What is spatial attribution DOOH?

Spatial attribution DOOH is the process of using physical sensors, like LiDAR and computer vision, to deterministically measure exactly who is looking at a digital screen, allowing for precise ROI and ROAS calculations.

How does LiDAR audience measurement work?

LiDAR uses active laser pulses to create a real-time 3D point cloud of a physical space. This allows the system to track shopper trajectories, proximity, and dwell times with sub-centimeter accuracy without relying on optical cameras.

How are computer vision advertising metrics used?

Computer vision processes optical data at the edge to determine gaze vectors, attention time, demographics, and emotional states. This allows advertisers to buy inventory based on verified attention rather than estimated impressions.

How is in-store ROAS tracking calculated?

In-store ROAS is calculated by taking deterministic screen-exposure data and matching it with POS transaction logs in a secure Data Clean Room. The conversion rate of the exposed group is then compared against a holdout control group to determine incremental sales lift.

What data does Trillboards collect via its screens?

Screens keep a record for each person detected: age range, gender, emotion, dwell time, and attention time. It computes a 512-number face-identity template to recognise people across screens, linking them to device keys and mobile advertising IDs. Selected frames go to Google Vertex AI for advanced estimation, and speech is transcribed on the device.

Spatial Attribution DOOH: LiDAR & Vision ROAS Tracking

The Evolution of Physical Screen Attribution

In 2026, the digital out-of-home (DOOH) industry has officially abandoned estimated, historical "Opportunity to See" (OTS) metrics. For decades, media buyers were forced to accept these inflated, probabilistic numbers, which were often based on outdated municipal traffic studies or broad cellular network pings. Advertisers now demand deterministic, real-time performance tracking that aligns with digital omni-channel budgets. As third-party cookies and mobile ad IDs (MAIDs) face strict deprecation in the online world, performance marketers have shifted massive budgets into the physical world,but only on the condition that physical screens can provide the same rigorous, mathematically sound attribution as a digital click.

The transition requires verifiable ground-truth data to measure exact audience exposure. Spatial attribution DOOH has emerged as the definitive standard for bridging physical environments and digital measurement. This methodology does not rely on guessing if a consumer was in the same zip code as a billboard; it mathematically proves that a specific consumer stood in front of a specific screen, at a specific angle, for a specific number of milliseconds.

By leveraging advanced sensor fusion, media buyers can finally execute precise in-store ROAS tracking. This shift moves the industry from probabilistic guessing to deterministic ROI calculation. Retail media networks, smart city kiosks, and transit hubs are no longer just brand-awareness vehicles; they are highly measurable, lower-funnel conversion engines that can prove their exact contribution to the bottom line.

The Problem with Legacy Measurement

Historically, DOOH relied on trailing mobile location data or static foot-traffic estimates. These methods provided a generalized view of venue traffic but failed to verify actual screen attention. A mobile location polygon drawn around a shopping mall might tell an advertiser that 10,000 devices entered the building, but it suffered from severe Z-axis limitations (failing to distinguish between floors) and GPS drift (where a signal bounces off concrete walls, placing a consumer 50 feet away from their actual location).

Advertisers could not definitively prove whether a shopper looked at an endcap display or simply walked past it. A consumer rushing past a digital endcap to grab a gallon of milk registers as an "impression" in legacy systems, even if their eyes never left their shopping list. This lack of precision made physical screen attribution nearly impossible for performance marketers who require strict causality between ad exposure and consumer action.

Key Insight: Without deterministic attention metrics, DOOH campaigns were often relegated to top-of-funnel brand awareness budgets, missing out on lucrative performance-marketing dollars. When a Chief Marketing Officer cannot draw a straight line from a $100,000 DOOH spend to a corresponding lift in point-of-sale (POS) revenue, that budget is inevitably reallocated to search or social media.

