The Problem with Programmatic as a Foundation
Programmatic advertising was built to solve how to price and place a digital ad impression in milliseconds. This it does extremely well, however it is downstream of media budget channel allocation. Programmatic decisions are limited to what is visible within digital, clicks, pixels, impressions and raw conversion APIs. Advertising effectiveness principles are not being applied at this granularity. The result is two decades of systematic undervaluation of brand-building media, over-indexing on last-touch attribution, and the effective exclusion of broadcast, audio, print, and out-of-home from automated planning and buying workflows.
Media planning and buying decisions, handled by agentic advertising, have the opportunity to do better if they adopt well-established foundation of media planning principles. To do so the industry must not repeat mistaking speed and granularity for marketing effectiveness. Instead it must design for the full scope of how advertising works within marketing.
Objectives First
Every media campaign begins with two interdependent objectives: grow sales and build brand equity. They operate simultaneously and reinforce each other. Product awareness, interest and purchase intent are driven by marketing. Sales are generated either in the short or long term then when product performance meets brand position, long term brand equity is built.
Media and creative are the means to achieve both. They are not objectives in themselves, they are instruments whose value can be judged against interim and eventual business outcomes.
This distinction matters enormously for how we design agentic planning and buying systems. A protocol limited to unverified audiences, impression availability and creative specifications is defining minutia that doesn’t aggregate into a full picture. A truly effective agentic media system must include the entire chain, from business objectives, through media and creative best practice such as reach and frequency management, through creative impact testing to sales and brand contribution while enabling governance and auditing of transactions. It must recognize trade body governance of audited and accredited systems and measures and joint industry committee currency and metrics where available.
To get there requires geo-structured large population models of virtual IDs and marketing mix modelling analysis. Both unify digital and traditional media in ways that personal IDs used in programmatic cannot, the former (VIDs) at the planning and modelling stage, the latter (MMM) at the outcome analysis stage.
Large population models, MMM and Geography as unifiers.
While campaigns can be national, a focus on local markets, each with its own media availability, and audience composition is more specific. The national view is simply an aggregation of markets which are advertiser specific.
A buying agent optimizing a national budget needs to reason at the market level by sourcing and choosing attention getting and memory forming media suppliers to build reach in the geographies where product distribution exists, modelling cross media reach and frequency against the local audience, while measuring sales contribution through a geo-hierarchical continuously updated MMM. The buying agent may create and deploy creative variations for language, product availability while testing creative performance in matched markets. The national picture emerges from laddering up those market-level campaigns and models.
Regional MMM models capture local conditions, distribution, competitive activity, and media weight, aggregated into a national view that preserves the signal rather than averaging it away.
Agentic advertising must support this architecture natively. Geography-as-campaign-dimension means the protocol needs to handle market-level briefs, market-level supplier discovery, cross media reach build, frequency management and market-level reporting that rolls up coherently to a national view.
The Planning Gap
An agentic planning system must solve a specific problem before a single brief is written: unduplicated reach in each geography.
This requires three things working together. First, knowledge of the target audience, who they are as defined by the advertiser. Second, the size of that target audience relative to the total in each geography. Third, the duplication profile across media suppliers expressed in a universally accepted across all media formats.
Without these three inputs, a buyer agent cannot make decisions that build reach. It can assemble a list of available suppliers and their prices but it cannot tell you whether spending $500K across five suppliers in a mid-sized market produces 60% reach or 35% reach against the target.
But reach and frequency is only half the planning problem. The other half is sales yield. This is where the buying agent must be able to access marketing mix models. A well-constructed portfolio of MMM’s produces a media yield curve for each channel and possibly supplier defining the relationship between media weight and sales contribution, net of base sales, price promotion, and distribution effects. Those reach and yield curves tell the planning agent not just how many people a channel reaches, but guide the buying agent to achieve sales forecasted by the MMM.
The planning gap is therefore a dual optimization problem. The plan must balance the reach and frequency curve against the sales yield curve from the MMM to ensure the buying agent is producing the most incremental sales and protecting future sales through brand building.
Today's agentic advertising protocols begin after this work is done. They assume the consequential decisions have been made, the media plan is complete and the buying agent’s task is simply executing the plan. Without planning capabilities, the buy side agents aren’t very intelligent.
Dynamic placement of ads is not required for effectiveness.
A very significant portion of media consumed by audiences is not dynamically placed by an ad server. Broadcast television, radio, out-of-home, print and sponsorships do not operate on impression-level tracking, real-time bidding, or ID based targeting.
