IAB Tech Lab opens agentic framework and on-device AI for publishers

The Interactive Promoting Bureau Expertise Laboratory launched its Agentic RTB Framework model 1.0 for public touch upon November 12, 2025, whereas concurrently selling on-device synthetic intelligence as a publisher-focused privateness resolution by way of a technical blog post. The twin bulletins place IAB Tech Lab on the intersection of containerized agent deployment and on-device processing applied sciences that would reshape programmatic promoting infrastructure.

The Agentic RTB Framework specification v1.0 enters a public remark interval extending by way of January 15, 2026. The framework defines standardized necessities for deploying agent-driven containers inside OpenRTB environments, addressing how containerized brokers can take part in real-time bidding infrastructure with minimal latency impacts. The specification establishes technical protocols for container runtime habits and defines utility programming interfaces for bidstream mutation, enabling delegation of processing duties to autonomous brokers working inside host platforms.

Miguel Morales, Director of Addressability & Privateness Enhancing Applied sciences at IAB Tech Lab, authored the framework announcement. The Container Challenge Working Group developed the specification by way of collaboration with Index Trade, OpenX, The Commerce Desk, and Chalice. Collaborating organizations embrace Amazon Adverts, Netflix, Yahoo, Paramount, Optable, HUMAN Safety, Magnite, PubMatic, WPP Media, and Foundation Applied sciences.

Concurrent with the framework launch, IAB Tech Lab printed a weblog submit on November 11 that includes evaluation from Anish Aravindakshan of Verve inspecting on-device cohort modeling. The submit argues that on-device synthetic intelligence represents an untapped alternative for publishers searching for privacy-compliant focusing on methodologies. On-device cohort modeling processes viewers segmentation straight on consumer gadgets quite than transmitting individual-level information to exterior servers, respecting platform privateness norms with out counting on third-party cookies.

How the know-how works

Understanding these applied sciences requires inspecting two distinct approaches to promoting automation. The ideas differ considerably however each deal with elementary challenges in digital promoting: velocity, privateness, and automation.

Containerized brokers perform equally to specialised staff dropped into an meeting line. Conventional programmatic promoting requires corporations to construct customized connections between their programs and each promoting platform. A fraud detection firm eager to verify commercials throughout 50 totally different platforms would want to construct 50 separate integrations. Containerized brokers get rid of this complexity. The agent arrives as a self-contained package deal carrying all the things wanted to carry out its job—fraud detection algorithms, model security guidelines, or information enrichment logic. The internet hosting platform gives the workspace and instruments, whereas the agent performs its specialised process. When an commercial alternative arises, the container springs into motion, analyzes the scenario in milliseconds, and returns its verdict with out slowing down the public sale. The standardized container format means the identical fraud detection agent works identically whether or not deployed on a supply-side platform, demand-side platform, or writer server.

On-device cohort modeling operates by way of a completely totally different mechanism targeted on privateness preservation. Present promoting usually works like this: web sites and functions transmit consumer habits information to central servers the place algorithms analyze tens of millions of customers concurrently, figuring out patterns and assigning commercials. On-device processing inverts this mannequin. The evaluation occurs on the smartphone, pill, or tv itself. A information utility on a smartphone may discover its proprietor reads automotive articles ceaselessly, watches residence enchancment movies, and engages with journey content material. As an alternative of sending this detailed behavioral historical past to exterior servers, the machine’s processor runs a light-weight algorithm figuring out the proprietor in all probability belongs to teams serious about automobiles, residence items, and trip packages. Solely these broad group memberships—cohorts—go away the machine. Advertisers obtain a sign indicating “this machine proprietor belongs to the automotive fanatic cohort” with out studying which particular articles had been learn, after they had been accessed, or any figuring out particulars concerning the particular person. The cellphone does the pondering, retains the delicate particulars native, and shares solely the conclusions.

