Why better viewer experiences start with better data, not better AI

There’s no shortage of data in the streaming industry, but turning that information into insights that improve engagement and user experience remains challenging. Too often, data is fragmented across disparate dashboards, making it difficult to consolidate.

AI is often positioned as the answer, but quality data is the bedrock of any AI workflow, and right now AI has a data problem.

Metrics in isolation don’t tell the whole story

Igor Oreper: We’re drowning in metrics, but there’s very little insight coming from them. A typical streaming setup has lots of systems creating data points, with encoding, player, QoE, and ad delivery metrics sitting separately, and each team working to optimize its own function. Dashboards tell us what happened, but business decisions depend on why it happened, who it affected and what it cost.

An ad server might say it’s delivered a million ads, but the player might show that only 700,000 actually played. What really matters is what the viewers actually saw, where those ads dropped off, and what’s needed to fix the problem. If the data isn’t joined together, you can’t see what’s really going on.

Bleuenn Le Goffic: We began looking at unified observability as a means of connecting the dots a few years ago, before agentic AI was on the radar. We wanted to move beyond reacting to incidents to proactively analyzing platform performance and patterns in audience behavior. The original assumption was that this required a unified data strategy with all the information in one place. But for most companies, bringing data together across different systems, vendors and existing contracts is so costly and complicated that it’s essentially non-viable, particularly for mature streaming platforms where rearchitecting the infrastructure simply isn’t realistic.

Understanding and monitoring user experience

BLG: Observability improves visibility, but many issues that impact user experience don’t surface in dashboards. They can anticipate incidents but may miss UI problems that aren’t business-critical, such as a button that doesn’t work or a lag when moving between screens. From a product management standpoint, these are extremely important.

Nothing beats testing the platform yourself to assess the user’s experience, but this is resource heavy and can’t be scaled. So instead, we’re using agentic AI workflows where agents act as virtual users testing the service in different ways depending on their trained persona.

IO: Even where dashboards agree, there can still be problems when you start digging. What was the quality of the delivered ad and how was the experience of moving from main content to ad and back again? This is why it’s important not to look at the data in aggregate, because averages hide the problems that matter. Instead, you have to look at data at session level, connecting playback data, user behavior and content metadata.

Advertisement

Playback data tells you what the user experienced, while content metadata tells you what they were watching, down to the scene. If viewers quit in episode three of a series, why is that? Is it a dud episode or a startup problem on a specific device? Now that we have more granular scene-by-scene metadata, we have the signals to interpret that behavior. Content metadata is the connective tissue. Combined with metrics from other systems, it turns behavior into insight.

AI can help but it’s only as good as the data that feeds it

BLG: Agentic AI has led us to look at product management and operations differently. You identify an objective, such as better issue management, incident management or faster development cycles, and then ask whether agentic AI can help you achieve it.

From there, you look at what signals you need to capture through observability and what context you need to bring into that workflow. Take incident management: you have the issue, but you also need to understand the platform, its architecture, how it’s behaving and how the operations team works. AI needs all that context to do the job properly.

IO: The data that matters most is tied to the moments that cost viewers and revenue: startup time, rebuffering, playback failures, ad errors, abandonment and completion rates. AI can help turn metrics into insight, but only if we get control over the data and can access it readily.

AI can help create good data, but the systems that generate it need careful training, testing and a lot of human validation to verify quality. With our AI Scene Analysis product, for example, we’ve gone through many iterations of getting our models to interpret video and audio in context, because the pacing of news, sports and a romantic comedy can be very different.

Bridging the data gap

BLG: For AI to support a workflow, it needs the right context and data, which requires investment in defining the workflow and making data accessible. But MCP shouldn’t be the default for every integration. It has strong use cases, but a direct API is sometimes the better fit, and not just on cost. I think there’s a realization now that as an industry, we need to go back to the drawing board and rethink how we work.

IO: The AI models we have are excellent, but to produce good results, you need high-quality data and a way for the models to access it. MCP lets agents query live operational data where it sits, instead of relying on someone to export it, so you don’t have to move everything into one place first. Agents can then consume data in a structured way at scale, as needed. But make no mistake, if they get bad data, they just make bad decisions faster. Data accuracy is paramount.

Key takeaway

Observability, experience monitoring and AI are converging to reshape streaming operations and product management, but AI’s value depends on surfacing the right data, with the context to turn it into useful insights.

Bleuenn Le Goffic, Accedo, and Igor Oreper, BitmovinBleuenn Le Goffic, VP, Business Transformation, Accedo Bleuenn Le Goffic has worked for Accedo since 2012, initially in Sales before heading up Global Partnerships & Innovation and has a deep understanding of video streaming and the wider media and entertainment industry. As part of her last role, she spearheaded Accedo’s sustainability strategy and continue working with ecosystem partners to accelerate the industry transformation. Prior to joining Accedo, Bleuenn held positions in software engineering and as an archeology researcher. Igor Oreper, Chief Strategy Officer, Bitmovin Igor has served in a number of customer-facing technology and product roles for the Media and Entertainment industry over the last 18 years. In his role as Chief Strategy Officer, Igor is a technical and business advisor to Bitmovin's strategic accounts. Immediately prior to Bitmovin, Igor managed the P&L and full product lifecycle of the Video Platform business at Comcast Technology Solutions (CTS).