i2c Executive: Before AI Agents Go Live, Category-by-Category Validation Matters More Than Knowledge Base Integration

Pam Guinn, Vice President of Operations at payment processor i2c, wrote in the PYMNTS ebook "Payments Leaders Rethink What Works in 2026," published on November 6, 2026, that when the company deployed agentic AI for customer service, it found that connecting AI to existing knowledge bases is only the starting point—the real barrier lies in operational validation and readiness.

Guinn said testing exposed gaps between documented processes and actual calls: cardholder disputed transactions, customer requests for program exceptions, and calls spanning multiple processes all require judgment built over years of experience—capability that previously existed largely as institutional knowledge rather than written procedures. i2c therefore organized this experience into actionable decision frameworks and clarified which steps are suitable for automation and which still require human intervention.

The deployment approach was adjusted accordingly, moving from a broad rollout to a call-type-by-call-type progression: before each call type is approved for the production environment, it must go through programming, testing, auditing, and validation to confirm its accuracy and compliance. Guinn said this approach requires higher upfront investment, but each validated call type becomes a reusable building block, reducing the risk of subsequent launches and accelerating expansion. She advised executives planning agentic AI projects to invest in validation and operational readiness on par with system integration.

Source: https://www.pymnts.com/opinion/2026/why-data-access-isnt-enough-for-ai-readiness/