Blog.

Global spend on contact centre voice and chat testing software is approximately $135 million in 2026, against an addressable market of $400-900 million. A bottom-up model built from statutory filings, funding disclosures and verified growth data — and an explanation of why the widely-repeated $6.5 billion figure is not real.

We can now test whether a voice agent sounds right, holds a conversation, survives four hundred concurrent callers and discloses that it's an AI. What nobody can currently evidence is what happens between the model and the system of record. That gap isn't a model problem or a prompt problem. It's a missing integration layer, and it's why security review keeps rejecting perfectly good agents.

CMP Research's Nicole Kyle has the data on what CX leaders are actually worried about in 2026. Fragmented analytics, self-service that doesn't resolve, and an AI workforce transition nobody is managing. All three are real. All three are also downstream of a question almost nobody funds properly, which is whether the journey works when a customer calls it.

NoJitter's orchestration piece nails the ownership mess - 68% of agentic AI decisions are made by business leaders, 4% by a centre of excellence. Fair. But the governance debate is all about the inputs, and skips the one output your customer actually hits: does the voice agent on your number work?

The 2026 Gartner Magic Quadrant for Conversational AI landed to a chorus of "who signed off on this?" Fair. But the vendor-ranking argument distracts from the thing that actually moves your risk - these tools now present as human, and the regulatory obligation for that lands on you, not on whoever's in the top-right box.

Omilia's Claudio Rodrigues argues your contact centre AI should be a dynamic asset that learns continuously, not a static one frozen at go-live. He's right. But a system that changes itself between releases breaks the assumption every voice QA process is built on - that you test when something ships.

A complete voice bot QA checklist covers twelve checks across three layers - conversational behaviour, telephony health, and regulatory compliance. Most teams have a partial version of layer one. Layers two and three are where production failures actually live.

Voice bot testing costs $0.07 or £0.07 to $0.50 or £0.50 per call minute on a self-serve platform like TotalPath. A typical agentic IVR with daily regression and quarterly load testing runs at $50 or £50 to $500 or £500 a month, versus a five-figure annual contract with legacy enterprise vendors. Here's the breakdown.

Loading testing an agentic IVR means simulating concurrent real phone calls - not synthetic SIP traffic - at expected production volume, and measuring time-to-greet, ASR confidence, end-to-end turn latency, and journey completion rate. Here's the four-step playbook.

The four credible Cyara alternatives in 2026 are Nopaque TotalPath, Hammer (Infovista), Klearcom, and Cekura. We compare them on pricing, agentic voice bot support, load testing, and time-to-first-test.

A real compliance audit of a real UK phone number. Eighteen tests, eight passed, ten failed. Every verdict is backed by a quoted line from the transcript and the rule it speaks to.

The classic testing pyramid was designed for software with clean APIs. Telephony doesn't work that way. Here is how we map customer-perspective voice testing onto the SDLC, and why the most valuable tests sit at the top, not the bottom.

Orsom & Nopaque, two organisations that bring beneficial experience and tooling to any CX transformation effort.

A first-pass observability sweep across publicly available AI Receptionist demos using TotalPath's Generic Explorer. Latency, ASR drop-outs, an unresolved template variable, and one bot that holds it all together.