Today, the integration of edge computing, hardware sensors, and programmatic infrastructure has solved this attribution gap. Platforms now capture exact shopper trajectories and gaze vectors in real time. By processing this heavy spatial data directly at the edge,on the screen's local hardware,the industry has eliminated the latency and bandwidth costs that previously made real-time physical attribution impossible.

Rearchitecting DOOH with Sensor Fusion

The cutting-edge solution for modern attribution is a sensor fusion stack combining active laser sensors and camera-based artificial intelligence. This architecture provides unparalleled accuracy in retail environments. Sensor fusion is a concept borrowed from autonomous vehicles and advanced robotics; it acknowledges that no single sensor is perfect. Cameras can be blinded by glare, and lasers cannot see color or read text. By combining them, the system achieves a flawless understanding of the physical space.

By layering multiple sensing modalities, networks can capture a complete, multi-dimensional view of the audience. The optical camera identifies the demographic and the attention state, while the laser sensor maps the exact physical distance and movement trajectory. This powers the next generation of computer vision advertising metrics, allowing screens to act as intelligent, context-aware nodes in a global programmatic network.

Deep Dive: LiDAR Audience Measurement

LiDAR audience measurement uses active laser sensors to generate real-time 3D point clouds of retail environments. These sensors emit rapid light pulses,often millions of pulses per second,to map the exact physical space and the objects moving within it. By measuring the "Time of Flight" (ToF) of these light pulses as they bounce off shoppers and return to the sensor, the system creates a dynamic, high-resolution topographical map of the venue.

This technology tracks exact shopper trajectories and dwell times with over 99% accuracy. Because LiDAR relies on depth mapping rather than optical imaging, it operates flawlessly in variable lighting conditions. Whether a screen is placed in a dimly lit nightclub, a sun-drenched airport terminal, or a retail aisle with flickering fluorescent lights, the LiDAR sensor maintains absolute precision. It can distinguish between two shoppers standing shoulder-to-shoulder, and it can track a consumer's path from the entrance of the store directly to the advertised product.

Furthermore, research published by the International Symposium on Digital Forensics and Security (ISDFS) confirms that utilizing LiDAR for human silhouette identification is 100% privacy-safe and GDPR-compliant, as it gathers no biometric or personally identifiable information (PII). The point cloud generated by a LiDAR sensor looks like a cluster of floating dots; it contains no skin color, no facial features, and no readable text. It is the ultimate privacy-first tracking mechanism.

Core LiDAR Capabilities:

Sub-centimeter spatial resolution, allowing for the tracking of micro-movements like reaching for a product on a shelf.
Real-time 3D bounding box generation, which mathematically isolates each shopper in the physical space.
Multi-target tracking across large retail floors, seamlessly handing off a shopper's trajectory from one screen's sensor to the next.
Immunity to ambient light fluctuations, eliminating the "blind spots" that plague traditional optical-only tracking systems.

Computer Vision Advertising Metrics

Simultaneously, edge-based computer vision is deployed to measure "gaze vectors",the exact angle of eyes and head tilt relative to a display. This transitions DOOH buying from raw impressions to "Cost Per Attention Second" (CPAS). Instead of paying for a theoretical opportunity to be seen, advertisers bid on guaranteed, verified human attention. If a shopper looks at the screen for 3.5 seconds, the advertiser pays for exactly 3.5 seconds of active engagement.

According to trade architecture publications by Amit Goel, rearchitecting DOOH around these deterministic metrics is fundamentally changing how inventory is valued. Screens that command higher verified attention now programmatically command higher CPMs. Goel's research highlights that a screen in a high-dwell-time area (like a checkout line or a transit waiting area) with verified gaze tracking is exponentially more valuable to a performance marketer than a massive roadside billboard with zero deterministic tracking.

By processing optical data directly at the edge, screens can instantly classify audience demographics and attention states. Specialized Neural Processing Units (NPUs) embedded in the screen's hardware run lightweight machine learning models that analyze frames in milliseconds. This data is then appended to programmatic bid requests in milliseconds, allowing Demand-Side Platforms (DSPs) to bid on a 25-to-34-year-old female who is actively looking at the screen in real-time.

