March 11, 2026
NiCE Puts Agentic AI into Action Across CXone and Cognigy
For many enterprises, the AI promise has run ahead of reality. Research shows that while 98 percent of contact centres use AI, 61 percent report handling harder conversations as a result, a sign that deployment alone does not guarantee outcomes. NiCE is targeting that gap directly, with two agentic AI innovations that turn enterprise interaction data into ready-to-deploy AI agents, transforming critical context into measurable results.
From Interaction Data to Deployable Agents
Launched at Enterprise Connect, an enterprise communications conference in North America, the CXone capability assesses structured and unstructured data spanning voice, chat, digital channels, workflows, and human interactions. Rather than stopping at insights, it identifies where automation would have the greatest business impact and goes straight to building and deploying agents to act on it, all within governed guardrails.
What once demanded specialist analysis, iterative testing, and months of validation can now move from opportunity to live deployment in hours. ROI is quantified before a single agent goes live, and every deployment is tracked against those projections once it does.
The system draws on billions of customer interactions processed annually across the CXone platform, continuously refining itself against the patterns of top-performing human agents and measuring whether live results match the promised returns. NiCE’s own research, which draws on large enterprises handling more than one million interactions per year, found containment rates above 80 percent and CSAT improvements of up to 20 percent. This is an early indication of what the closed-loop approach can deliver at scale.
According to Robin Gareiss, CEO and Principal Analyst at Metrigy, 82.4 percent of companies see value in a unified platform for CX and AI capabilities. Gareiss continues: “By connecting enterprise data directly to deployment within a unified platform, NiCE’s closed-loop approach enables enterprises to scale AI with confidence.”
A Shared Direction at Cognigy
NiCE Cognigy made a parallel announcement the same day, at Nexus 2026, its global CX AI Summit. Its automation discovery capability mines engagement data across call and chat transcripts, routing signals and performance metrics to surface where agents would add most value, then generates them automatically.
The result is a closed loop that takes an enterprise from raw interaction data to a working AI agent without the usual cycle of manual analysis and protracted testing. Genesys is another company to be advancing agentic AI capable of completing tasks across multiple systems, but what makes NiCE’s approach different is that it identifies and quantifies the opportunity before a line of agent logic is written.
Philipp Heltewig, General Manager of NiCE Cognigy and Chief AI Officer, highlights the importance of moving beyond the testing phase: “Agentic AI is becoming the operating layer of the enterprise. The challenge is no longer experimentation. It is how to run AI with visibility, accountability, and measurable performance.”
The Nexus event also saw NiCE Cognigy launch additional innovations, including embedded multivariate testing, which enables controlled comparisons across prompts and routing logic before agents go live, plus enhanced conversation analytics powered by LLM-based evaluation. Both capabilities are designed to shift AI management from static quality assurance to continuous performance engineering. NiCE Cognigy also announced expanded integration with the Model Context Protocol (MCP) for secure interoperability with external AI tools.
Moving Beyond Experimentation
Both announcements point in the same direction. NiCE is positioning itself as the platform that turns AI ambition into execution, across both CXone and Cognigy. With Gartner predicting that agentic AI will autonomously resolve 80 percent of routine customer service interactions by 2029, there is pressure to start seeing AI-driven results, and the window for experimentation is closing.
