Everything you need to know about HyphaMetrics

HyphaMetrics measures what people actually watch. We recruit consented households across America and install off-the-shelf capture devices running our proprietary AI software. This lets us observe and identify all media content — across every screen, every platform, and every person — and license that data to media companies, brands, agencies, and AI companies who need ground truth about human attention.

AI marketing tools are only as good as the data they're trained on. Today, most media measurement relies on what platforms self-report, demographic guesses, or asking people what they remember watching. None of that captures what actually happened on screen or who was actually in the room. HyphaMetrics closes that gap by using computer vision to identify content frame-by-frame, giving AI systems the ground truth they need to optimize effectively.

Hypha is the thread-like vegetative part of fungi that forms mycelium networks. Just as hypha enables forests to communicate, we are the connective tissue that enables the media ecosystem to communicate.

The CoreMeter is a small hardware device installed on each screen in a household. It captures video frames and uses multiple layers of AI—video recognition, hashing, spatial correlation—to identify content in real time. The data is processed through our cloud-based Unified Neuromedia Identification Engine (UNIe) and matched against our content database. We capture what is being watched, on which device, by which person, at what time—across broadcast TV, cable, streaming, gaming, social video, FAST services, and GenAI content. We are on version 7 of our hardware after seven years of iteration.

It starts with how the coreMeter actually works. The coreMeter is a patented hardware device installed in consenting households that uses computer vision and a purpose-built AI stack to identify on-screen content regardless of whether audio is present, whether a platform has consented to third-party measurement, or whether the content carries an embedded watermark. It does not listen for what is on the screen. It looks. That distinction is not semantic. It is the difference between counting and guessing. But technology alone cannot compensate for a sample that does not reflect the country. A representative panel does not emerge from random selection. Getting the counting right requires both the right technology and the right sample design. HyphaMetrics builds and manages its panel in partnership with Ipsos, with explicit outreach to the populations that legacy measurement methods have consistently underrepresented, including younger adults, urban dense areas, multilingual households, and communities that have historically been harder to reach and retain in traditional panels. Currently in over 1,000 households and on track for 5,000 by mid-2026, the panel is designed to mirror the demographic, geographic, and technographic diversity of the U.S. population. Major clients benchmark 5,000 households as the threshold for statistical validity. That is the target, not a stretch goal. The simplicity is intentional. Counting is either right or it is wrong. Audio-based measurement could not get it right, not because the engineers who built it failed, but because audio cannot see a muted screen, a set of earbuds, or a streaming platform that never agreed to be measured on a fixed schedule. The hardware and AI stack HyphaMetrics invented exist because the industry needed a system capable of doing what prior approaches structurally could not. The count is the foundation. Everything else follows from it.

CoreMeter devices capture video frames in each household. Computer vision identifies all content frame-by-frame. Our ML engine (UNIe) maps it to a content database in real time, while our metalibrary data labeling team fortifies the models continuously. The data product covers content identification, device-level viewing, person-level presence, and ad exposure across broadcast TV, cable, streaming, gaming, social video, FAST services, and GenAI content. Clients access data via our Personicore API or dashboard, with daily data delivery once 1,000+ households are active.

Because we use computer vision rather than audio fingerprinting, we can measure: all streaming services (Netflix, Prime, Disney+, etc.), short-form video (YouTube, TikTok, Instagram), video gaming, dynamic and programmatic ads, FAST services, GenAI-generated content, muted content, and cross-device journeys. Legacy measurement cannot reliably capture most of these categories because audio ACR requires platform cooperation and pre-built content libraries.

Privacy is foundational to our architecture. We are designed to be privacy-compliant by default, operating within all relevant regulatory frameworks including GDPR and CCPA. Our technology works with consented data and anonymized signals, ensuring measurement accuracy without compromising user privacy.

We sell across three primary segments—publishers, agencies, and brands—plus measurement companies, ad-tech platforms, and emerging segments like AI labs, hedge funds, political campaigns, and retail media networks. Our product sits at the intersection of research budgets (typically OpEx with fixed, long-term contracts) and programmatic budgets (typically COGS with transactional/usage-based pricing), allowing us to capture wallet share across multiple line items.

Publishers like NBCUniversal, Disney, and Comcast can use our data as a foundational source that informs content strategy, ad product development, and competitive positioning against rival networks and streamers. They get respondent-level data for cross-platform consumer insights, competitive analysis, and content-audience clusters.

Agency holding companies like WPP, Publicis, and Omnicom can use our data to develop media strategies, build planning tools, and provide competitive intelligence to their clients. Our data serves as a truth-set that calibrates their planning models with observed cross-platform viewing behavior rather than modeled estimates.

Brands like Procter & Gamble, Nike, and Coca-Cola can use our data to inform media strategy, track competitive share of voice, and develop intelligence on where rivals are reaching consumers. Our data enables brands to see actual ad exposure across every screen and platform, understand cross-device journeys, and optimize creative and placement decisions.

Ad-tech aggregators like The Trade Desk, DoubleVerify, and IAS evaluate whether our data fills a product gap, then move to licensing terms. We serve as a truth-set calibration layer for programmatic trading models. Measurement companies like Comscore license our data to power their own products—panel building is not their core competency and they lack the capital to replicate our approach.

