Featuring: Moshe Pridan, Chief Product and Information Officer, SiriusXM, Jerome Kapelus, Chief Executive Officer, Afiniti
Most contact centers have moved past the question of whether to use AI. They have the tools. What they don’t have, in a lot of cases, is the coordination to make those tools work together toward an actual outcome. That’s the gap SiriusXM Chief Product and Information Officer Moshe Pridan and Afiniti CEO Jerome Kapelus sat down to talk through on stage at Customer Contact Week Las Vegas 2026, six-plus years into a partnership that’s had to answer that question in practice, not theory.
Evaluate From the Inside Out, Not the Vendor In
With hundreds of AI vendors competing for a seat at the table, Pridan’s evaluation process starts nowhere near the vendors.
“You’ve got to know from a strategic perspective what your goals are and what you’re trying to accomplish with this amazing technology… We stay very close to our business goals and keep thinking about how AI can be used as a tool.“
Chasing a technology because a competitor uses it, or because a vendor happened to knock on the door first, is, in his words, “a very easy mistake.“ Instead, Pridan spends much of his time with early-stage founders and VCs – looking for the companies that aren’tready to pitch him yet, rather than reacting to the ones that are.
When It Works, and When It Doesn’t
Pridan pointed to SiriusXM’s multi-year design partnership on conversational AI for customer service as a case where the technology delivered on every measure that matters: resolution rates, containment, and, the metric he’s proudest of, customer effort, which came in at less than half of what a human interaction requires and less than a third of web self-service.
He was just as direct about where AI hasn’t worked. An earlier attempt at predictive, next-best-action decisioning didn’t fail because of the model. It failed because of the organization around it.
“It hasn’t been a failure, but it hasn’t been a glowing success either… It had nothing to do with the technology, but more about us as an organization, the change management, or the lack of change management with implementing a technology that essentially makes decisions on your behalf.“
The lesson: mature operations can be the hardest place to introduce a non-deterministic system, precisely because everything has always been measurable and predictable before.
Six-Plus Years of Being Judged on the Same Metric
SiriusXM’s relationship with Afiniti goes back more than six years, and Pridan was blunt about why it’s lasted. He described a “triangle” of misaligned incentives that plagues most vendor relationships, the software provider paid by seat, the systems integrator paid by the hour, and the customer left holding the goal of getting the implementation done on time and on budget.
“You get paid only if you deliver. Moreover, you make it very easy to isolate the single effect of Afiniti in an ecosystem that is completely full of noise… We can tell, day one, but to this day, what is exactly the benefit that is uniquely attributed to your technology and how it’s implemented.“
That’s the foundation Kapelus built on in describing where Afiniti is headed next: an orchestration layer that sits above any given vendor’s stack, agnostic to CCaaS platform, on-prem or cloud, and focused on managing outcomes rather than any single interaction.
The Contact Center Needs a Traffic Cop
The near-term future isn’t one AI agent per contact center, it’s many. Agents built by CCaaS partners, agents built in-house, agents from best-of-breed point solutions, each good at a narrow task. Pridan was clear about what happens without something coordinating them:
“You risk complete chaos unless you have an orchestration layer and decisioning layer that is tied to company goals, that has the ability to pivot on a dime… Am I optimizing for revenue? Am I optimizing for subscribers? …That’s where orchestration comes in.”
Kapelus tied that directly back to Afiniti’s twenty years sitting at the routing layer: “We understand the plumbing of the contact center better than most companies… You need a traffic cop to manage the agents as well.”
Building the Contact Center From a Blank Sheet of Paper
Asked what the relationship looks like five years out, Pridan pushed back on the idea that AI simply replaces human agents. The more useful frame, he argued, is designing the contact center of the future from scratch, squads of human and AI agents built to work together from day one, rather than bolting new technology onto old workflows.
“If you go about it the very traditional way, then you’re going to end up teaching this super, super smart agent to do the same things you’ve always done before. You’re bringing all of your old biases into the new models… That is usually destined to fail.“
His prescription: stand up small innovation pods outside the core process, prove measurable value, and let the results, not the mandate, move volume from the old system to the new one.
Why Outcomes Win
When asked what single metric matters most to his own board, Pridan didn’t hesitate: customer satisfaction. Not containment, not deflection, not average handle time.
“Better containment doesn’t help you if you didn’t really resolve the customer’s need on that same call. If they have to call back again, or even worse, they leave you as a customer, that is a really bad outcome… Our philosophy on the technology side, but also on care operations, is doing the right thing for the customer, everything else will follow.“
Kapelus picked up on it immediately, calling back to a line he’d heard earlier that captured the whole conversation:
“None of us wants to be contained or deflected. We want to be served.”
He added: “There are definitely folks out there who are getting drumbeats from the C-suite and from the board to drive efficiency. But efficiency at the cost of customer satisfaction, at the cost of lifetime value, it’s a fool’s errand.“
Key Takeaways
- Evaluate from strategy, not from the vendor pipeline. Start with the business goal, then find the technology, not the other way around.
- AI failures are usually organizational, not technical. Change management determines whether a non-deterministic technology succeeds inside a mature operation.
- Outcome-based, auditable value is a durable differentiator. A six-plus year relationship was built on being able to isolate exactly what a partner delivered, not just what it promised.
- Multiple agents need a decisioning layer above them. As agent sprawl grows, CCaaS-native, in-house, and best-of-breed, something has to coordinate them against shifting business goals.
- Customer satisfaction outranks activity metrics. Containment and deflection are only good outcomes if the customer’s need actually gets resolved.
Want to see how an outcomes-based decisioning layer could work across your own contact center stack?
Book a demo with our team today.
See the full session here.