The Trillboards Architecture and Data Practice

Trillboards is a screen aggregator and audience measurement platform for digital signage and DOOH publishers. It provides the critical infrastructure needed to capture, process, and monetize these advanced spatial signals. Building a sensor fusion stack from scratch is prohibitively expensive for most venue owners and digital signage operators. Trillboards democratizes this technology by offering a unified, plug-and-play architectural layer.

SSPs and exchanges are complementary partners that connect screen inventory to programmatic buyers. Trillboards acts as the foundational measurement and aggregator layer, never operating as a replacement for an SSP or exchange. Instead, it enriches the bid requests flowing into those SSPs with deterministic, sensor-verified data, making the inventory vastly more attractive to premium buyers.

The platform offers free screen-management and ad-serving software, meaning publishers pay $0/screen/month. This is a massive disruption to legacy Content Management System (CMS) providers who traditionally charge exorbitant monthly SaaS fees per endpoint. Instead of charging software fees, the platform monetizes through ad demand, aligning its success entirely with the publisher's success.

Programmatic ad revenue is split 60/40 in the publisher's favor: the venue/publisher keeps 60%, Trillboards keeps 40%. This transparent, performance-based model removes the barrier to entry for deploying advanced spatial attribution, allowing networks of all sizes to upgrade their infrastructure without upfront capital expenditure.

Proprietary Network Scale

To understand the impact of this architecture, one must look at the sheer scale of the deployment. In 2026, there are 1,154,000+ screens reachable through Trillboards and 29,000+ screens registered on our own network. This unprecedented scale creates a global, decentralized sensor network that maps consumer behavior across the physical world with the same fidelity that web analytics map the digital world.

This massive footprint spans 40 countries and 3,267 distinct cities, covering 38 venue categories under the IAB OpenOOH taxonomy. From EV charging stations in Oslo and smart city kiosks in Tokyo, to grocery store endcaps in Chicago and pharmacy checkout screens in London, the network captures a diverse, global dataset of human movement and attention. It represents a fundamental shift in how physical screens are aggregated, moving away from fragmented, localized networks into a unified, globally targetable ecosystem.

On a technical level, there are 194 IAB audience segments observed in live impressions in a trailing 30-day window. This means buyers can target highly specific cohorts,such as "Fitness Enthusiasts," "Business Travelers," or "Primary Grocery Shoppers",based on deterministic physical presence rather than inferred digital behavior. Furthermore, there are 4,000+ screens that measured their audience by camera or nearby-phone radio in a single day, creating a massive sensing census that buyers target as segtax=600 audience segments. This segtax=600 classification ensures that DSPs know they are bidding on hard, sensor-verified audience data, not modeled lookalike audiences.

The Authoritative Data Practice

The true power of spatial attribution lies in the rigorous, privacy-first data pipeline. Screens measure the people and devices near them. A screen keeps a record for each person its camera detects: estimated age range, gender, dominant emotion, dwell time and time spent looking at the screen. This data is computed entirely at the edge, ensuring that raw visual data is instantly abstracted into anonymous metadata.

On the full sensing profile it also computes a face-identity template (a vector of 512 numbers) that Trillboards uses to recognise the same person across screens and days, and Trillboards links each person to the device keys of the phones seen with them and to mobile advertising IDs. This 512-dimensional latent space embedding is a mathematical representation of facial geometry, not a photograph. It cannot be reverse-engineered into an image, but it allows the system to deterministically understand frequency capping and cross-venue visitation (e.g., knowing that the same anonymous profile visited a coffee shop screen at 8 AM and a gym screen at 5 PM).

Selected camera frames go to a cloud model (Google Vertex AI) that estimates ethnicity, age band, gender, social composition, purchase-intent stage, activity, attire, carried items, lifestyle segments and mood. By leveraging Google Vertex AI's massive multimodal capabilities, the platform can infer deep contextual insights,such as recognizing that a shopper is carrying a competitor's shopping bag, wearing athletic gear, or exhibiting high-intent browsing behavior in a specific aisle.

