The boardroom mandate: 91% are told to deploy AI — the phone call is the hard part
Gartner: 91% of CX leaders are under board pressure to deploy AI. The mandate is easy; the phone call isn't. Pilots die in production — telephony, voicemail, barge-in, handoff — not in the model.
According to Gartner, 91% of CX leaders are under board pressure to deploy AI.
So the mandate is everywhere. "Deploy AI" gets written into a slide, handed down, and added to someone's objectives. The trouble is that "deploy AI" and "deploy an agent that survives a real phone line" are not the same job — and the gap between them is where most projects quietly die.
The mandate is easy. The phone call isn't. Pilots die in production — telephony, voicemail, barge-in, handoff — not in the model.
The mandate is the easy part
A board directive to "deploy AI" creates urgency, budget, and a deadline. What it doesn't create is a phone agent that works. The directive treats the model as the product — pick a good LLM, write a prompt, ship it. But on a real call, the model is the easy 10%.
Where pilots actually die
The 90% is everything that never shows up in a polished demo:
- Telephony / SIP — real networks add jitter and delay you don't control.
- Voicemail — about half of outbound calls hit a machine; the agent has to know.
- Barge-in — callers interrupt, and an agent that talks over them feels broken.
- Handoff — when the agent should pass to a human, it has to do it cleanly, with context.
- Latency under load — the difference between a human-band reply and a robotic one, on every turn, at P95.
A pilot passes because it dodges all of these in a controlled setting. Production runs straight into them on call one.
How to actually deliver the mandate
Flip the project around: make the production layer the work, not an afterthought. Before you scale, prove the agent handles real telephony, real interruptions, real voicemail, real handoff, and real network variance. The board asked for AI; what earns the renewal is an agent that survives the call.
For the specific failure modes, see answering machine detection: the silent killer of outbound AI, barge-in: the feature that separates a demo from a product, and the 700ms wall.
Build on a platform designed for real phone calls. Free to start — $5 in credits, no card.
Read next
- McKinsey's State of AI: voice agents are the highest-value use case — the data behind why scaling, not piloting, is where the value is.
Frequently asked
Q.Why do voice AI pilots fail in production?
They rarely fail on the model. They fail on the unglamorous production layer: telephony and SIP variability, voicemail detection on outbound, barge-in when callers interrupt, and clean handoff to a human. A pilot that demos perfectly in a controlled setting hits all of these the moment it's on a real phone line — which is where most enterprise voice AI projects stall.
Q.What does the Gartner 91% figure mean for CX leaders?
It reflects that the vast majority of CX leaders are under board-level pressure to deploy AI. The directive — 'deploy AI' — is easy to issue. The actual job is deploying an agent that survives a real phone line: handles interruptions, detects voicemail, hands off gracefully, and stays in the human latency band. The mandate and the engineering are very different problems.
Q.How do you move a voice AI project from pilot to production?
Treat the production layer as the project, not an afterthought. Validate telephony/SIP behavior, voicemail detection, barge-in, latency under real network jitter, and human handoff before you scale. The model choice is the easy 10%; the surviving-a-real-call part is the 90%.
Q.What makes Call2Me production-ready for phone calls?
Call2Me is built around the production layer — sub-500ms latency targets, barge-in, voicemail detection, SIP/telephony, and human handoff — so an agent behaves on a real line, not just in a demo. Start free with $5 in credits.
Keep reading
All posts
Voice AIMcKinsey's State of AI: voice agents are the highest-value use case — if you can get past the pilot
McKinsey's State of AI report: 88% of organizations use AI but only about a third have scaled it, and customer-service and conversational interfaces are among the highest-value uses. Here's why most voice-AI pilots stall — and what 'in production' actually takes.
Jul 6, 20264 min
Voice AIWhy smart teams are moving from chatbots to voice agents
Text chatbots made customers wait and type. When a reply takes more than 3 seconds, a large share of people drop off. Voice agents answer in the human band — here's why teams are making the switch.
May 20, 20262 min
AI ReceptionistHow much does an AI receptionist cost? (2026)
An AI receptionist runs about $0.10–$0.15 per minute — roughly $1–2 per answered call — with no setup fee and no per-call human charge. Here's the honest cost breakdown vs a human answering service.
Jul 3, 20262 min