Fresh off a World Cup where Conviva monitored tens of millions of concurrent streams across 20+ platforms, Conviva CEO Keith Zubchevich joined the Profluence Sports Podcast to talk about the future of live-streamed events.
His main point: the Quality of Experience (QoE) discipline that built streaming can’t stop at the stream. It has to extend to every agentic and interactive layer publishers are now building around it.
“If the stream buffers right when viewers are leaning in to see Messi make a penalty kick, it’s emotional. Publishers don’t want to deal with that,” said Zubchevich.
That same real-time discipline is why Conviva built its World Cup monitoring to work at the moment fans are watching, not after the fact. And this year (during Conviva’s fifth World Cup) that real-time monitoring and analysis happened with an AI agent, Nexa. Zubchevich calls this shift — using AI agents to do the analysis instead of humans — agentic analytics.
Now here’s where it gets interesting. Today, 83% of Conviva’s customers are using Nexa to analyze data about a viewer’s experience across their apps, websites, and streams. Now, they’re extending that to other ways fans are interacting with publishers, including via AI agents. Now, we’re talking about an agent (Nexa) analyzing another agent (think a content recommendation agent, a betting agent, or another fan experience).
“As people start talking to AI agents, publishers need to know if the agent is actually improving the customer experience. Are they happy, or is it miserable?” Answering that isn’t about whether the agent’s response was accurate (the viewer asked for basketball and I showed them a list of basketball content), it’s about measuring whether the experience was efficient, helpful, and personalized to the person it aimed to serve.
Personalization Has to Be Built on Behavior, Not Demographics
Zubchevich argues that most personalization stops at a category — a fan’s favorite team or sport — when the more useful signal is how that specific person actually watches.
“Ultimate personalization means you make it about Scott. You personalize to Keith’s preference. Now you’re looking at millions of different personalization requirements. A human can’t respond to that. So AI has to be the personalization engine.”
He describes the payoff as knowing a viewer well enough to act before they ask: “It’s Tuesday, and I know what Scott wants to watch on Tuesdays. Why? Because I have the historical data. I know it’s Scott and I can act on that.” That’s not a favorite-sport preference — it’s a sequence: maybe a fan checks in on First Take, drops into Fox Sports, then catches part of a studio show, all in a pattern specific to that person and that day of the week.
For publishers, that same behavioral lens doesn’t have to stop at content recommendations. The same viewer-level pattern data that predicts what someone wants to watch next is the foundation for understanding how engaged they are with the ads inside that experience, and how their in-app behavior outside the video player connects back to it. Treating those as one continuous behavioral picture, rather than three separate measurement problems, is what makes personalization actually specific to a person instead of a segment.
Structured, Time-Synced Data Is the Prerequisite for Reliable Agents
None of the above works if the underlying data is inconsistent, says Zubchevich.
“Bad data is never cleaned up by AI… You will spend so much money on getting an agent or an AI service deployed on bad data. You could have the highest-performing AI agent built, but if you give it 75% gas, 25% water, you’re screwed.”
For Conviva, the fix starts with something as basic as time: every stream, across every platform the company monitors, is processed against a single clock, so an issue that surfaces in one region can be pinpointed to the second and cross-referenced against everything else happening at that moment. As publishers extend measurement beyond the player — into app engagement, ad experiences, and agentic interactions — that same clock-level consistency is what will make it possible to see all of it as one experience instead of a set of disconnected dashboards.
Key Takeaways
- The quality-of-experience discipline built for streaming applies to every layer around it — app behavior, ad engagement, and any agentic experience a publisher launches all deserve the same real-time standard as video playback.
- Using AI agents to monitor and analyze at scale (agentic analytics) is what makes real-time response possible — at the volume of a global sporting event, no human team can scan for issues fast enough; agents can.
- Measuring a consumer-facing agent’s impact (agent analytics) is a different question than whether it answered correctly — it’s about whether the person on the other end was actually helped, not whether the response was accurate.
- Personalization has to be built on individual behavior patterns, not demographic segments — knowing what a specific fan watches, when, and in what sequence is a fundamentally different signal than knowing their favorite team.
- Viewer behavior, ad engagement, and app activity are one continuous picture, not three separate measurement problems — the same pattern data that predicts content preference is the foundation for understanding engagement everywhere else in the experience.
- Structured, time-synced data is a prerequisite for reliable AI agents, not an optimization — as measurement expands beyond the stream, that same consistency is what keeps everything correlated to a single, trustworthy timeline.