Personalisation is a continuous capability
The biggest players in the streaming industry have spent years refining and personalising the user experience; tier 0 and 1 operators have invested billions in improving personalisation. As far back as 2006, Netflix ran a competition offering a prise of $1 million to anyone who could improve the accuracy of its recommendation by 10%. Twenty years on and its AI personalisation engine reportedly saves the company $1 billion in annual savings by reducing churn.
Top-tier OTT players personalise their services with a level of granularity that has remained out of reach for the majority of smaller services. However, as AI tools have become more accessible and powerful, regional broadcasters and smaller OTT services have an opportunity to apply similar thinking and personalise the user experience to a much more granular level than was possible before.
Yet many still treat personalisation as a problem that can be fixed once and for all by deploying an AI recommendation engine. There’s a tendency to think that once a personalisation model has been implemented, the personalisation requirement is met, and the model can be left alone to run. When actually, what services need to understand is that personalisation is an ongoing capability, not a one-off problem to be solved.
How far personalisation has come
In 2026, effective personalisation means personalising almost every element of the experience to individual user level. This includes recommendations, thumbnails, title descriptions, trailers, when and how a title is suggested, reordering the content rails and selecting the categories to display on the home page.
Take thumbnails as an example: for the same piece of content, a romance fan might see a thumbnail featuring a couple, someone who prefers strong protagonists may see the lead character, while a viewer that likes a particular actor, might see that actor featured instead. Even the expression of the character on the thumbnail can impact engagement. Having tested a vast number of thumbnails, Netflix knows that the ones with a polarising character with a facial expression displaying strong emotions, such as anger, fear or sadness, prompts more viewers to press play.
But while AI tools are now readily available that offer smaller services an affordable means of better personalising the user experience without billions of dollars investment or huge teams of engineers, they don’t work in isolation. They still need the right underlying foundations with a certain depth of viewer data otherwise personalisation engines are effectively blind.
Understanding the barriers to achieving better personalisation
The fragmented nature of viewing today means that viewers often watch the same service across different devices, whether on smart phones, CTVs, or on web browsers. For personalisation engines to do their job properly, they need a solid understanding of a user’s behaviour, and this is only possible if data from across these different platforms is fused together, otherwise personalisation models are basing their assumptions on incomplete data.
It’s hard to make good content recommendations if you don’t have enough behavioural data available to build up a really good accurate picture of the user. This can be a real problem for smaller or niche services which users don’t necessarily access daily.
Another problem comes if content metadata is lacking or inaccurate, because even with the best personalisation engine and or the most in-depth user behavioural data, the system can’t locate and surface the right content if the information needed to identify it is lacking.
Another critical factor often overlooked is the need to verify that a particular personalisation tactic is working. The big players didn’t get where they are today without experimenting and working out what works and what doesn’t. This ability to systematically test the impact of personalisation activities, then reiterate as required until the desired outcome is achieved, is often missing from many medium and small services.
The right foundations
A lot of streaming services only have comparatively small user datasets and limited resources to invest in personalisation. However, even then, they can still achieve meaningful results provided they have the right foundations in place.
Building up a solid understanding of users starts with being able to identify individual users. This is only possible through requiring users to register and set up individual profiles. Services can also gather user data by asking questions about preferences and interests at this point. Once users are signing in to their individual profile to use the service, the next priority is to ensure viewing behaviour is tracked consistently across devices. Searches, browsing activity, playbacks, pauses and abandoned titles all provide useful signals, but usage across mobile, CTV, and web needs to be analyzed together to build a more complete picture of each viewer.
Good quality content metadata is another essential component so it’s important to make sure titles need to be accurately categorised with detailed descriptions. There are various tools on the market that can automatically generate metadata across entire catalogs.
Finally, services need to build in the ability to test and refine their approach. Even relatively simple rules-based personalisation can deliver results if it’s properly implemented and regularly reviewed.
Making the most of existing capabilities
OTT providers can also improve personalisation by building on existing capabilities and making better use of the data already collected. Take search and discovery: if a service can understand what viewers search for, which titles they select and where they abandon the browsing process, this helps identify opportunities to improve content discovery. Personalising search results and the way content is surfaced according to individual viewing behaviour makes it easier for viewers to find something relevant, reducing the likelihood that they leave without watching anything.
Adjusting the homepage offers another opportunity for personalisation. Rather than presenting every subscriber with the same selection and ordering of content rails, services can use viewing history and preferences to determine which categories and titles should be most prominent. Additionally, it’s also important to consider the user’s side of personalisation. Transparency in how data is used is very important to users; one approach to help achieve greater transparency is to label recommendation rails with "because you watched/ enjoyed show X" rather than the standard "recommended for you." Enabling controls that give users a way of managing their viewing history and content recommendations can also help, as can balancing algorithmic picks with editorial curation.
Notifications are another area where video services can make relatively simple improvements. Instead of sending the same messages to everyone at set times, by analyzing viewing patterns and behaviour, services can identify when individual users are most likely to engage and what they may be most interested in. The timing of push notifications, and the recommendations they make, can then be tailored to get better results.
Making personalisation an ongoing process
While smaller services don’t have the same resources as streaming giants, they now have the opportunity to offer a more personalised experience than was possible a few years ago. This starts with making better use of data and capabilities already available, then testing and refining different approaches over time. Personalisation is an experience and operational challenge, not just an algorithm problem. The engine decides what to show, but whether that actually changes behaviour depends on how it’s presented, measured and looked after over time. Many services jump straight to a vendor when their identity, profile, metadata, or measurement foundations aren’t ready. A better approach is to establish that baseline first, so investment goes where it will actually move results. The algorithm and interface have to be designed together, with shared goals that are continually tested and developed.
[Editor's note: This is a contributed article from Accedo. Streaming Media accepts vendor bylines based solely on their value to our readers.]
Related Articles
Redefining viewer engagement demands ongoing evolution. Streaming entertainment should feel effortless. Discovery should be smart. Subscriptions should be simple. And operators should have the tools to act with precision and agility.
09 Jul 2025