To win the fight against churn and build lasting engagement, media and entertainment companies are leaning heavily into content personalization. In fact, they’re committing themselves to it.
“We’ve reached this inflection point where AI hyperpersonalization is transitioning from experimental into foundational,” said Nina Walsh, global leader of business development, Media & Entertainment, Games and Sports, at Amazon Web Services (AWS). “Consumers are drowning in content and are looking at how best to find what they want when they want it. Passive discovery is dying. Channel surfing is gone.”
Personalization is also one of the most effective defenses against churn, added Mrugesh Desai, chief revenue officer at Accedo.“When viewers can’t quickly find something worth watching, they begin to question the value of their subscription,” he said. “The best streaming services know that content discovery, not content volume, is what drives engagement and retention.”
This means treating television like any other modern application. “People expect the same level of convenience from television that they get from every other digital service they use,” explained Elodie Levrel, corporate marketing and communication director for Broadpeak. “That means recommendations should reflect their interests, live and on-demand content should be easy to find and the experience should feel consistent whether they’re watching on a connected TV, tablet or smartphone.”
Breaking the Metadata Bottleneck
Today’s hyperpersonalized content delivery relies on a massive amount of high-quality data. Unfortunately, the efficiency of these systems is being hampered by legacy databases and the very nature of household viewing.
“One of the problems with personalization is the fact that a television is a shared device—one per household, rather than a truly personal device like a phone or a tablet,” said Dimitri Tarassenko, senior vice president of product management for LTN. “To make a step from addressable [i.e. targeted at a household] to personalized [targeted at a viewer] you need more data, and programming context [what the viewer has watched to date] is often a very good clue.”
Extracting that context in a useful way is the industry’s biggest hurdle. “This is the No. 1 conversation that I have with media companies all over the globe: Legacy metadata is the biggest obstacle that they encounter,” Walsh said. “The reason for that is that there are decades of inconsistently tagged, multiformat, multilanguage content libraries that weren’t designed for algorithmic consumption.”
Desai agreed. “Poor metadata is still the quiet killer of personalization projects, and it’s rarely fixed by buying a better recommendation engine,” he said. “The real work is foundational: First, an honest audit of where metadata is inconsistent, incomplete or siloed across legacy CMS systems, then establishing a single normalized schema that every content source (live, VOD, licensed and owned) maps into.”
Titan TV President Heidi Steffen stressed the importance of reliable data for feeding content personalization systems. “You need the data that is being fed into AI to be consistent,” she noted. “‘Garbage in’ results in ‘garbage out.’ If the data being used is inconsistent and confusing, AI will not be able to produce reliable, useful content recommendations.”
Curing Choice Fatigue
Conventional wisdom says churn happens because viewers are tired of paying for too many streaming services, which isn’t necessarily the case.
“Churn in streaming, much like piracy, is a convenience problem rather than price,” said Jacob Arends, senior product manager for Bitmovin. “If one of those services isn’t meeting a viewer’s needs or keeping them engaged, then it will likely be on the short list for the next service to drop.”
The same convenience problem occurs when platforms present viewers with too many recommendations. “Casual viewers do not want to be the director, so giving them too much choice and options actually creates a less rewarding experience,” Arends told TV Tech.
Instead, these viewers want to be told what to watch, and those recommendations had better be spot on—or else. “The data shows that if a viewer hasn’t had a ‘wow’ moment in 72 hours, it’s probably a 30-day churn from there,” Walsh said.
Ironically, making it hard for viewers to drop services can actually increase churn over the long term. “For the Gen Z and millennial cohorts driving the 50%-plus churn rate, the fix is rarely a smarter algorithm alone but rather to remove the friction around leaving and returning,” Desai said. “The platforms with the lowest churn make cancellation painless and reactivation a single click, because a subscriber who leaves easily is far more likely to come back than one who feels trapped.”
Ultimately, keeping those subscribers engaged requires an end-to-end approach to the user experience. “The broadcasters most likely to reduce churn are the ones that combine personalization with a simple interface, value-added features such as multiview for live sport and advertising that feels relevant without interrupting the viewing experience,” Levrel said.
Live Sports Demands Clean Data
Live sports and hyperlocal news present unique challenges for content personalization algorithms.
A case in point: Broadcasters that decide to route a live game to a secondary over-the-air channel can wreak havoc on smart-TV algorithms and viewer retention. “They can’t put it on their 9.1 due to other demands, so they put it on their 9.2, which is traditionally a pass-through,” Steffen said.
Making such switches confuses algorithms and viewers alike. This is why broadcasters and other content producers need to ensure their real-time metadata is properly synchronized so that content personalization systems can function correctly.
Poor metadata is still the quiet killer of personalization projects, and it’s rarely fixed by buying a better recommendation engine.”
Mrugesh Desai, Accedo
With the right data, delivered cleanly and consistently, these systems can work wonders. “We’re now at a point where productions can offer real-time user preference graphs, where you can build live models of each fan’s teams, leagues, favorite players and what their viewing patterns are,” Walsh said. “This enables them to personalize the ‘what’s on now’ experience for fans—but it requires ingesting real-time schedule data, rights, and metadata.”
To further optimize this experience, the delivery infrastructure must keep pace. “Technologies such as Media over QUIC (MoQ) are helping improve live streaming by reducing zapping time, lowering latency and limiting rebuffering,” Levrel explained. “During major tournaments, features such as multiview also give fans more control by letting them follow several matches or different camera angles at the same time.”
From ATSC 3.0 to Agentic AI
As content personalization expands, traditional broadcasters face technical hurdles—particularly in implementing ATSC 3.0.
“For traditional broadcasters, the biggest roadblocks are addressable advertising and interactive overlays onto ATSC 3.0 hybrid delivery orchestration,” Walsh explained. “Legacy infrastructure coexists, so broadcasters can’t abandon ATSC 1.0 overnight. This creates massive operational challenges, not to mention the cost.”
Added Tarassenko, “While the ad replacement and ad decisioning technologies are largely there, one of the biggest gaps lies in connecting the ad sales systems used for digital campaigns with the traditional infrastructure used to originate the linear channels.”
The good news: In the future, improvements in agentic AI, predictive behaviors and even physical biometrics are poised to make content curation entirely seamless.
The next real shift is agentic AI, according to Desai.
“Agentic AI closes this gap by letting specialized agents work together toward a shared outcome rather than each optimizing in isolation,” he said. “Get that right, and personalization stops being just a feature to become the core of a service’s operating model.”
Added Walsh: “Sensors in wearables can provide real-time emotional signals to enhance recommendation engines. A system that senses viewer stress, for example, could surface more calming content. Or, if your heart rate spikes during a scene, it could use that as a preference signal.
“Privacy is an important piece of all of this,” he continued. “The platforms that win long-term will be those that give users the transparent control over the personalization of their data, creating a trust-based value exchange.”
Of course, deploying such wearables will require a high level of consumer trust, because tapping into biometric and emotional signals moves content personalization from viewing behavior into sensitive personal data.
Overall, Walsh believes the M&E industry is moving towards agentic, predictive and generative personalization.
“As for how far it will go, it’s likely we will reach a point where the default entertainment experience will feel handcrafted for each viewer,” she said. “When this happens, the generic broadcast experience will feel dated, like the old-school telephone. As we move this direction, it’s vital to ensure that ethical frameworks are in place to build alongside the technology that’s evolving at a rapid rate.”
However, the ultimate goal isn’t total automation. “Personalization will continue to improve, but the goal isn’t to personalize every part of the viewing experience,” Levrel said. “The real opportunity is using personalization where it adds value, so people stay engaged and broadcasters can better understand what’s working.”
