The easiest part of an enterprise AI project is often getting the model to work. Proof of concept can be built in weeks, and a demonstration can generate excitement across an organization. Those early results frequently look promising enough to justify additional investment. However, many projects never progress much further.
Despite years of investment and enthusiasm surrounding artificial intelligence, organizations are still struggling with turning pilots into measurable business outcomes. But it’s not usually because the technology itself fails. More often, companies discover that building an AI model and operating AI at scale are two very different problems.
Why Most AI Pilots Never Reach Production
That has become important as companies rush to deploy AI across business functions. According to Jason Kurtz, CEO of Basware, 61% of organizations are rolling out AI agents experimentally, while studies indicate that the overwhelming majority of AI pilots never accelerate revenue growth.
The problem begins with how success is defined.
“If you cannot name the metric you are trying to move, you are running a science project,” Kurtz said. “And science projects do not get budget twice.”
The companies that successfully scale AI tend to focus on operational outcomes first. In finance, those outcomes might include touchless invoice rates, faster exception resolution, improved working capital management, or lower processing costs. The technology serves the objective rather than becoming the objective itself.
The Model Isn’t the Hard Part
That lesson has become increasingly relevant as organizations discover that the biggest obstacles often emerge after a pilot succeeds. Many executives still assume the model is the hardest part of the equation, but in practice, the challenge is often elsewhere.
“They think the hard part is the model. It is not,” Kevin Kamau said, Director of Product Management at Basware. “Models are commoditizing fast. The hard part, and the durable advantage, is the proprietary data, the workflow integration, and the trust infrastructure around the model. Most of the value and most of the difficulty live there.”
Where AI Fits in Accounts Payable
Those challenges become particularly visible in enterprise finance environments, where AI systems are expected to operate inside existing workflows rather than alongside them.
Accounts payable provides a useful example. Organizations have spent years automating invoice processing through rules, workflows, and business logic. Many have already achieved high levels of automation for straightforward transactions. But the next phase is not simply more automation. It is invoice lifecycle management: governing every invoice end to end with the controls needed for compliance, financial integrity, and enterprise visibility.
Kamau describes it as the long tail of exceptions where manual effort, delays, and operational risk tend to concentrate.
“You can get to 80% plus touchless processing for well-structured invoice flows without AI,” Kamau said. “But there’s always a stubborn long tail of exceptions and complexity where rules alone break down. That’s where manual work, delays and risk concentrate, and where AI can unlock the next step change.”
Basware’s experience in that area is informed by scale. The company, which focuses on invoice lifecycle management, says its platform has been trained on more than 2.5 billion invoices and now supports more than one billion AI actions annually across its customer base.
From Periodic Checks to Continuous Compliance
That volume creates opportunities that extend beyond simple automation.
According to Anssi Ruokonen, VP of Data & AI at Basware, many finance processes historically operated in monthly cycles, not because that cadence was ideal, but because deeper analysis required too much time and effort to perform continuously.
“Agentic AI will change this by making the execution of those tasks less costly,” Ruokonen said. “As operational tasks become less of a burden, the focus shifts onto more strategic tasks. Compliance and audit checks are a textbook case. Moving from a periodic scramble to something checked continuously is exactly what we mean by Continuous Compliance.”
That same logic applies to financial integrity. As AI moves deeper into finance workflows, organizations need confidence that each transaction is accurate, compliant, validated, and protected from risk before money moves.
The Hidden Cost of Moving From Pilot to Production
Even when organizations identify the right use cases, another challenge often emerges. The economics of production AI can look very different from the economics of a pilot.
A proof of concept typically operates on a limited dataset, with a small group of users and relatively clean conditions. Production environments introduce additional requirements, including integration, monitoring, oversight, orchestration, and failure recovery.
“The pilot runs on a few users and clean data, and the business case gets built on pilot economics,” Kamau said. “Production then adds integration, monitoring, oversight, and consumption-based running cost, and the true figure can be several times the estimate.”
As organizations move beyond experimentation, they are reassessing what it takes to bring AI into production. The build-versus-buy question is increasingly less about building the intelligence itself and more about the cost and complexity of operating it, including the governance, controls, integrations, and oversight required to support business-critical processes.
That reality has become increasingly important as organizations move toward agentic AI systems capable of performing work rather than simply generating recommendations.
The deeper AI becomes embedded inside operational processes, the greater the need for visibility into both performance and cost.
“There is a cost dimension here that most people miss,” Kamau said. “Risk and cost both scale with how deeply you embed the AI.”
What Separates Companies That Scale AI
The companies that successfully move beyond experimentation tend to recognize this early. They invest in data quality. They redesign workflows instead of simply layering AI on top of existing processes. They establish ownership and accountability. Most importantly, they focus on solving business problems rather than chasing technology trends.
Jason Kurtz, Chief Executive Officer of Basware, believes too many organizations are still approaching AI from the opposite direction.
“The real risk in the AI investment story isn’t that companies are spending too little,” Kurtz said. “It’s that they’re spending in the wrong places.”
That perspective reflects a broader shift happening across enterprise AI. The conversation is gradually moving away from what models can do and toward what organizations can operationalize. In finance, that shift is from automating isolated tasks to governing the invoice lifecycle end to end.
The companies generating measurable returns are rarely the ones running the most experiments. More often, they are the ones that identify a specific problem, build the processes needed to support AI in production, and then scale from there.
“Execution beats experimentation,” Kurtz concluded.
As AI adoption continues to accelerate, that may prove to be the difference between organizations that showcase successful pilots and organizations that build lasting operational advantages.
Photo by BoliviaInteligente: Unsplash
Marcus Whitfield writes about developer tools, programming languages, and the software trends shaping how engineers build. Before joining DevX, he spent five years as a full-stack developer and two more running a small dev-tools newsletter that topped 10,000 subscribers.