Trillboards keeps those frames for 400 days, and the frames captured between 5 May and 21 June 2026 are kept as a model-evaluation set with no scheduled deletion. This specific date range serves as a critical baseline for ongoing machine learning model training, ensuring that the AI's accuracy in demographic and contextual estimation remains rigorously benchmarked against a verified historical dataset. Speech is transcribed on the device and Trillboards stores the transcript. No video or audio recording is stored or transmitted. This on-device NLP (Natural Language Processing) extracts keyword intent from ambient conversation (e.g., a customer asking a pharmacist about allergy medication) without ever violating wiretapping or privacy laws, as the raw audio is instantly destroyed.

Screens send Trillboards the Bluetooth, Wi-Fi and local-network identifiers nearby phones broadcast, and their GPS position; Trillboards turns radio and GPS sensing on by default. By capturing these hashed MAC addresses and radio signals, the platform creates a deterministic bridge between the physical screen exposure and the consumer's digital device. Bid requests carry the screen's IP address, location and live audience measurements, providing DSPs with the ultimate contextual payload.

Call the companies that receive bid requests DOOH exchanges and demand-side platforms. Trillboards sells and shares audience measurement and location data with its data partners, including 375ai and Intuizi. These partnerships allow the deterministic spatial data to be ingested into broader omni-channel identity graphs, enabling advertisers to retarget a consumer on their mobile device or CTV 15 minutes after they were exposed to a physical DOOH screen.

Measuring In-Store ROAS Tracking Deterministically

With deterministic audience data flowing from the edge, the next step is calculating a verifiable return on ad spend. Modern platforms route this localized exposure data into secure Data Clean Rooms (DCRs). A Data Clean Room is a secure, cryptographic environment where two parties (e.g., the DOOH network and the retailer) can match their datasets without ever exposing the underlying PII to each other.

By matching anonymous, timestamped screen-exposure events with actual POS transactions, advertisers can measure incremental sales lift against holdout control groups. If the 512-number vector and hashed mobile ID from the screen exposure match the hashed loyalty card ID or digital wallet used at the checkout register 10 minutes later, a deterministic conversion is logged. This calculates an auditable, precise return on ad spend that satisfies even the most rigorous performance marketing CFOs.

Platforms like Doohlabs and Ariadne emphasize that this deterministic matching is the holy grail of retail media. It proves exactly how many shoppers saw a physical ad and subsequently made a purchase. Ariadne's advanced footfall analytics, combined with Doohlabs' retail media monetization engines, rely entirely on this deterministic bridge to prove that physical screens are driving actual, register-ringing sales, not just theoretical brand lift.

The Mathematics of Incrementality

Incrementality testing requires isolating the specific impact of the DOOH exposure. It is not enough to simply say that a shopper saw an ad and bought a product; the advertiser must prove that the shopper would not have bought the product anyway. This is achieved by comparing the conversion rates of the exposed group against a statistically identical holdout group that visited the store but was not exposed to the screen.

The ROAS Formula:

Identify exposed shoppers via deterministic spatial attribution (utilizing the LiDAR and computer vision logs).
Match exposure logs to POS transaction data (via Data Clean Room cryptographic hashing).
Calculate the baseline conversion rate of unexposed shoppers (the control group).
Subtract the baseline from the exposed conversion rate to find incremental lift, and multiply by the product margin to determine true ROAS.

Pro Tip: Always ensure your holdout groups are geographically and temporally aligned with your exposed groups to account for seasonal or regional sales variations. A control group from a Tuesday morning in a suburban store cannot be accurately compared against an exposed group from a Saturday afternoon in an urban flagship location.

