Why AI adoption stalls

Adding artificial intelligence (AI) to workflows is on many businesses’ to-do list for 2026. According to McKinsey & Co.’s State of AI in 2025 survey, 88% of respondents report their business is experimenting or piloting AI, using it in at least one business function. However, nearly two-thirds of respondents say their organizations haven’t begun scaling AI across the enterprise. What’s keeping them from moving to full-scale implementation?

Jason Rosenfeld, chief growth & alliances officer at NewRocket, a Service Now AI enterprise workflow platform partner, explains why so many enterprise AI initiatives stalled in 2025 and what organizations are doing differently heading into 2026.

He clarifies a few terms: Agentic AI is intelligent, task-oriented agents working together behind the scenes to automate processes and reveal knowledge to deliver faster, more accurate outcomes. A first step toward fully scaled AI adoption is the pilot – a time-constrained, limited-scope deployment of a narrow task, such as forecasting demand, summarizing maintenance logs, or detecting production anomalies. A co-pilot embeds AI into a workflow to assist in real time – offering best actions, document generation, scheduling, or helping with root-cause analysis. Fully scaled AI adoption operates across processes to automate decisions within defined guardrails and escalate exceptions for human review.

“After years of AI pilots, copilots, and workflow automation, many enterprises are realizing the technology itself isn’t the limiting factor,” Rosenfeld explains. “The bigger challenge is service models, operating assumptions, and partner strategies haven’t evolved at the same pace. As a result, AI initiatives often stall before delivering measurable, enterprise-wide impact.”

The human factor can slow AI adoption. “Many enterprises used AI to generate better answers but still relied on humans to coordinate the work: moving tickets, reconciling master data, chasing approvals, translating recommendations into ERP [enterprise resource planning] transactions, and documenting decisions for audit,” Rosenfeld says. “In manufacturing, humans also serve as the glue between IT systems and physical execution on the floor. If AI can’t trigger actions in a governed way, then every ‘insight’ still turns into a manual project. That dependency becomes the bottleneck at scale.”

To move beyond pilots to embed AI into everyday decision making requires organizations to improve their enterprise data and standardize and simplify the workflows around decisions before trying to automate them. “They’re also investing in governance patterns that make autonomy safer: human-in-the-loop where it matters, and straight-through processing where it doesn’t,” Rosenfeld says. “In manufacturing and ERP, that includes improving master data discipline, defining process ownership across facilities, and creating a closed-loop measurement approach so AI is judged on outcomes, not demos.” In this way AI is being designed as part of the operating model, not as a separate improvement.

“Companies that scale AI don’t win because they find a smarter model. They win because they ensure their underlying enterprise data is clean and redesign how decisions get executed across real workflows, with governance that makes speed and safety compatible,” Rosenfeld concludes. – Eric