Inside AI Day 2026: Five Sessions, One Vision for Voice AI Agents

AI Day 2026

Five sessions, five takeaways

Voices of trust 

A customer's perspective: self-service 

TruConnect uses Five9 AI Agents to automate a sensitive process, government benefits support, while pushing toward 50% containment. The session digs into the two questions any leader in that position asks first, “How do you keep an autonomous agent accurate and on-script for high-stakes interactions?” And “How do you get customers to actually use self-service instead of defaulting to a live agent?” 

Claudia Guerrero put the adoption question to rest simply: customers don't care whether it's an AI; they care whether it works. If self-service means troubleshooting a device or buying a data top-up in a few minutes instead of waiting on hold, customers use it.  

The bigger surprise for TruConnect was how much authentication mattered. Once customers got through that step, they generally finished what they came to do. So instead of adding more features, the team spent more time making the first part of the journey easier. As Sarika Prasad, Director, Product Marketing at Five9, summed it up on stage, adoption isn't about whether it's an AI agent or a human agent. It's about whether the issue gets resolved quickly. 

The result: Five9 Agent Assist helped TruConnect reduce after-call work by 40% and improve average handle time by 9%, from 542 to 494 seconds. What made this stick? Claudia traced it back to the agents themselves. Agent Assist wasn't another step in their workflow. It took work off their plate by handling notes and call summaries for them. TruConnect brought supervisors into the process, listened to calls, gathered agent feedback and kept refining the experience. Once agents trusted the summaries, taking notes gave way to focusing on the customer, and it became a natural part of how they worked. 

A customer's perspective: quality at scale 

Advia Credit Union serves nearly 200,000 members across Michigan, Wisconsin and Illinois, with about $4 billion in assets, putting it in the top 3% of credit unions nationwide. It's also the best reference for what Agentic QM surfaces once you stop sampling and start seeing everything. 

Before Agentic QM, quality management at Advia looked like it does at most contact centers. A quality specialist and the leadership team manually reviewed calls, constrained by time to a small slice of interactions. Cassandra Murray, Manager of Operations Systems & Projects at Advia, put a number on that slice. About 1%. That meant coaching decisions and trend calls were based on a tiny fraction of what was actually happening, with no way to know if an issue was one bad call or a pattern repeating across a thousand conversations. 

Agentic QM changed the math. Advia now evaluates 97% of its call volume, and the shift isn't just about coverage. It's about the kind of question a supervisor can ask. 

Instead of, "Did we happen to review the right call?", it's "What pattern is showing up across all of them?" 

Coaching opportunities surface automatically with evidence attached, synced to transcript and audio, instead of waiting on a manual find. That hands supervisors back the time they used to spend searching for trends so they can spend it developing people instead. The result Cassandra didn't fully expect: agents get real-time visibility into their own quality, so they can adjust mid-conversation instead of waiting on a review that comes back weeks later. 

What closes the loop is where the findings go next. Agentic QM feeds what it learns back into routing and agent assist, so a pattern caught on Tuesday changes how the next call is handled rather than sitting in a report. 

Cassandra's own summary is the cleanest way to say it. Before Agentic QM, Advia managed quality through sampling. Now, they manage it through visibility. A shift from hindsight to insight, and from a compliance exercise to the kind of continuous feedback loop quality management was always supposed to be. 

A partner's perspective 

CDW's Ken Drazin, Director of Employee & Customer Experience, helps customers stand up Five9 AI Agents, Five9 AI Agent Assist and Five9 AI Insights across all kinds of environments, and has seen enough rollouts up close to identify the pattern. What separates deployments that reach full production and real outcomes from the ones stuck at proof of concept: clear goals tied to a real business problem, strong data with clear ownership and change management, and starting focused before proving value and expanding. The most common mistake between proof of concept and full deployment, in CDW's experience, is treating the proof of concept like a demo instead of planning for it to become production. 

An analyst's perspective 

For analyst Derek Top, the real test of any AI Agents rollout is whether it turns into results a company can see and measure. That means putting the right tracking in place from the start: how often issues get resolved, how many calls get transferred, how long resolution takes, customer satisfaction and cost. 

The numbers behind that point are worth sitting with. Only 39% of organizations can today point to measurable business impact from CX AI. 85% of companies say fewer than 40% of AI pilots ever reach production, but once a pilot does reach production, only 16.6% get discontinued . The wall sits before production, not after it. And for the companies that get through, 83% realize ROI on AI projects in 12 months or less. The technology showed up. The trust took longer. 

Four vantage points, one conclusion. Two customers with very different workloads, a partner who's watched hundreds of rollouts and an analyst who studies what moves the needle all land in the same place. The gap between AI pilots and AI that pays off isn't a technology problem. It's a trust problem, and trust gets built in specifics. 

Four of those specifics ran through every session: 

One platform for every CX interaction. AI agents, agent assist and conversational intelligence on shared data and orchestration, not point tools stitched together. When a request reaches the edge of what an AI Agent is licensed to do, the full history transfers to Five9 AI Agent Assist. The customer never repeats themselves. What the human resolves makes the system sharper next time. This humantic data is the missing data layer for agentic AI. 

1. Self-service that not only responds but also resolves. Many contact centers already utilize automation, but the key question is how effectively it can be completed. Five9 Voice AI Agents engage in real conversations, manage multi-step requests that may change in intent and take action across various business systems. This is how TruConnect aims to achieve 50% containment on regulated workflows. 
2. Data and guardrails, not as an afterthought. Fragmented data is the top cause of AI project failure, so Five9 meets data where it already lives with one knowledge and RAG layer shared across AI Agents, IVA, and Agent Assist. Real-time guardrails, entity redaction, least-privilege permissions, a full audit trail and continuous evaluation. SOC 2 Type 2, ISO 27001, PCI DSS Level 1, HIPAA, GDPR. Autonomy grows at the pace trust does. 
3. Quality management for the agentic era. Traditional QM reviews a sliver and calls it a trend. Five9 Agentic QM evaluates up to 100% of interactions, flags coaching moments with evidence attached and feeds what it learns back into routing and Agent Assist. Advia went from 1% to 97%. The executive panel called it the fastest ROI in CX today. 
4. Point solutions automate a task. It takes one platform, built on unified data, real guardrails and a habit of measuring instead of assuming, to earn the trust that lets AI scale past the pilot. Build the trust first, and the results follow. That's Smarter CX with every interaction; one resolved call at a time. 

Want to learn more?

All five sessions are available on demand. Watch them here.