Ask a room of executives whether they'd be comfortable making decisions with 1% of the information they need, and no hands go up. Yet that's close to how most organizations treat video: the richest record of what people watch, say, show and buy, and almost entirely unread.
This brief brings together public research and Qortex's own findings to size the problem.
Key findings
- 17,000+ years: the time it would take to watch everything on YouTube alone (Lawfare, 2023).
- Less than 1%: the share of footage from more than 1.5 billion cameras that is ever reviewed by humans (NVIDIA, March 2026).
- About 80%: the share of enterprise data that is unstructured (IBM).
- 72 hours: how long most media and advertising companies keep stored video before deleting it (what we hear consistently in Qortex's conversations).
- 88.98%: the share of 50,000+ video pages where the page's IAB category didn't match the video on it, in a study Qortex ran for a leading brand-safety verification company.
- 57%: AI projects not reported to be delivering their objectives, even as 97% of organizations deploy or pilot AI agents (Nasuni, 2026).
1. There's too much video to watch
Lawfare estimated in 2023 that watching all the video then on YouTube would take more than 17,000 years of nonstop viewing. That's one platform. In January 2025, Qortex projected that, with AI-generated content, the total could reach 100,000 years of viewing by the end of that year (a Qortex estimate).
It isn't only media. NVIDIA reports that more than 1.5 billion cameras capture footage globally, and less than 1% of it is ever reviewed by humans.
2. What isn't watched gets thrown away
Video is expensive to store and hard to search, so most of it doesn't last. What we hear consistently in our conversations with media and advertising companies: most delete stored video within 72 hours. Whatever that video knew about the audience, the moment and the products on screen goes with it.
It fits a wider pattern. IBM estimates that about 80% of enterprise data is unstructured, and video is the hardest kind of unstructured data to put to use.
3. What does get read is often read wrong
When video is analyzed at all, it's usually through titles, tags, keywords and descriptions rather than the content itself. In a study Qortex ran for a leading brand-safety verification company, we took more than 50,000 URLs from its data. For each one, we compared the IAB category the page had been given with the IAB category our models assigned to the video on that page. In 88.98% of cases, the two didn't match.
That isn't a reporting footnote. Targeting models, brand-safety filters, recommendation engines and measurement all inherit the error. Buying faster on the wrong understanding just gets you to the wrong answer sooner.
Context is where blanket rules break down. As Qortex CEO Zack Rosenberg puts it, "One brand's bad news is another brand's perfect fit." A car-accident story may be the wrong place for an automaker and the right place for a body shop, an insurer or a law firm. In sports, "they're killing them" means one team is winning by a lot. A keyword list can't tell the difference. Understanding the video can.
4. AI can't use what it can't read
Organizations are moving fast on AI. Nasuni's 2026 survey of 1,000 purchasing decision makers found that 97% of organizations have deployed or are piloting AI agents, yet 57% of AI projects are not reported to be delivering their objectives. Models and agents are only as good as the data they can reach. While video stays unread, or read wrong, the most detailed record a company has of its customers and its content stays outside every model it builds.
What it takes to read video
Reading video at scale means turning each video into structured data a product can use:
- what the video is about, in a standard taxonomy such as IAB categories
- what's said (transcript and summary) and how it feels (sentiment)
- what's shown: scenes, objects, on-screen text, brands and logos
- risk and brand suitability, with the reasoning behind each call
- which advertisers, products and messages it fits
Done consistently across a whole library, not clip by clip, that turns stored video from a cost into a dataset.
How Qortex helps
Qortex runs hosted video models built for media, advertising and entertainment. Send video URLs, choose your signals and get structured JSON back, at 100k+ videos per hour. Our models are our own: we've been working on them for over five years, taking in billions of data points and human-curating them, and your video isn't sent to a big LLM provider.
Sources
- Lawfare, "Exploring Tradeoffs in Ranking and Recommendation Algorithms," Sept. 29, 2023: lawfaremedia.org
- NVIDIA newsroom, March 16, 2026: nvidianews.nvidia.com
- IBM, on unstructured enterprise data: ibm.com
- Nasuni press release, May 18, 2026: survey of 1,000 purchasing decision makers at organizations with 1,000+ employees in the US, UK, France and DACH.
- Qortex: 100,000-year projection (Qortex blog, January 2025), 72-hour retention (what we hear consistently in conversations with media and advertising companies), and the miscategorization study (50,000+ URLs, page IAB category vs. video IAB category; study partner confidential).