The technical magnificence lies in matching processing energy to privateness necessities. Trendy smartphones include processors extra highly effective than desktop computer systems from a decade in the past. This computational functionality beforehand sat idle between utility launches. On-device processing harnesses that dormant energy for privacy-preserving evaluation. Machine studying fashions sufficiently small to suit on shopper gadgets can determine pursuits and preferences by inspecting native utility utilization patterns. A streaming tv machine is aware of which exhibits its proprietor watches, which genres get deserted mid-episode, and when viewing usually happens. Processing this data regionally produces cohort assignments—”night drama viewer,” “sports activities fanatic,” “household content material shopper”—with out transmitting the precise viewing historical past past the machine. Publishers profit as a result of they management the algorithm figuring out cohort assignments. A information writer might develop proprietary curiosity detection superior to generic third-party options, differentiating their stock by way of higher viewers understanding whereas sustaining consumer privateness.

Each applied sciences deal with bottlenecks in programmatic promoting infrastructure. Containerized brokers remedy the combination complexity drawback: how do dozens of specialised service suppliers effectively connect with tons of of promoting platforms with out constructing 1000’s of customized connections? On-device processing solves the privateness paradox: how do publishers monetize by way of focused promoting when platforms prohibit monitoring particular person customers throughout functions and regulators prohibit private information assortment? The options function at totally different factors within the promoting workflow however pursue the identical objective—sustaining programmatic promoting effectiveness as technical and regulatory constraints get rid of earlier methodologies.

Measuring efficiency in opaque programs

The shift towards autonomous brokers and on-device processing creates measurement challenges absent from conventional promoting infrastructure. Advertisers accustomed to granular efficiency information face a elementary query: how do you confirm effectiveness when synthetic intelligence makes choices autonomously and privateness protections obscure the mechanisms producing outcomes?

Conventional measurement relied on observable inputs and outputs. An advertiser might study which key phrases triggered commercials, which viewers segments obtained impressions, what inventive variations carried out finest, and the way bidding methods affected prices. Transparency enabled optimization. Marketing campaign managers adjusted based mostly on clear cause-and-effect relationships seen in reporting interfaces. Autonomous brokers disrupt this mannequin. When an AI system decides which stock to buy, which audiences to focus on, and what bids to submit—all inside milliseconds and throughout tens of millions of selections each day—human remark turns into impractical. The query shifts from “what did the system do?” to “did the system obtain desired outcomes?”

Final result-based measurement focuses on enterprise outcomes quite than course of metrics. An advertiser working campaigns by way of autonomous brokers cares whether or not gross sales elevated, value per acquisition declined, or return on promoting spend improved. The particular bid changes, viewers expansions, or stock alternatives that produced these outcomes matter lower than the combination efficiency. This strategy mirrors how advertisers consider human media patrons. No person calls for minute-by-minute documentation of each choice a marketing campaign supervisor makes. Efficiency opinions study outcomes: did the campaigns hit targets? Autonomous brokers obtain comparable analysis based mostly on outcomes quite than granular choice logs.

On-device processing introduces totally different measurement constraints. Cohort assignments occur regionally with out transmitting the underlying behavioral information that knowledgeable these assignments. An advertiser is aware of an impression reached the “automotive fanatic” cohort however can’t confirm how the machine decided that classification. Was the cohort project correct? Did the algorithm correctly determine real curiosity alerts or misread unrelated behaviors? Validation turns into probabilistic quite than deterministic. Advertisers should infer cohort high quality from marketing campaign efficiency. If automotive commercials delivered to the “automotive fanatic” cohort generate robust conversion charges, the cohort project course of in all probability works accurately. Poor efficiency suggests both defective cohort logic or misaligned inventive messaging.

Management teams present one verification methodology. Advertisers can examine campaigns utilizing on-device cohorts towards campaigns utilizing conventional focusing on strategies or no focusing on in any respect. A statistically important efficiency enchancment demonstrates that on-device cohorts add worth whatever the particular algorithms producing these cohorts. Incrementality testing measures the causal affect of promoting publicity by randomly withholding commercials from a portion of the target market, then evaluating outcomes between uncovered and unexposed teams. The methodology works identically whether or not cohorts derive from server-side processing or on-device algorithms. Outcomes point out whether or not the focusing on strategy successfully reaches responsive audiences.