Yet these channels are powerful drivers of brand awareness, creative impact, purchase intent and sales. In addition to their inherent benefits, they affect the performance of other media types especially downstream digital channels such as search and social.
Currently, non-ad-servable media operates on different rules. Delivery is contractual, not impression-based. Measurement is though a panel possibly a joint industry committee. Optimization happens between flights, not within them. Creative is trafficked to a station or publisher, not served dynamically from a tag.
An agentic media protocol must serve both dynamically and non-dynamically place ad environments to serve the planning and buying agents cross media scope and the MMM’s evaluation. To consistently represent all media fairly, the large population model must exist at the individual level and aggregate up into household, postal code, FSA, city, province, national and eventually international.
Attribution & Measurement
Crediting a media supplier with a sale requires first removing every other explanation for why that sale occurred. Skip that step and you get attribution that misleads the buying agent. Buying agents must work with attributed conversion events instead of gross conversions.
Enabling buying agents to work with last-touch attribution, view-through conversion, and multi-touch attribution will continue the systematic overvaluation of bottom-funnel digital channels and an equally systematic undervaluation of brand-building efforts.
The correct methodology is Marketing Mix Modelling. A well-constructed MMM decomposes total sales into their constituent drivers: base sales reflecting brand equity and distribution, promotional lift that is temporary and price-driven, and media contribution that is channel-specific and weight-dependent. Only after that decomposition can a media supplier be credibly credited with sales outcome.
But MMM introduces its own planning requirements. The model must match the geographic structure which planning and buying operates in. It must be updated regularly enough to reflect changing market conditions. And critically, the yield curves it produces must feed back into the planning system so that budget allocation decisions are informed by measured effectiveness, not assumed effectiveness.
This creates an all inclusive closed loop that agentic advertising is uniquely positioned to close: plan against reach and yield curves, execute across all media suppliers, measure contribution through geo-hierarchical MMM, test creative variations in matched geographies, feed results back into the next planning cycle through use of priors in MMM. Each campaign makes the next one smarter.
AAMP, AdCP and Joint Industry Committees
Both The IAB Tech Lab and Agentic Advertising Org are proposing and building protocols to support agentic planning, buying, selling and evaluation of advertising campaigns. It is presumed that IAB Tech Lab’s initiative exploits IAB’s Project Eidos direction as guidance. Additionally there are private initiatives such as the Trade Desk’s Koa agents.
Standards work because they are universally applicable and useful. For this reason, Joint Industry Committees were developed in each country to handle media currency and other broadly used metrics. While the US is not a market where JICs are active, all the other major media markets are. For this reason, JICs will need to engage deeply with the develop of agentic planning, buying and evaluation standards and may need to develop country specific protocols over and above a base.
Additionally, JICs face a question about cross media audience definitions. Countries where JICs are active often have multiple JICs each addressing a specific media type. Agentic audiences definitions need to be standard across media type. Either the JICs sort out a way to talk to each within the one agentic protocol or merge their data to enable cross media reach and frequency calculations. Differences in media quality, consumer attention and other qualitative aspects of each medium need to be represented for the agent’s decision making.
A Reference Architecture for Agentic Media
A holistic agentic media system has five distinct layers. Each is necessary. None can substitute for another.
1. Objective Setting
The system begins with business objectives — sales targets and brand equity goals, expressed at the market level. Each geography where the brand operates is treated as a distinct campaign with its own targets, media suppliers, creative, reflecting local sales baselines, competitive conditions, and brand health.
2. Planning
Against those objectives, the planning layer solves two problems simultaneously. The reach and frequency problem: which combination of available media suppliers, in each market, produces unduplicated reach against the target audience universe at the required frequency? The yield problem: which channels and markets produce the highest incremental sales per dollar invested, as determined by the regional MMM?
The output of the planning layer is not a brief. It is a market-level plan, a recommended distribution of budget across suppliers, geographies, and channels, grounded in both audience measurement and measured sales effectiveness. The agentic brief is derived from that allocation, not the other way around.
3. Execution
The execution layer handles the full range of media types — ad-servable and non-ad-servable. For digital inventory, it manages programmatic and direct-sold placements, creative trafficking, and impression-based delivery reporting. For broadcast, audio, out-of-home, and print, it manages contractual commitments, manual creative trafficking, panel-based delivery confirmation, and makegood handling when contracted delivery is not met.
