Paid and owned marketing reach audiences and customers on different surfaces, under different economics and controls. This map shows what has to connect underneath them.
Version 2 · Reviewed August 2026 · A point-of-view model, not the universal stack
A shared operating core is what lets separate delivery systems produce one consistent experience for the customer. When a connection in that core fails, the systems downstream act on incomplete or conflicting information.
Trace an event follows one real customer event across the layers. Explore the architecture opens the full map.
A scenario is already running; choose another below.
Each capability here may be supplied by one platform, by several systems together, or by a manual process.
The lit capabilities below are the ones this event has to cross. The order, and the point where it breaks, are in the panel.
Where it crosses a boundary
What fails first
What happened. Someone who has been getting your ads buys, online or in a store.
Which shared capabilities carry it:
Where it crosses a boundary. Commerce owns the transaction and media owns the ad spend, and nothing suppresses the ad unless the purchase signal crosses from the commerce team's systems into the media team's platforms fast enough to matter.
What fails first. The suppression is what fails first. If the purchase signal is slow, or the identity match misses, the customer who just bought keeps getting served the ad to buy it for weeks, so the spend goes toward irritating the person you just won.
What happened. A customer calls in angry, or opens a furious chat, about an order that went wrong.
Which shared capabilities carry it:
Where it crosses a boundary. Service owns the complaint and marketing owns the promotion, and marketing only knows to stand down if the service signal crosses into the profile and the decision layer treats it as a suppression rule rather than a data point that just sits there.
What fails first. The break is the handoff between service and marketing. Service already knows the customer is furious, and marketing sends the upsell anyway, because the complaint never became an input the decision layer could see, so one of the strongest risk signals you have stays shut inside a ticketing system.
What happened. A customer's AI assistant, not the customer, arrives to research and compare on their behalf.
Which shared capabilities carry it:
Where it crosses a boundary. This decision leaves the organization entirely. A machine you do not own interprets you for a customer on the other side of it, reading what you publish alongside everything else it can reach. Preparation improves your side of what it finds. The rest sits outside your control.
What fails first. Your product knowledge fails first. The assistant answers from whatever it can read, and what it cannot read from you it takes from somewhere else. The surfaces now report some of the answers you appear in. The ones that leave you out produce nothing to count. Share of Model, defined in Modern Measurement, is one way to measure that visibility.
What happened. A retail media network sends you an invoice and a report claiming its ads drove an incremental sale.
Which shared capabilities carry it:
Where it crosses a boundary. The retailer owns the measurement of its own media, which means it is effectively checking its own results. Testing the claim independently is on you, and when the test needs their data, it crosses into a collaboration you both have to agree to and pay for.
What fails first. Independent measurement fails first. Without a causal test of your own, you are left funding whatever the network says worked. What counts as incremental, channel by channel, is not a question the seller can answer about itself. Modern Measurement addresses that.
The surfaces where the customer actually meets the brand. Paid tends to sit toward the anonymous end, owned toward the known end, but the same core feeds all of them. Everything below this layer exists so these behave like one brand instead of a dozen disconnected ones.
Identity tends to be thinner on the paid side and richer on the owned side, but the pattern bends both ways. Paid surfaces include logged-in environments; owned sites get anonymous traffic; a known customer goes unrecognized on a new device. The core's job is moving what is known across surfaces, in both directions, without breaking consent.
What it does. Buys audiences at scale across the open web through DSPs and ad platforms: display, contextual placements, automated bidding against the segments the core produces.
The decision it serves. Where the next dollar of reach is best spent, down to which audiences are worth paying a premium to sit in front of.
What breaks without it. Without it, you are buying reach against someone else's definition of the audience, and because nothing coordinates the separate buys, one person gets hit over and over while you are left guessing what the spend actually bought.
What it does. Captures declared demand at the moment someone types it, across keyword campaigns and the automated formats that blend search, shopping, and display inventory.
