Now Streaming… Your Forgotten Thoughts
Leveraging BLUR to transform fuzzy customer inputs into clear results by building AI systems that reason, browse, and adapt like humans.
Mallory wasn’t looking for a moonshot. As fictional director of content discovery at StreamWorks (a fictional streaming platform with a global user base), her job was to help subscribers find the right show or movie, faster and with less friction. For years, she had led the charge on recommendation engines, watchlist optimization, and metadata tagging. The platform had rolled out everything from genre-specific carousels to curated collections powered by machine learning (ML). Yet despite all the innovation, one stubborn problem kept showing up in user feedback and support tickets: people still couldn’t find the content they already had in mind—because they couldn’t remember its name.
“I’m looking for that sci-fi movie where it’s always raining and there’s neon everywhere,” one support ticket read. Another simply asked, “What’s that old fantasy film with the flying white dog?” These weren’t search failures due to bad tagging or a lack of options. They were memory problems (common, very human “tip-of-the-tongue” experiences), and the platform had no intelligent way to handle them.
Mallory and her team tracked these vague-recall searches over time and noticed a troubling pattern. Customers were expressing intent (sometimes very clear visual or emotional clues), but the AI systems failed to translate them into usable queries. Instead, they returned irrelevant results or defaulted to trending content, which only deepened the user’s frustration. Even worse, these moments led to increased support tickets and more search session abandonments. Mallory knew this was a signal, not noise. Despite all their investments in AI, StreamWorks had a blind spot that impacted discoverability, satisfaction, and ultimately retention.
When More Content Makes Things Worse
The problem wasn’t just that the catalog was large; it was also that it was growing faster than the platform’s discovery capabilities. As international licensing deals and regional original productions multiplied, the sheer volume of content became overwhelming. Customers searching for older or niche titles struggled even more when titles had been renamed in different markets or had no obvious metadata tags connecting them to modern search behaviors.
Adding to the challenge was a shift in how users were interacting with search. People were no longer typing in clean titles like “The Godfather” or “Bridgerton.” Instead, they were entering ambiguous, mood-driven descriptions: “a dark political drama with lots of betrayal,” or “the one with the twist ending on the train.” Natural language processing (NLP) tools built for keyword matching or topic modeling weren’t built to decode memory fragments like these. And the search engines that powered results weren’t designed to navigate uncertainty or piece together evidence from multiple clues. They needed a name, or at least a close match.
Internally, Mallory also faced a looming competitive pressure. Rival platforms like Flixter and Viewtopia (also fictional) were rumored to be experimenting with smarter AI assistants, ones that could potentially “understand you better than you understand yourself.” If they got recall-based discovery right first, StreamWorks would fall behind not because of weaker content, but because of a clunkier experience.
And perhaps most frustrating for Mallory’s team was that they had the clues. Users were offering detailed, evocative prompts. But their systems lacked the interpretive capacity to do anything meaningful with them.
Ignore It, and Pay the Price
The risks of inaction weren’t theoretical. Over the previous quarter, support teams reported a 40% spike in unresolved content discovery requests. User satisfaction scores around “ease of finding content” dropped noticeably. Worse, session lengths and view-to-search ratios were starting to show signs of fatigue, with users bouncing between tiles and menus before abandoning the app entirely.
Mallory could feel the pressure building across leadership: growth had slowed, acquisition costs were climbing, and there was no room to lose users because of a problem that felt fixable. If customers didn’t feel seen (or understood) by the platform, they would walk. Worse, they’d tell others.
The question was no longer whether recall-based discovery was important. The question was: how fast could they catch up to the future customers already expected?
Curious about what happened next? Learn how Mallory applied a recently published AI research, stopped chasing titles (and started reconstructing intent), and achieved meaningful business outcomes.