Aggregated reporting replaces individual-level monitoring in privacy-preserving programs. As an alternative of detailed consumer journeys exhibiting precisely which commercials a particular particular person noticed earlier than buying, advertisers obtain abstract statistics indicating how many individuals from every cohort transformed inside numerous time home windows. The aggregation preserves privateness whereas enabling efficiency evaluation. An advertiser may study that 3.2 % of impressions delivered to the “residence enchancment fanatic” cohort resulted in conversions inside 30 days, in comparison with 1.8 % for the “basic curiosity” cohort. This information suffices for funds allocation choices with out requiring particular person consumer monitoring.

Attribution modeling turns into more difficult when autonomous brokers management marketing campaign execution throughout a number of platforms concurrently. Conventional attribution assigns credit score for conversions to particular touchpoints within the buyer journey. Multi-touch attribution may credit score 40 % to the preliminary consciousness commercial, 30 % to mid-funnel consideration content material, and 30 % to the ultimate retargeting impression. Autonomous brokers managing campaigns holistically throughout channels complicate this evaluation. The agent may concurrently modify search bids, increase social media focusing on, and enhance video commercial frequency. Isolating which particular motion drove incremental conversions turns into tough when the agent makes interdependent choices treating the marketing campaign as an built-in system quite than discrete channels.

Media combine modeling gives an answer for measuring agent-driven campaigns. This statistical strategy analyzes historic relationships between promoting investments and enterprise outcomes, accounting for seasonality, aggressive exercise, and exterior components. The fashions estimate how a lot every advertising and marketing channel contributed to general outcomes with out requiring individual-level attribution. An advertiser utilizing autonomous brokers throughout search, social, show, and video can make use of media combine modeling to know channel effectiveness and optimum funds allocation. The strategy treats agent-controlled campaigns as inputs to the mannequin, evaluating mixture efficiency quite than dissecting particular person choices.

Benchmark comparisons set up efficiency expectations for AI-driven programs. An advertiser implementing on-device cohort focusing on for the primary time lacks reference factors for evaluating outcomes. Is a 2.5 % conversion charge good or poor for that cohort methodology? Business benchmarks present context. If comparable advertisers obtain 2.8 % conversion charges utilizing comparable approaches, the two.5 % end result suggests room for enchancment. If the trade common sits at 1.9 %, the two.5 % efficiency signifies efficient implementation. Benchmarking requires cautious comparability. Cohort definitions, inventive high quality, provide competitiveness, and seasonal timing all have an effect on outcomes. Direct comparisons work finest when advertisers share detailed implementation approaches, although aggressive issues usually restrict transparency.

Certification and audit frameworks might deal with belief deficits in opaque AI programs. IAB Tech Lab’s framework improvement consists of necessities for brokers to declare their intents and modifications. This transparency allows verification that brokers behave as specified. An advertiser deploying a model security agent can audit whether or not the agent really filters stock in line with acknowledged standards or permits violations by way of. Unbiased auditors may certify that on-device processing implementations genuinely hold information native quite than transmitting detailed consumer data to exterior servers regardless of privateness claims. Verification requires technical capabilities past most advertisers’ inner sources, creating alternatives for specialised audit companies.

The measurement problem displays broader tensions between automation effectivity and operational transparency. Autonomous programs promise superior efficiency by way of processing capabilities exceeding human evaluation. Machine studying algorithms determine patterns in billions of information factors, optimizing towards targets sooner than handbook marketing campaign administration. This functionality requires delegating choice authority to the AI system. Extreme constraints or reporting necessities undermine the effectivity beneficial properties motivating automation adoption. Advertisers should stability the need for clear, explainable choices towards the efficiency benefits of permitting AI programs substantial autonomy. Totally different organizations will strike totally different balances based mostly on threat tolerance, regulatory necessities, and aggressive positioning.

The on-device strategy contrasts with server-side processing fashions which have dominated programmatic promoting. Conventional strategies mixture consumer information on centralized servers the place focusing on choices happen earlier than commercial supply. On-device processing retains delicate consumer data native to the machine whereas enabling contextual and behavioral focusing on by way of cohort assignments computed straight on smartphones, tablets, or linked tv gadgets.

Technical implementation requires processing energy ample to execute machine studying fashions on shopper {hardware}. Trendy cell gadgets and streaming gadgets possess computational capabilities that had been beforehand accessible solely in information middle environments. Graphics processing models embedded in smartphones allow real-time inference from neural networks skilled to determine consumer pursuits and buying intent based mostly on utility utilization patterns, content material consumption, and machine exercise.

