This article was first published on 9th January 2026: Autonomy without accountability: The real AI risk – AI News
An interview with Jason Roos, CEO of Cirrus
Contact centre leaders have heard no shortage of bold promises about AI. Faster resolution. Lower costs. Happier customers. Better agents.
Yet despite widespread adoption, many organisations are struggling to turn that promise into everyday performance. According to Jason Roos, CEO of Cirrus, the issue is not capability. It is readiness.
In a conversation with Contact Centre Helper, Roos shared why so many AI programmes stall, what readiness really looks like on the contact centre floor, and why the biggest barriers to success are rarely technological.
AI Has Become Normal. Readiness Has Not
AI is no longer experimental in contact centres. McKinsey’s latest State of AI research shows that nearly nine in ten organisations now use AI in at least one business function. But most of that activity remains stuck in pilots, with limited enterprise-level impact.
“The industry has spent the last two years buying capability,” Roos explains. “Very little time has been spent making that capability usable day to day.”
This gap shows up clearly in operations. AI tools are present, dashboards are full, but the way work gets done has not changed. Insights arrive too late, teams lack confidence in decisions, and the same problems repeat week after week.
“The technology is there,” says Roos. “It just isn’t embedded into the work.”
When AI Arrives Faster Than Operations Can Absorb It
Many contact centres are trying to layer AI onto workflows that are already under strain. Benchmark data from Call Centre Helper highlights the tension.
While more than a quarter of centres rank AI as their top priority for 2026, fewer than half use sentiment analysis, only a third provide real-time guidance to agents, and just one in five use AI to improve workflows.
“That tells you a lot,” Roos says. “The ambition is clear but the foundations are often weak.”
Without stable knowledge, consistent processes, and time to coach, AI tends to expose weaknesses rather than resolve them. Pilots succeed in controlled environments, then struggle when exposed to the realities of peak demand, compliance pressure, and operational complexity.
Why Pilots So Often Stall
According to Roos, pilots usually fail because they are protected.
“You choose a clean use case. You give it extra attention. You involve your strongest supervisors,” he says. “Then you try to scale, and the wider operation simply can’t carry it.”
Independent research supports this view. Studies suggest that the vast majority of generative AI initiatives fail to deliver their expected ROI, not because the technology breaks, but because it gets swallowed by the day job.
“The technology was never the hard part,” Roos adds. “It’s in knowledge being right, workflows being clear, people owning decisions, having time to coach, and being able to act quickly.”
What AI Readiness Really Looks Like
Rather than focusing on technology, Roos defines readiness through observable behaviours.
An AI-ready contact centre typically has:
- A dependable source of truth for knowledge, with clear ownership
- A small number of agreed best-known ways to handle common intents
- Decision pathways that do not rely on multiple approvals
- Coaching that happens while the work is fresh
- A steady rhythm of small operational improvements
“When those things are in place,” he says, “AI stops feeling like an initiative and starts feeling like how the contact centre runs.”
Designing for Service Moments
Roos also frames readiness through what he calls service moments. Every interaction, he explains, has three phases: before the contact, during the conversation, and after the interaction ends.
“Most centres focus on the ‘during’ because that’s where the pressure is,” he says. “But the biggest opportunity often sits before and after, where you prevent repeat work and close the loop.”
The data supports this. Many organisations have centralised data, but far fewer use it to guide agents in real time or improve workflows. The plumbing exists. The joints are weak.
Five Human Skills AI Can Support, But Not Replace
At the heart of readiness are five human skills: listening, thinking, speaking, doing, and improving.
AI can support each of these, but it cannot compensate for gaps in how they are practiced. Weak listening leads to missed intent. Weak thinking slows decisions. Weak speaking creates confusion. Weak execution piles up follow-ups. Weak improvement ensures problems return.
“That’s why readiness is a leadership issue,” Roos says. “Tools don’t fix those skills. Management does.”
Where Leaders Should Start
For leaders wondering where to begin, Roos advises against grand transformations.
“Pick a small number of outcomes that really matter in the next two quarters,” he says. “Then find the work that drives them.”
Reducing after-call work through better summaries is one example. It’s not glamorous, but it improves data quality, frees up time, and makes coaching easier.
“Space is the currency of change,” Roos adds.
Measuring Readiness on the Floor
Rather than relying solely on dashboards, Roos encourages leaders to measure signals the operation can feel:
- After-call work trending down
- Repeat contacts for top intents decreasing
- Faster time-to-fix for recurring issues
- Increased use of trusted knowledge
- More frequent coaching
- Shorter decision cycles for meaningful change
“If the numbers look good but the day still feels chaotic,” he says, “you’re measuring the wrong things.”
The Unsung Hero
Asked for his most unpopular opinion, Roos is characteristically direct.
“A lot of AI strategies are for show,” he says. “Readiness doesn’t photograph well. It doesn’t make a flashy slide. But it does make AI pay back.”
His advice for 2026 is simple. Decide what really matters, fix the foundations that block adoption, and sequence AI into reality rather than into a vendor roadmap.
“When service is treated like a system,” Roos concludes, “AI stops being a side project. It becomes the unsung hero you can feel in queues, quality, and the mood of the team.”
