Enterprise AI Moves From Pilots to Procurement and Process Governance

The conversation around enterprise AI has shifted. After two and a half years of experimentation, organizations are no longer asking whether AI can perform specific tasks. They're asking who pays for it, who's accountable when it fails, and how to integrate it into existing business processes without creating new operational headaches.

Two recent industry discussions—one from procurement platform Levelpath, the other from customer experience vendor NiCE—arrived at the same conclusion from different angles: proving the technology works is the easy part.

The procurement reality check

Levelpath surveyed enterprise software buyers and found that 57% experienced at least one AI spending issue during the previous six months. The most common problem was invoices arriving higher than budgeted, followed by teams hitting usage caps and organizations redirecting budget from other priorities to cover unexpected AI costs.

"The most consistent thing we're hearing from procurement leaders, across different industries and company sizes, is that the governance gap catches them off guard," said Stan Garber, president and co-founder of Levelpath.

Buyers are responding by demanding transparency rather than imposing hard limits. The survey found buyers were twice as likely to negotiate greater visibility into AI usage and spending (32%) as they were to implement spending caps (16%). Organizations are also building flexibility into contracts: 39% added exit or transition clauses, while 36% shortened contract terms—signs of reluctance to make long-term commitments while pricing models and technologies continue to evolve.

From chatbots to orchestration

The governance challenges in procurement mirror operational challenges in customer service. Dan Belanger, president of NiCE Americas, said customer conversations have changed significantly over the past eighteen months.

"Eighteen months ago, most customer conversations were anchored in pilots focused on point use cases like chatbots or agent assist," Belanger said. Now discussions focus on how AI fits into customer service operations and broader business workflows.

Organizations are moving beyond isolated deployments toward what NiCE calls "end-to-end orchestration"—linking customer intent, workflows, and resolution in a single system. Customer service teams want to know if AI can authenticate a customer, update an order, process a refund, or schedule a service appointment without creating additional work elsewhere.

Fabletics, a NiCE customer, moved beyond traditional rule-based bots that could answer questions but not complete tasks. The company's AI deployment now supports customer authentication, order management, and other workflows that allow interactions to move from conversation to resolution.

"We didn't start this project to add AI to the contact center," said Jack Roberts, senior global director of GMS technology and applications at Fabletics. "We started it to give our customers faster, more flexible service, and to see what happens when AI can actually make decisions in real customer interactions."

Why it matters

The shift from pilots to production reveals a maturity gap that many organizations underestimated. The technical proof of concept—demonstrating that AI can perform a task—is fundamentally different from the operational proof of concept: demonstrating that an organization can manage AI costs, measure outcomes, assign accountability, and integrate automation into existing processes without creating new failure points. Organizations that treated pilots as tests of both technology and operating model are now ahead of those that focused solely on the former.

Testing the operating model, not just the technology

Garber believes the organizations making the most progress tested more than the technology itself during pilot phases. "The leaders who are getting this right treated the pilot as a proof of concept for the operating model, not just the technology," he said.

That means building processes for cost tracking, usage monitoring, and outcome measurement before AI deployments scale. It means deciding in advance who is responsible when automation falls short. And it means understanding how AI connects to existing workflows before those connections create dependencies that are difficult to unwind.

These details were first reported by No Jitter.