The Decision Layer: Win the Audience Interest Graph. The Transaction Follows.
Part 3 of the RMN/CMN series: why agentic commerce will accelerate commerce media by moving the battlefield upstream, and what the decision engine connecting consumers to brands actually looks like
Just joining? Part 1 mapped the market Retail Media Network, CMN, and FMN. Part 2 explained the infrastructure war the Renters vs. the Owners. This is Part 3: what the winners do as they build/ own that first party audience infrastructure. They stop optimizing only for the transaction. They start shaping consideration, nurturing their audience across “attention surfaces” before any transactions. Because in the agentic era, the transaction is no longer where brands win. It is where the system cashes out everything that happened earlier.
Everyone talking about agentic commerce is asking the wrong question.
They’re asking: “If an AI agent makes purchases on behalf of consumers, do ads still matter?”
And a second question people are quietly asking: “Is influencing that agent ethical or just consumer manipulation with extra steps?”
Both are the wrong frame. They assume the battle still happens at the point of purchase. It doesn’t.
The right question is: “Before an AI agent decides what to recommend/buy whose interests is it acting on? And were those interests earned or manufactured?”
Those interests were formed somewhere. Through content, experience, and brand exposure over time. Agents do not invent desire. They do not create taste from thin air. They act on signals, memory, and prior context. Brands that earn a place in the audience interest graph by being genuinely useful in trusted contexts are not manipulating the agent. They’re being remembered. Trust is not just a moral requirement in this model. It’s a technical one.
The New Funnel
The old marketing funnel: Awareness → Consideration → Decision → Purchase
Every layer had a media vehicle. TV and display owned awareness. Search and social owned consideration. Commerce media owned the purchase moment — the sponsored listing, the digital shelf, the checkout nudge.
Agentic commerce is collapsing the middle. When a consumer says “buy me the best protein powder under $40 that fits my diet” - Consideration and Decision happen inside the AI model. The consumer never visits a product page. Never sees a sponsored listing.
The new funnel: Interest Formation → Agent Activation → Agent-Executed Purchase
That framework only works if three very different stakeholders win at the same time.
Three Stakeholders. One System.
The decision layer works because it aligns three parties who have historically pulled in different directions:
The Consumer wants useful discovery in trusted contexts not creepy tracking across the open web. The deal is implicit: you have my data because I use your service use it to show me things that are genuinely useful. When that deal is honored, brand discovery feels like a recommendation. When it’s violated, trust breaks and that brand gets filtered out of the consumer’s consideration set.
The Brand needs presence before the purchase moment. If a consumer’s AI agent has never encountered a brand in a trusted context, that brand is invisible at the moment of agentic decision. Not just “be on Amazon at the point of purchase” but “be in hospitality, finance, delivery, and travel so our brand is already in the consideration set when the agent acts.”
The CFO has a new measurement problem. It’s not attribution anymore it’s admissibility. Not just: did this spend cause a sale? But: did this brand earn a place in the set the agent was willing to choose from? That requires moving from ROAS to incremental consideration — the measure of whether your brand actually moved from unaware to considered. Harder to measure. The only metric that survives the agentic era.
The rest of this post is about the system that satisfies all three.
What It Looks Like in Practice
Here’s what the new funnel looks like across real attention surfaces.
Sarah is a fitness-conscious consumer. Over the past 90 days, without ever actively searching for a protein brand, she encountered the same brand in three high-trust, high-attention contexts:
On Instacart, while planning her weekly grocery shop, she saw a contextually relevant recommendation for a protein powder that fit her usual basket — plant-based, mid-price, high-rated. She didn’t buy it. But she noticed it.
On United Airlines, during a 4-hour flight, a brand ran a health and wellness ad through Kinective Media while she was in a relaxed, receptive state with nothing competing for her attention. She read it. She still didn’t buy.
In Chase Offers, she saw a cashback deal from the same brand. She added it to her wallet — which told Chase that this consumer was warm, a signal the brand could access through Chase Media Solutions.
Three touchpoints. Three trusted contexts. No transaction yet.
Then, two weeks later, she told her AI agent: “find me a good protein powder, plant-based, under $40.”
Those were not failed impressions. They were interest-forming exposures across trusted attention surfaces.
The agent does not start from zero. The brand that showed up consistently in the right contexts is already inside Sarah’s admissible consideration set. The brand that only bid on the Google keyword never made it onto the starting line.
The transaction may happen in one place. But the decision was built somewhere else.
Why Agentic Commerce Expands Commerce Media, Not Shrinks It
This is why agentic commerce does not kill commerce media. It expands it.
But the value shifts upstream.
The winners will not just be the networks closest to checkout. They will be the networks strongest at interest formation.
Instacart is not just a commerce surface. It is a planning surface. Chase is not just a payment surface. It is a trust surface. Marriott, United, PayPal, and DoorDash are not just channels. They are high-context environments where the consumer is already generating intent, preference, and timing signals before any purchase task is delegated.
That makes them powerful before the transaction. Not just at the transaction.
