge cases can be as complex as the primary use cases, requiring substantial additional development and operational planning.
Technical performance must also meet user expectations. Fahmi points to a Queensland Government example involving a department that developed an AI chatbot to provide information to an external community. The department wanted to minimise the risk of the chatbot providing incorrect information, so it introduced additional controls to verify accuracy, which significantly slowed responses and degraded the user experience. “A technically functional AI system isn’t necessarily a successful product,” he says.
Enterprise AI can outgrow the departments that develop it
AI applications frequently depend on capabilities distributed across IT departments, business units, and operational teams. Bringing those capabilities together becomes difficult when established responsibilities leave no single team accountable for the resulting service.
Fahmi encountered this problem when a Queensland Government department developed an HR chatbot capable of answering employee questions using internal policies and expert knowledge. The proof of concept worked, but the team lacked the resources and expertise to operate a custom AI application. A whole-of-government deployment could have spread operating costs across more users, but the department had no mandate to deliver such a service, and no other unit was positioned to take ownership. "Traditionally, there's a distinction between IT departments and business units. AI sits right on the boundary between the two, and sometimes it's unclear who's in charge," Fahmi explains.
Scaling AI introduces costs that initial budgets rarely anticipate
Small-scale AI experiments typically involve limited users, defined workloads, and contained development costs. Production introduces customization, infrastructure, maintenance, and ongoing service requirements that can substantially increase spending. "When organizations want to move from a small-scale test to full-scale implementation, they suddenly realize the cost increases significantly in ways they didn't anticipate," Fahmi says.
Executive enthusiasm can secure initial funding, but production often requires decisions that affect multiple departments, budgets, and established business priorities. Without a clear strategic direction, organizations risk supporting technically successful initiatives that can’t be integrated into their operations. "You need to integrate AI into your strategic plan, implementation plans, and operations, and you need senior executives to back it," Fahmi explains.
Executive interest in AI opens the door for broader IT modernization
Fahmi sees an opportunity in the attention generative AI has attracted from senior leadership. Executives who previously showed limited interest in technology are now engaging with questions about data, infrastructure, security, and legacy systems, creating opportunities to address the underlying constraints that have long complicated enterprise technology adoption.
"If you get a sponsor who understands AI, that's a sponsor who understands data, IT infrastructure, and electronic services. That's a rare sponsor, but it's the sponsor you really want," Fahmi concludes.