The cohort modeling methodology teams customers with comparable traits into segments with out revealing particular person identities. A tool may decide that its proprietor belongs to cohorts serious about out of doors recreation, residence enchancment, and premium automotive merchandise based mostly on utility interactions and content material engagement. These cohort assignments grow to be accessible to promoting programs with out transmitting personally identifiable data past the machine, sustaining privateness whereas enabling related commercial supply.

Platform operators together with Apple and Google have carried out privateness frameworks limiting cross-application monitoring and limiting information assortment practices. Apple’s App Monitoring Transparency requires express consumer permission for cross-app monitoring, leading to opt-in charges beneath 25 % amongst iOS customers. Google introduced plans to deprecate third-party cookies in Chrome a number of occasions earlier than halting implementation in July 2024, creating uncertainty round different focusing on methodologies.

On-device processing gives publishers management over information processing whereas complying with platform restrictions. Publishers can implement proprietary algorithms that analyze consumer habits inside their functions or web sites, producing cohort assignments that inform commercial choice with out sharing uncooked behavioral information with advertisers or intermediaries. This strategy preserves the writer’s information asset whereas enabling programmatic monetization.

The Agentic RTB Framework addresses totally different technical challenges associated to autonomous decision-making inside bidstream processing. Brokers deployed by way of the framework can modify bid requests, improve information alerts, apply model security guidelines, or execute fraud detection algorithms. Container-based deployment allows unbiased software program distributors to distribute specialised processing logic that integrates into supply-side platforms and demand-side platforms with out requiring customized integration work.

5 crucial necessities govern agent deployment underneath the framework. Brokers should take part within the core bidstream, specializing in entities transacting in real-time auctions. Every agent should declare particular intents and any public sale modifications, permitting orchestrating entities to simply accept or reject modifications. Container construction follows Open Container Initiative compliance, manageable by way of Kubernetes, Docker Compose, or cloud-based programs together with Amazon’s Elastic Container Service. Efficiency structure achieves low to sub-millisecond response occasions by way of container-based deployment dropped into host networks. Containers talk by way of high-performance messaging mechanisms constructed on gRPC, OpenRTB Patch, and Mannequin Context Protocol suitable with real-time bidding necessities.

IAB Tech Lab’s framework improvement follows the group’s 2025 technical standards roadmap introduced January 29, which outlined 31 new specs or updates addressing privateness laws, information dealing with, and streaming media challenges. The group accomplished 79 initiatives in 2024, developed with enter from over 800 member corporations.

The emphasis on writer advantages within the on-device synthetic intelligence weblog submit displays broader trade issues about writer monetization amid platform modifications. IAB Tech Lab launched its Content Monetization Protocols working group on August 20, 2025, responding to proof that synthetic intelligence applied sciences threaten digital publishing financial sustainability. AI-driven search summaries scale back writer site visitors by 20-60 % on common, with area of interest websites experiencing losses as much as 90 %.

Anthony Katsur, Chief Govt Officer at IAB Tech Lab, emphasised the group’s function in technical requirements improvement. The requirements physique beforehand challenged Prebid transaction ID implementation modifications in August 2025, declaring that modifications materially violated OpenRTB specs and risked “undermining the integrity and consistency of open technical requirements.” That enforcement motion adopted established procedures for technical specification evolution.

The convergence of containerized agent deployment and on-device processing represents parallel approaches to promoting know-how challenges. Containerized brokers allow centralized processing logic deployment throughout distributed infrastructure, whereas on-device processing distributes computation to edge gadgets. Each methodologies deal with scalability, latency, and privateness necessities that constrain conventional server-based architectures.

Business motion towards agentic synthetic intelligence has accelerated all through 2025. McKinsey information signifies $1.1 billion in fairness funding flowed into agentic AI throughout 2024, with job postings associated to the know-how growing 985 % from 2023 to 2024. Six companies launched Ad Context Protocol on October 15, betting that open-source technical requirements might allow AI brokers to speak throughout platforms and execute promoting duties autonomously.