OpenRTB 2.6 and Supply-Chain Transparency

To facilitate this massive, real-time data transfer, the industry relies on standardized programmatic protocols. Trillboards features full OpenRTB 2.6 compatibility for screen inventory and audience signals shared with programmatic partners. OpenRTB 2.6 introduced vital enhancements for DOOH, including pod bidding (allowing multiple ads to be auctioned for a single screen playlist loop) and enhanced context signaling.

Every bid request undergoes rigorous validation to prevent the rampant fraud that plagues the digital display ecosystem. The platform executes a 14-check OpenRTB 2.6 supply-chain validation runbook, ensuring sellers.json, ads.txt, and schain compliance in every VAST request. This runbook actively checks for domain spoofing, mismatched publisher IDs, and unauthorized inventory reselling, ensuring a perfectly clean supply path.

This transparency guarantees that DOOH exchanges and demand-side platforms receive verified, fraud-free inventory. It is critical for maintaining buyer trust in physical screen attribution. If buyers cannot trust the supply chain, they will not trust the ROAS metrics.

Furthermore, the system has performed 264,904 creative-level classifications across 44 IAB Content Taxonomy top-level categories, ensuring total brand safety for advertisers. This automated classification ensures that an ad for alcohol is never programmatically served on a screen located near a school, and that competing brands are not played back-to-back in the same ad pod.

Implementing Physical Screen Attribution via SDK

For developers and media enterprises managing 100+ screens, integrating these capabilities requires robust infrastructure. Building a custom programmatic and sensor fusion stack from scratch would take years of engineering and millions in R&D. The Trillboards Partner SDK (@trillboards/ads-sdk) provides programmatic infrastructure-as-a-service, abstracting away the immense complexity of edge-to-cloud data routing.

This SDK supports TypeScript, React, React Native, Flutter, CTV, and server-side implementations. It allows digital signage software companies to easily add ad monetization and spatial attribution to their existing stacks. Whether a developer is building a custom Android application for a kiosk or a web-based React player for a smart mirror, the SDK provides pre-built hooks for impression tracking, sensor data ingestion, and programmatic ad fetching.

This ease of integration is vital for rapidly expanding networks, as Naki Power demonstrates with their power bank sharing kiosk network across europe with screen-equipped charging stations in bars, restaurants, and malls. By dropping the @trillboards/ads-sdk into their existing React Native kiosk application, Naki Power instantly transformed their utility kiosks into a highly monetizable, spatially aware retail media network, generating new revenue streams without altering their core hardware.

Webhook-Driven Event Architecture

The platform utilizes a webhook-driven event architecture to manage device status, impressions, payouts, and audience spikes in real time. Traditional DOOH systems rely on batch processing, where proof-of-play logs are uploaded at the end of the day. Modern spatial attribution demands real-time responsiveness, which is where webhooks excel. This allows developers to build highly responsive, data-driven applications that react to the physical environment instantly.

When a screen detects a specific audience segment, it can instantly trigger a webhook payload. This alerts AI buying agents or dynamic creative optimization (DCO) engines to swap the creative payload. For example, if the computer vision system detects a sudden influx of families with young children, the webhook fires, and the DCO engine instantly swaps the on-screen creative from a luxury car ad to an ad for a family-friendly theme park.

Example Webhook Capabilities:

Real-time impression firing (delivering verified CPAS logs to DSPs within milliseconds of the ad finishing).
Hardware status and heartbeat monitoring (alerting network operators instantly if a screen goes offline or a LiDAR sensor is obstructed).
Audience spike alerts (e.g., sudden influx of a specific demographic, triggering surge pricing in the programmatic auction).
Automated payout notifications (triggering smart contracts or financial ledgers the moment an ad play is verified).

AI Agents and the MCP Server

In 2026, AI agents play a massive role in media planning and buying. Human media planners no longer manually select screens on a map; instead, autonomous AI agents analyze spatial data and execute buys to maximize ROAS. Trillboards provides an MCP server (https://api.trillboards.com/mcp), acting as an AdCP sales agent that an AI agent can plan and buy through. The Model Context Protocol (MCP) standardizes how Large Language Models interact with external data environments.