Beyond traditional media measurement buyers, we see demand from AI labs and hedge funds that need truth-set data on attention and media consumption, political campaigns and agencies that need unique insight into voter media behavior, retail media networks that can prove incremental reach beyond owned platforms, and sports leagues and broadcasters that need to understand total cross-screen viewership.

Hypha’s panel build and run costs are a fraction of its competitors’ thanks to more durable hardware, a connected panelist experience, and CV models that get smarter and faster with every frame. Ask the Hypha team for specific costs!

The 5,000-household threshold is an industry benchmark established through decades of panel-based measurement practice. Major clients consider 5,000 households to be the size at which panels reliably represent the demographic, geographic, and technographic diversity of U.S. households across nine Census Divisions while maintaining sufficient sample within sub-segments to produce stable estimates. We own the supply chain and can increase our panel easily with a low marginal cost. We are planning to do this and offer the expanded dataset as an upsell to clients.

Our Chief Panel Officer draws on 20+ years of domain experience at Nielsen, GfK, and Comscore to execute a proven playbook. Ipsos continuously refines enrollment survey design and incentive structures. Our SuperApp v2 includes step-by-step video guides, live chat support, and a gamified incentive marketplace. Real-time monitoring dashboards flag issues like devices offline for 4+ days.

We have no direct competitors offering the same combination of panel-based, computer-vision, cross-platform measurement at the household, viewer, and device level. We have never encountered a sales scenario in which we are being directly compared with another company with comparable products. Comscore is our customer, not our competitor. They license our data for their measurement products. VideoAmp is a measurement and optimization platform that uses licensed data sets rather than a proprietary panel. These companies offer currencies and ratings, whereas we offer first-party data.

TVision builds a panel using audio ACR plus a camera facing the couch for person-level data. TVision still faces the same underlying audio ACR limitations as other legacy providers like Nielsen.

Comscore is our customer, not our competitor. They license our data for their measurement products. VideoAmp is a measurement and optimization platform that uses licensed data sets rather than a proprietary panel. These companies offer currencies and ratings, whereas we offer first-party data.

There is a substantial gap between what a patent describes conceptually and what it takes to build a production system. A well-funded competitor starting today would need to replicate years of hardware and software iteration, build panel recruitment infrastructure, develop proprietary onboarding software, establish data processing pipelines, and earn the trust of panelists, clients, and industry bodies—all while we continue to iterate ahead of them.

Three granted U.S. patents, all sharing a July 2019 priority date: Patent No. 10,932,002 (granted February 2021) covering multiple content identification rules; Patent No. 11,570,513 (granted January 2023) covering ML models for logo and brand identification via trigger events; and Patent No. 12,238,372 (granted February 2025) covering advertisement identification after machine-learned frame trigger events. All received first action allowance or were allowed after a single round of examination. Standard 20-year patent term places expiration around 2039.

Nielsen filed four serial patent infringement suits against us between November 2021 and April 2024. All are resolved with no infringement findings. Case 1 (filed November 2021): Nielsen voluntarily dismissed before trial. Cases 2 and 3 (filed 2023): went to trial July 28–31, 2025, with jury verdicts of non-infringement on August 1, 2025. Case 4 (supplemental complaint filed April 2024): Nielsen dropped the claim one week before trial. Nielsen did not file post-trial motions and did not appeal by the September 2025 deadline.

Cross-platform is at the core of what we do. Our AI unifies measurement across devices and platforms into a single deduplicated view. A viewer who watches content on their TV, then picks up on their phone, and sees an ad on a digital billboard is measured as one continuous journey, not three separate events.

Hypha measures media consumption across all major platforms and formats, including linear TV, streaming (SVOD/AVOD/FAST), social video, digital display, audio, gaming, and out-of-home. Our technology is platform-agnostic and content-format-agnostic and our coverage is continually expanding as the media landscape evolves.

Yes. Our measurement covers all business models in media — ad-supported (AVOD/FAST), subscription (SVOD), hybrid, transactional (TVOD), and free ad-supported television. We measure content engagement regardless of the underlying monetization model.

Several converging forces drive demand: the fragmentation of media consumption across platforms, the deprecation of third-party cookies and identifiers, the growth of connected TV and streaming, advertiser demands for better measurement, and regulatory pressure for more transparent, privacy-compliant data practices. On the demand side, every AI-forward business needs as much data as possible on their consumers to keep all models as accurate and effective as possible.

The streaming revolution has broken the legacy measurement paradigm. Traditional panel-based measurement was designed for a 200-channel cable world. With thousands of streaming services, millions of content titles, and viewing happening across every screen, the market needs a fundamentally new approach — which is exactly what Hypha provides.

Everyone on earth consumes media and our technology is inherently scalable across geographies. We are building toward international expansion, prioritizing markets with mature media ecosystems and strong demand for independent, technology-driven measurement.

The panel has a defined steady-state run cost, and breakeven requires a modest number of enterprise clients relative to the large addressable market. Given thousands of brands and media companies who would benefit from this data, and a strong active pipeline across multiple segments, the path is achievable. Every segment we’ve tested has shown demand, and we’ve had no trouble sourcing and progressing sales opportunities.