The Skift Takeaway
SKIFT DATA + AI SUMMIT 2026
How Agentic AI Is Solving The Billion-Dollar Hold Queue
Nirmal Mukhi, Chief Architect, ASAPP
Moderated by Jeremy Kressmann
THE ARGUMENT
The question in AI-enabled customer support is no longer whether to automate — it’s where to draw the line. Mukhi argued the answer is architectural: keep the AI in control of the conversation, but route high-stakes decisions silently to a human in the background. That model, he said, is what allows companies to expand from simple policy questions to complex, high-stress transactions like rebooking a family during a winter storm. Trust is built incrementally, guardrails are removed as performance proves out, and the aperture of what AI can handle widens over time.
THE EVIDENCE
- An airline customer scaled up AI agents during a multiday winter storm event in early 2026 that shut down U.S. airports, saving approximately $69,000 in labor costs.
- One ASAPP customer reduced its repeat call rate for the same issue by 90% after deploying the platform.
- During a winter storm deployment, satisfaction scores on AI-handled conversations ran more than 8 points higher than on human-handled conversations.
- On policy questions, AI answered accurately at a rate 26 times higher than humans.
- ASAPP’s typical first deployment runs three weeks, after which customers independently add an average of five new use cases per month.
THE SO WHAT
If your customer support roadmap treats AI as a containment tool — something that deflects calls before handing off to a human — you’re leaving most of the value on the table. The model that’s producing much bigger ROI keeps AI in control end-to-end, with humans tapped silently for specific high-impact decisions, not handed the conversation.
Access all the Skift Takeaways from this event, as well as full session videos, presentation documents, and more editorial coverage at our Skift Data + AI Summit attendees hub.