The old question was: did this placement capture demand?
The new question is: did this environment help build consideration before the agent was activated?
That is a much harder question. It is also the one that matters more.
What Interest Formation Actually Looks Like
Three things build brand interest before a purchase decision:
Contextual relevance at the right moment. Marriott showing a Visa offer at check-in. United Airlines offering a Norwegian Cruise Line promotion during a long-haul flight. Instacart surfacing a brand while someone is planning meals. These aren’t interruptions — they’re suggestions arriving when the consumer is receptive. That’s what makes them stick.
The networks with the richest context — travel, hospitality, finance, delivery — aren’t just better at targeting transactions. They’re structurally better at reaching consumers during interest formation windows. That’s a different and more valuable capability.
Repetition across trusted surfaces. A consumer who sees a brand in their Chase Offers, Marriott Bonvoy app, and Instacart recommendations has encountered it in three high-trust environments. The agent reads the interest graph. That interest graph was built by those three touchpoints.
This is why Financial Media Networks aren’t just the highest-signal data layer. They’re the highest-trust environment for interest formation. Chase and PayPal aren’t interruptions in a social feed — they’re inside a consumer’s financial life. That context carries a completely different level of trust.
Nurturing, not just activating. Most brands think in campaigns: Q4 push, product launch, seasonal spike. Interest formation requires a different mindset — continuous presence in high-context environments, building the association over time, not just capturing demand at its peak.
The shift: from transactional commerce media to relational commerce media. The question to ask alongside ROAS: am I present in the interest layer before the agent acts?
The Infrastructure That Makes It Work
The winning stack has one job: connect brand presence to interest formation to eventual purchase in a loop that gets smarter over time.
Before getting to the layers, here is the loop in plain terms. The decision layer does five things over and over:
Captures signals across trusted surfaces
Resolves those signals to a real person
Updates a live view of evolving interests and context
Determines what action should happen next — where, and why
Records the outcome so the next decision improves
Not campaigns. Not dashboards. A loop.
This is not a platform you buy. It’s a composable architecture you own- five distinct layers, each best-of-breed, connected by a single intelligence spine. Swap the components as the market evolves. Never surrender the architecture.
Layers 1 and 2 are the data foundation the infrastructure underneath everything. Layers 3, 4, and 5 are the Decision Engine where audience interests are mapped, decisions are made in real time, and learning compounds with every cycle. Together, they are the system that connects a consumer’s interest graph to a brand’s media activation.
Layer 1 — Data Cloud: The Foundation Your first-party data in Snowflake, BigQuery, or Databricks. In an environment you control, not a vendor’s black box. Every signal transaction, behavior, context, outcome flows back here. Everything else is built on top of it.
Layer 2 — Identity Graph: The Visibility Layer A real person stitched from transaction history, CRM records, app and web behavior, and partner data not a device proxy, not a cookie, not a probabilistic guess. Without this, the same customer on three surfaces looks like three different people. Identity resolution is what turns fragmented data points into a single, durable view of a real person.
Layer 3 — Audience Interest Graph: The Brand Asset This is what layers 1 and 2 build together and what most brands don’t know they’re not building. An identity-resolved, time-aware map of evolving affinities, consideration signals, and context across trusted surfaces. A consumer’s personal interest graph captures what they want. Your Audience Interest Graph captures who among your audience wants it — and how that’s shifting. This is not a segment. It’s a compounding asset. Brands that own it stop renting audiences. They start knowing them.
Layer 4 — Context Engine: The Intelligence Layer Layer 2 tells you who. The context engine tells you what they need right now. It reads the Audience Interest Graph and determines whether this is the moment to reinforce, suppress, recommend, wait, or escalate — and across which surface, message, and moment. That is the difference between blunt targeting and actual decisioning.
Layer 5 — Decision Traces: The Memory Every exposure leaves a signal. Most brands discard it after the campaign report. Decision traces are structured records linking signal → action → outcome → learning. Each cycle makes the next decision smarter. That’s the compounding loop and the moat no competitor can replicate by buying a better ad server.
The five layers are composable. The intelligence they produce is not. That’s the moat.
The Series in One Frame
Part 1: What retail media is and who the players are. Part 2: Who wins the infrastructure war and why most brands are paying an identity tax they don’t know exists. Part 3: What the winners do with owned infrastructure shape consideration before the agent acts, and learn from every exposure.
Whether you’re selling protein powder or kitchen cabinets, the brand that won wasn’t the highest bidder on the search keyword. It was the one that showed up relevantly, consistently during interest formation.
The transaction is not the battlefield anymore. It is the endpoint.
The real contest is upstream:
Who shaped consideration before the agent optimized purchase. Who built memory that survives automation. Who can prove they moved interest, not just captured demand.
If you are not building that layer, you are not building a moat. You are renting demand.
Build your audience. Own your audience. Understand their evolving interests so when the agent acts, your brand is already there.
An LLM helped with research, citations, and light proofreading without changing the content. All ideas, arguments, voice, and em dashes are mine. :)