The first NiCE CXone–Watson AI integration — built for IBM's healthcare contact centers before native AI existed

THE CLIENT

IBM's healthcare contact center operation managed patient inquiries at significant scale — thousands of daily interactions spanning appointment scheduling, urgent medical coordination, and healthcare programme navigation. 

IBM set a clear strategic target: AI should handle 90% or more of interactions without human intervention, with Watson AI (IBM's own platform) as the intelligence layer. NiCE CXone was the CCaaS environment. The challenge was that at the time of the engagement, NiCE had no native AI capabilities whatsoever. There was no Mpower, no Enlighten, no Virtual Agent Hub. The integration IBM needed had never been built.

THE CHALLENGE

Three structural problems defined the engineering challenge. First, there was no integration path between Watson AI and NiCE CXone — every component would need to be built from scratch, without APIs, orchestration frameworks, or any prior implementation to reference. 

Second, IBM's healthcare interactions involved protected patient health information (PHI), requiring HIPAA compliance, end-to-end encryption, and custom authentication protocols that NiCE did not support natively. 

Third, the AI-to-agent handoff had to be completely invisible to callers. Internal data showed that 60% of patients would abandon a call if they perceived a system transfer, such as hearing a phone ring when being passed to an agent. 

WHAT WE BUILT

Primero Group built a custom AI orchestration layer inside NiCE — first of its kind. A custom AI-driven IVR classified calls, determined patient intent, and provided self-service resolution or escalation without any native platform support for this workflow. 

Watson processed chat, SMS, and eventually voice interactions; Primero Group experts manually coded approximately 80% of the AI training pipeline, iterating Watson's responses across thousands of interactions to reach the 90% deflection target. 

When Watson's near-instant response times felt unnaturally fast to agents in chat interactions, the team introduced randomized response timers to mimic the conversational rhythm of a human operator — a detail that mattered for agent adoption. 

End-to-end encryption for PHI was implemented via custom TLS SIP trunk configuration, custom hashing algorithms inside NiCE for authenticated data exchange, and secured REST API communication — capabilities NiCE did not support at the time and that required bespoke engineering. 

The AI-to-agent handoff was designed to produce no audible transition — no ring, no hold tone, no signal that the patient had moved between systems. The solution was deployed progressively: chat and SMS first in 2019, voice in 2020. 

IBM estimated they would have needed to double their workforce without this solution. During the COVID-19 pandemic, the system scaled to handle 125,000+ voice calls per hour at peak capacity across vaccination scheduling and testing coordination across multiple US states.

WHAT THIS SHOWS

The IBM engagement is the most consequential proof of capability in the Primero Group's CX engineering portfolio. It was a first-of-its-kind integration — built before NiCE's own AI product roadmap had produced the capabilities that now exist natively on the platform. 

The patterns established here (custom AI orchestration layers, seamless AI-to-agent handoffs, HIPAA-compliant contact centre engineering) directly informed how we approached every subsequent AI–CCaaS integration project.