Amazon launched agentic capabilities throughout its promoting platform on November 11, reworking instruments from question-answering programs into autonomous brokers that monitor accounts, optimize stock, and handle campaigns. The system processes pure language directions to execute advanced workflows together with marketing campaign creation, viewers focusing on, and analytics question era.

The Agentic RTB Framework specification consists of pattern code within the IAB Tech Lab GitHub repository. Organizations serious about reviewing technical specs and offering suggestions can entry documentation by way of the IAB Tech Lab web site. The general public remark interval concluding January 15, 2026, gives alternative for trade members to contribute to the refinement of deployment requirements.

IAB Tech Lab beforehand launched Publisher Advertiser Identity Reconciliation protocol in September 2024 for first-party information matching, Attribution Data Matching Protocol in October 2024 for privacy-preserving conversion measurement, and ID-Less Solutions Guidance in July 2025 for promoting with out conventional identifiers. The framework improvement parallels the group’s Content material Monetization Protocols working group addressing AI-driven search impacts on writer income.

The technical specs matter for advertising and marketing professionals managing programmatic promoting operations. Containerized brokers might automate model security enforcement, fraud detection, and information enrichment duties at the moment dealt with by way of handbook processes or platform-specific implementations. Standardized deployment allows interoperability throughout supply-side platforms and demand-side platforms, lowering integration complexity for unbiased software program distributors growing specialised processing logic.

On-device processing addresses privateness compliance necessities whereas sustaining focusing on capabilities crucial for programmatic monetization. Publishers implementing on-device cohort modeling can differentiate their stock by way of proprietary viewers insights unavailable to opponents counting on third-party information suppliers. The methodology aligns with regulatory frameworks in jurisdictions together with the European Union, California, and different privacy-conscious markets limiting private information assortment and cross-context monitoring.

Implementation challenges embrace computational necessities, mannequin distribution, and measurement verification. Processing machine studying fashions on shopper gadgets consumes battery energy and reminiscence sources that machine producers and working system suppliers could prohibit. Distributing up to date fashions to tens of millions of gadgets creates bandwidth and model management complexity. Verifying that on-device processing happens as declared quite than transmitting uncooked information to servers requires attestation mechanisms and audit capabilities.

The IAB Tech Lab bulletins place publishers on the middle of privacy-preserving promoting know-how evolution. On-device processing provides publishers management over information processing whereas sustaining consumer privateness. Containerized brokers allow deployment of publisher-specific logic throughout programmatic infrastructure. Each approaches reply to platform restrictions and regulatory necessities reshaping digital promoting technical structure.

Timeline

Abstract

Who: The Interactive Promoting Bureau Expertise Laboratory, a nonprofit consortium growing digital promoting technical requirements, launched specs developed by its Container Challenge Working Group. Anish Aravindakshan from Verve contributed evaluation on on-device cohort modeling printed within the IAB Tech Lab blog.

What: IAB Tech Lab launched Agentic RTB Framework model 1.0 for public remark, establishing standardized specs for deploying containerized brokers inside real-time bidding infrastructure. Concurrently, the group printed evaluation selling on-device synthetic intelligence for publisher-focused privacy-compliant viewers focusing on with out third-party cookies.

When: The Agentic RTB Framework specification was launched on November 12, 2025, with a public remark interval extending by way of January 15, 2026. The on-device AI weblog submit was printed on November 11, 2025.

The place: The specs have an effect on international programmatic promoting infrastructure, with collaborating organizations together with Index Trade, OpenX, The Commerce Desk, Amazon Adverts, Netflix, Yahoo, Paramount, Optable, HUMAN Safety, Magnite, PubMatic, WPP Media, and Foundation Applied sciences. Implementation targets supply-side platforms, demand-side platforms, and writer functions working throughout cell, linked tv, and net environments.

Why: The framework addresses the necessity for standardized containerized agent deployment as agentic synthetic intelligence adoption accelerates throughout promoting know-how, with $1.1 billion in fairness funding throughout 2024. On-device processing responds to platform privateness restrictions and regulatory necessities limiting third-party cookies and cross-application monitoring, providing publishers privacy-compliant focusing on methodologies whereas sustaining programmatic monetization capabilities amid site visitors declines from AI-driven search.