Creative is not an afterthought in this layer. The execution system must know which creative assets are available in which formats, which formats are accepted by which suppliers, and how creative weight interacts with frequency to manage wear-out across the campaign flight.
4. Measurement
The measurement layer operates in parallel with execution and feeds back into planning. Regional MMM models decompose sales into base, promotional, and media-attributed contribution for each market. Reach and frequency delivery is tracked against the large population model-based plan. Creative effectiveness is tested in matched market and recall metrics. All measurement is methodology-declared — every metric explicitly states how it was produced, so the planning system knows which signals to trust and how.
5. Optimisation
The optimisation layer closes the loop. It compares actual delivery against the reach and frequency plan and flags markets where adjustments are needed — shifting creative weights, reallocating budget between suppliers, adjusting frequency targets. It feeds, updates and uses MMM results to inform planning and buying system.
This is not a vision for a distant future. Every component of this architecture exists today, in some form, in the tools that sophisticated media planners and analysts use. What does not exist is a protocol that connects them — that allows a buyer agent to reason across all five layers, communicate with suppliers across all media types, and produce a closed loop from business objective to measured outcome and back.
Recommendations for the Agentic Roadmap
The gap between what agentic advertising is today and what a holistic agentic media system requires is significant. It is not, however, unbridgeable. The protocols have the right architectural instincts. What is needed is tripartite governance where advertisers, agencies and broadcast/publishers determine the standards of protocols. Audit systems, which may be agentic themselves, also need to designed.
The following recommendations are offered in priority order.
1. Define a Pre-Brief Exploration Primitive
Agentic must support a planning mode that operates upstream of the brief — allowing a buyer agent to query available suppliers, model unduplicated reach against a target audience universe in a defined geography, and iterate on budget allocation before any commitment is made. This is not a refinement of the existing brief flow. It is a distinct protocol layer that requires new primitives for audience universe declaration, reach curve modelling, and cross-supplier duplication estimation.
2. Adopt Large Population Models as Cross Media Spine
The protocol must support large populations models which include panel-based audience measurement. This means supporting target audience declaration in composable terms from virtual IDs to households, to postal codes, FSAs, cities, CMAs, geography-level universe sizing, and delivery reporting expressed in reach and frequency against the declared target, not just impressions delivered. The roles of JICs, their data, media currency and audited accreditations must be included.
3. Support Non-Ad-Servable Media Natively
Contractual delivery, manual trafficking confirmation, panel-based delivery reporting, and makegood management must be first-class protocol primitives — not workarounds. Non-ad-servable media is not a niche. In many markets it is the majority of available inventory. A protocol that cannot handle it is a protocol for a subset of media.
4. Introduce Methodology-Declared Metrics
Every performance metric flowing through the agentic protocol should declare its methodology — pixel-fired, panel-estimated, MMM-derived, incrementality-tested. Without this declaration, a buyer agent cannot know whether a conversion signal should inform future allocation decisions.
5. Support Geography as a Campaign Dimension
Market-level briefs, market-level supplier discovery, and market-level reporting that rolls up to a national view must be supported natively. The protocol should allow a buyer agent to manage a single brand campaign as a portfolio of market-level campaigns — each with its own suppliers, creative weights, reach objectives, and measurement — while maintaining coherent national reporting.
6. Close the MMM Loop
The provide_performance_feedback task must be extended to support modeled attribution coefficients, not just impression-level signals. The protocol should allow MMM-derived yield curves to be fed back into the planning layer so that budget allocation decisions in the next campaign reflect measured effectiveness from the last.
7. Position Agentic as the Investment Layer, Not the Execution Layer
This is as much a strategic recommendation as a technical one. Agentic’s differentiated value is not faster programmatic execution, it is the ability to reason about media investment holistically, across all channels and markets, from business objective to measured outcome. Every protocol design decision should be evaluated against that ambition. Features that optimize impression-level execution belong at the impression layer.
8. In Silico Testing of the Protocol, the Agents and the Outputs
Agentic protocols need to be throughly tested in experiments before being fully rolled out. Specific cases such as a national brand launch, grocery retail, telco customer acquisition, online retailer, travel platforms and many others need to run through their paces to examine if the agentic system’s plans make sense before unleashing them as buyers.
Conclusion
Agentic advertising has the opportunity to automate the right things only if it is designed from the business objective down, not from the digital impression up. The choices being made now will determine whether agentic advertising becomes a genuinely new strategic architecture for media and creative investment, or a more sophisticated version of a RTB.