Media products are used differently from most software: for hours at a time, habitually, often by people who are not concentrating. Small friction compounds, and a control that is merely acceptable on first use becomes irritating by the hundredth.
We have designed interfaces at that scale, including work on STV Player. The same attention now applies to what sits behind the interface: the metadata, search and recommendation systems that decide whether anyone finds the content at all.
Insights
Recommendation is a metadata problem
Recommendation projects are scoped as machine learning work. They usually stall on a taxonomy nobody owns and a back catalogue tagged by whoever ingested it.
Read the articleDesigning for people who are not concentrating
Media interfaces are used habitually, for hours, by people doing something else. Friction that is acceptable on first use becomes intolerable by the hundredth.
Read the articleSearch is what people use when recommendation fails
Discovery investment goes to recommendation. Search gets what is left over, despite being what people reach for when they know exactly what they want.
Read the article
How we support changemakers in entertainment and media
Set up and run teams
We work inside your teams rather than at arm's length from them, at whatever pace the release schedule demands.
Launch new products
We build to production standards from the first sprint, with the performance budget treated as a requirement rather than something measured afterwards.
Discover new propositions and customer experiences
Long-session products are judged on the hundredth visit, not the first. We design for the habits that form after the novelty wears off.
How we strengthen your team
Interface design for long sessions
We have designed and built streaming and content experiences, where small friction compounds over hours of use and a clumsy control is felt every single session.
Metadata and search that people can use
Recommendation and discovery are only as good as the metadata underneath them. We do the unglamorous work of getting that right before layering models on top.
AI applied where it helps the audience
Automatic tagging, transcription and semantic search over back catalogue, built with the same governance and explainability as any other production system.
Search that works before recommendation does
Typo tolerance, structured entities for contributors and series, synonyms for what audiences actually call things, and availability applied to results. Zero-result queries are read rather than discarded — they are the clearest statement an audience makes about what it expected you to have.
Rights and availability as data
Windows, territories and platform rights held as queryable data rather than as rules applied at the end, so nothing can be recommended, searched or promoted that cannot legally be played. It is the one discovery failure an audience never forgives.
Measured where the audience actually is
Start-up time, rebuffering, playback failure and time-to-first-play tracked by device and connection rather than in aggregate. In a habitual product every second between opening the app and content starting is paid daily, by everyone.
White paper
Metadata foundations for content discovery
Why recommendation projects stall, and what a content business needs in place before any model is worth training: taxonomy, tagging, rights data and search that already works.
Our work
Let’s talk
Have a problem you need to solve or an idea you want to explore? Let’s talk about how we can make it happen.