When the world sneezes, aviation catches a cold.
Louise Phillips, Virgin Atlantic’s Vice President of Customer Centers has worked with the airline for 16 years and recounted the situation the airline found itself in when COVID caused the ultimate disruption. That meant fleets sitting idle on tarmac and phones ringing off the hook with people trying to rebook, refund or just understand the situation. At the time, Virgin Atlantic’s contact centers were running on a telephony platform that had been push-button, voice-only for 28 years. Three-quarters of the airline’s North America traffic was grounded overnight, giving the centers roughly 100,000 unresolved tickets and a customer base whose main question was whether they’d ever fly again. In a conversation at Genesys Xperience in Las Vegas, Phillips told me that this was the moment that the team decided it had to be bold. The archaic system had left the team with no way of seeing what customers were contacting the airline about, never mind respond to the demand.
Both Phillips and Darren Kelly, Virgin Atlantic’s Head of Customer Contact, spent time on the front line before progressing into leadership roles – Kelly as an agent, Philips joining as cabin crew before returning later to lead the customer centers. These experiences shaped their next decision. Rather than automating everything, they examined which interactions belonged with a human and which would be better suited to a channel designed for the task. Document-heavy work such as boarding passes, name changes and ticket reissues moved to WhatsApp, where handling time is about half what it takes on voice. Emotive scenarios such as cancelled honeymoons, lost luggage, or disruption to destinations that people were traveling to for reasons that mattered were prioritized for voice calls. As Phillips explained:
You can’t class a cancelled honeymoon as a transaction. If we lose your bag or delay your bag, that’s not a workflow. We are deliberately tactical about where we want to place human agents versus virtual agents.
Phillips recalls sitting around a table with Genesys in 2023 at a previous Xperience to talk about predictive routing:
We said, “yeah, lets just go home and switch it on.” And within a month, we were saving 15% on our handling time.
Some of the changes were structural. Where the airline’s contact centers had previously run across multiple suppliers for live chat, voice and social, everything now moved into a single Genesys platform. This resulted in a single knowledge repository that now serves web self-service, bots and human agents from one source, making a big difference to frontline staff who can take up to nine months to get fully up to speed. Genesys Copilot was turned on in 2024, and is designed to surface the right knowledge article to the agent in the moment, along with an internally-built concierge on the airline’s website which hands off to human agents with rich contextual data attached. The measurable outcomes include web messaging traffic up 220%, customer satisfaction up 25 points year over year, agent attrition down 40%, and 10% more customer conversations handled with the same resources.
It is easy to see why Genesys is keen for Virgin Atlantic to share it’s story, but it would be a disservice to assume that it’s just about the numbers. The decisions have definitely paid dividends but they were also made out of empathy – later in the conversation, Phillips emphasizes:
We are very passionate because we’ve come from the front line upwards over 16 years, and we remember how difficult it is to be on the front line.
What Phillips and Kelly have built at Virgin Atlantic is a practical example of what the industry now calls agentic orchestration.
Chaos is building – and the customer is waiting
Virgin Atlantic isn’t the only organization wrangling with this decision making challenge. At the embargoed media briefing earlier this week, Mike Szilagyi, Genesys’s Head of Product, described it as the AI paradox: the more AI an organization deploys, the harder it is to deliver a seamless customer experience. A slide in a session on turning intent into outcomes with case management, titled “Chaos is building and the customer is waiting” explained the situation clearly. Most enterprises now have dozens of AI workflows running in customer experience (AI agents, copilots, automation) and each of them, individually, creates value. Put together, they fragment the customer journey. Customers have to repeat themselves because no context is shared across back-office teams and systems. Work slows down because every handoff means a time lag. This leads to teams operating reactively and losing sight of the bigger picture. Which brings us onto where the context went.
CEO Tony Bates argues that most enterprises don’t have an AI problem, they have an operating model problem. In his view, companies have been using AI as a helper – bolting it on top of processes designed for human execution. This only works up to a point – if you want AI to actually take on the work, you have to redesign the way the business runs, not just add AI to the existing processes.
He sets out five things that have to change:
- Interactions have to connect with each other, so that every conversation feeds the next one, rather than starting over every time.
- Your workforce has to manage a blended team of AI agents and people, not two separate ones.
- Stop measuring how fast someone answered the phone – start measuring what it cost you to win the customer and how much money you kept.
- Not all workflows can be scripted in advance. Sometimes AI has to be able to make its own decisions inside parameters you have set, what Bates calls “governed autonomy”.
- Transformation never ends, it’s the way you now operate – always changing and improving.
Putting a framework on a slide is one thing – building it into an enterprise is another. Bates is emphatic that organizations have to overhaul their entire operating model, and do it with empathy, trust, and orchestration. Plenty of vendors are using “orchestration” as the buzzword du jour. So how does Genesys propose to deliver it?