Developers who build buying agents can leverage the OpenAPI reference to access full REST APIs for device, audience, venue, webhook, and analytics management. This allows AI to programmatically evaluate spatial attribution data and execute buys. An AI agent can query the API to find all screens in a 50-mile radius currently indexing highly for "Business Travelers," negotiate the floor price, and deploy the creative,all in a matter of seconds.

There are three API tiers available to developers to support this automated ecosystem: Basic (200/min), Developer (1K/min, which includes venue intelligence), and Enterprise (5K/min, offering raw exports and an SLA). This tiered approach ensures that indie developers building niche AI tools have access, while enterprise DSPs have the massive throughput and service level agreements required for global trading.

Optimizing the Publisher Experience

While advertisers focus on in-store ROAS tracking, venue owners and publishers need seamless operational tools. A highly advanced sensor network is useless if the venue owner cannot easily manage their playlists, monitor screen health, and forecast their revenue. Trillboards provides comprehensive screen onboarding, playback, venue controls, and reporting through an intuitive, centralized dashboard.

The platform integrates natively with Google Ad Manager (GAM) using IVT-compliant VAST 4.2 tags. It also utilizes the OM SDK (Open Measurement) for MRC-compliant ad verification, ensuring every impression is independently auditable. The integration of VAST 4.2 ensures seamless video playback with advanced error reporting, while the OM SDK allows third-party verification firms (like DoubleVerify or Integral Ad Science) to mathematically certify viewability and fraud-free delivery to the buyer.

Publishers can view published floor CPMs for every audience, access the interactive planner, and manage deal IDs for a buyer's own DSP. This transparency empowers venue owners to maximize their yield. If a publisher notices that their screens are consistently over-indexing for high-value demographics during the lunch rush, they can dynamically adjust their floor CPMs or set up private marketplace (PMP) Deal IDs to sell that premium inventory directly to specific brands.

Connecting to Demand

It is crucial to reiterate that Trillboards is the measurement and aggregation layer. SSPs and exchanges are the complementary partners that connect this premium, sensor-verified inventory to programmatic buyers. Trillboards provides the "brains" and the "eyes" at the edge, while the SSPs provide the vital plumbing to the global programmatic marketplace.

By providing DOOH exchanges and demand-side platforms with rich, deterministic data, the entire ecosystem benefits. Buyers get verifiable ROAS, and publishers get higher fill rates and optimized eCPMs. When a DSP knows with 100% certainty that their ad is being viewed by their exact target audience, they are willing to bid significantly higher, driving up yield for the venue owner.

This symbiotic relationship is what makes modern spatial attribution DOOH so powerful. It aligns the incentives of the physical venue, the measurement platform, the programmatic exchange, and the final advertiser. Everyone wins when the data is deterministic, privacy-safe, and instantly actionable.

Conclusion and Next Steps

The era of estimated out-of-home measurement is over. The days of selling physical screen inventory based on generalized municipal traffic data or flawed mobile location polygons have been permanently replaced by mathematical certainty. By combining LiDAR audience measurement, computer vision advertising metrics, and rigorous data clean room matching, the industry has finally achieved deterministic in-store ROAS tracking.

For CTOs, architects, and media buyers, the mandate is clear: adopt sensor-fused spatial attribution or fall behind. The infrastructure is available, the programmatic pipes are standardized via OpenRTB 2.6, and the AI buying agents are ready. Networks that fail to upgrade to deterministic measurement will see their inventory commoditized, while those who embrace sensor fusion will capture the massive influx of performance marketing budgets entering the physical world.

To begin leveraging deterministic physical screen attribution, developers should review the OpenAPI specifications and integrate the Partner SDK. Advertisers should demand sensor-verified inventory from their DOOH exchanges. The physical world is now fully digitized, measurable, and programmatic. It is time to build the future of spatial advertising.