Plain English, please
Let’s start with trust. Note to vendors everywhere – this does not mean “the customer likes us.” A business should be able to let an AI system act on its behalf and know what the system did, why it made that decision, and what would have stopped it if needed. Trust in agentic AI is what you can observe, audit, and constrain – not what you can hope for. As Genesys CTO Glenn Nethercutt put it during the keynote:
You can’t govern what you cannot observe.
Empathy is trickier. It is one of the company’s stated values, and Bates has published a book called Empathy in Action. In this context, empathy is a design principle rather than a feeling. The system knows enough about the customer and about the moment they’re in, in order to route them somewhere appropriate to both. Bates contextualized the point with an example:
My mother is 83 years old. She’s not very digitally savvy, and if she has to change the password on her bank account, it’s a very scary moment for her. It’s not something that she wants to do through an automated self-service chatbot. She needs to speak to a human being.
Onto the third word, orchestration. As Szilagyi said: “There’s lots of businesses out there talking about agentic orchestration. But the nuance is for what.” In Genesys’s case, it’s from the moment a customer arrives with an intent to the moment their problem is actually resolved. That requires much more than one conversation, an AI agent or a single workflow can hold on its own. Nethercutt explained it during the keynote:
In AI, we’ve just built the vessel. The rest, the part that took humanity the longest, is what we’re building now.
The body of the architecture
Edward Calvesbert, VP of Product Management for AI and Digital, walked through the architecture. Navigator is the eyes and ears of the system – the LLM-based front door that works out what the customer is trying to do. Orchestrator is the brain – it takes the intent and builds a plan by reasoning across policies, business goals and the journey itself. Agentic Virtual Agents are the arms and legs, and they execute the plan by calling on tools and taking action within guardrails. Underneath all three is Contextual Intelligence which Elcenora Martinez, SVP of Product Management for Analytics and Journey Management, described as “more than a Customer Data Platform (CDP), and more than a Customer Relationship Management (CRM)” – the persistent enterprise memory that holds every interaction that came before.
The AI Control Plane sits across all four of these components and is the governance layer that decides who gets access to which AI tools, what the tools can do, how decisions are observed, and how they can be switched off if something goes wrong. Szilagyi described it as an industry-wide direction of travel rather than a Genesys invention, but the main point is that this is built into the architecture of the platform rather than layered on top. This is a design decision that Genesys says is the difference between AI you can operate at enterprise scale and AI that you can only pilot.
Beyond the buzzword
So what makes the Genesys version of orchestration structurally different from the raft of AI-first vendors chasing this market? A few things stood out to me. Navigator, Orchestrator and the Agentic Virtual Agents (AVAs) are architecturally separate. Navigator handles intent, Orchestrator handles the plan, and AVAs handle execution, rather than one general-purpose agent trying to do everything. While plenty of vendors are talking about agentic orchestration, Genesys is building it specifically for customer experience, from a customer’s intent through to the outcome that resolves it. That focus, Szilagyi argued, is what makes the difference between orchestration as a technology capability and orchestration as a business platform.
The other thing that stood out is the ecosystem work. Jack Nichols, VP of Product Management for ecosystem strategy, explained:
We know we’re not the center of the universe. We’re part of an ecosystem.
Essentially, that means Agent-to-Agent (A2A) interoperability – an open protocol that lets AI agents from different vendors coordinate work – is already in beta with ServiceNow, and in progress with Salesforce Agentforce. Model Context Protocol (MCP) support, which lets AI systems reach tools and data through a standard interface, comes via the Pinkfish acquisition and opens up over 25,000 MCP-compatible tools to Genesys customers. Nichols told a story from a customer breakfast that morning: a large financial services company was running Genesys AVAs, Sierra agents, and Salesforce agents in fragmented isolation – three different systems with no unified voice – and could now bring them into one journey through Genesys.
Scaling up
At a CEO roundtable, Bates spoke about what the AI Control Plane actually unlocks for customers: they stopped hallucinating. Early LLM deployments had well-publicized failures (we’ve all seen the examples of the airlines offering discounts they shouldn’t have and chatbots making up policies) – because the models had no governance wrapper around them. Once Genesys moved to large action models running inside a configured framework of what the AI could and couldn’t do, the failure mode changed. Bates argues this is the difference between AI as a pilot and AI as a production system:
Governance isn’t actually the brake that you’re putting on autonomy. It’s the license to scale it. It gives you the license to go faster, but you need it in place.
My take
Genesys is setting out its stall, and it’s bringing solid customer use cases to back up where it differentiates. I’ve picked up on five different customer examples this week and none of them feel like “we built a bunch of agents that can do this.” There’s more emphasis on intent and outcomes which makes sense given the customer base. I’m looking forward to speaking with more practitioners and executives this week and digging into the rationale behind the scenes – and finding examples of empathy, trust, and the line between agents and humans. It’s turning into an interesting event.