The decision it serves. Whether a given search dollar is winning new demand or just paying to hold onto demand you had already earned.
What breaks without it. Search keeps winning clicks, but with no line back to the wider stack you cannot tell, from search data alone, which demand you created and which you would have captured anyway, so the channel ends up tuned to numbers the other teams rarely look at.
What it does. Runs the feed side of paid: platform ads, creator partnerships, and the UGC that platforms now price and rank like ad inventory.
The decision it serves. Which creators and communities are worth putting budget behind, and how that spend compares to the value of media you own outright.
What breaks without it. Creator and social spend ends up measured only by the platform that sold it, and a feed impression is hard to connect back to a profile you could act on later, so it sits outside the same accounting as everything else you run.
What it does. Places sight, sound, and motion across connected TV, online video, digital out-of-home, and streaming audio, bought against the same audiences as display.
The decision it serves. Whether the premium you pay for CTV and streaming actually buys more addressable reach than the linear budget it replaced.
What breaks without it. The platforms each count reach and outcomes their own way, the methods do not line up across them or with the rest of the plan, and the premium you are paying for CTV stays unproven.
What it does. Buys placement inside retailers' properties and audiences: sponsored products, on-site display, and off-site extensions built on the retailer's purchase data.
The decision it serves. Which retail networks earn their budget once the test is incremental sales measured independently of the network's own reporting.
What breaks without it. Each network reports its own success in its own terms, the results rarely line up side by side, and retail media slides toward a cost of doing business with the retailer that you cannot really weigh against the rest of the plan.
What it does. Covers the surfaces where assistants and answer engines research, compare, and recommend on the customer's behalf, drawing on your catalog, your content, and third parties you do not control.
The decision it serves. Whether the brand is legible enough to the machines to make their shortlist in the growing number of cases where the customer never reaches your site at all.
What breaks without it. When the brand is not legible to it, the assistant researches and recommends without ever surfacing you. The surfaces are starting to report where you appear. Where you do not appear is harder to see, because an answer that leaves you out produces no click to count. Share of Model is one defined measure of that visibility, and it lives in Modern Measurement.
Agent mediation. Increasingly, a customer's assistant sits between them and the brand: summarizing search, comparing listings, reading your product information, filtering your messages, acting through your service and commerce interfaces. It works as a layer of interpretation laid across the surfaces you already have, reading and acting on them for the customer. What the assistant can read, trust, and act on is governed by what the context layer exposes to it. Where AI discovery appears as its own surface, the answer it gives can combine paid placements, earned citations, and your own catalog in a single response, so it belongs cleanly to neither the paid nor the owned side.
What it does. Delivers owned, addressed communication, sequenced by the decisioning and orchestration layer against consent and frequency rules.
The decision it serves. How often this brand gets to land in a known customer's inbox before the welcome wears out.
What breaks without it. Every team with a send button ends up competing for the same inbox, the frequency caps sit in separate tools that contradict each other, and the customer feels all of it at once as one brand that will not let up.
What it does. Runs the properties you own outright, where content, offers, and experience change per visitor and every interaction writes a signal back to the foundation.
The decision it serves. How differently the site behaves for someone it recognizes versus someone showing up cold for the first time.
What breaks without it. A returning high-value customer gets treated like a first-time stranger, the personalization slowly turns into a pile of rules no one owns, and the behavioral signal your own site generates struggles to reach the profile where it would be useful.
What it does. Handles the conversations customers start: account self-service, the contact center and live chat, and the conversational AI now answering first.
The decision it serves. How much of what a customer says to service should follow them into what marketing does next.
What breaks without it. The service team can see a customer is at the end of their patience while marketing, with no line into that signal, sends the upsell anyway, because one of the strongest signals of intent and risk you have stays walled off inside a ticketing system the wider stack cannot see.
What it does. Captures the transaction wherever it happens: online, in the store, and through the in-store digital that connects the two.
The decision it serves. How online and offline purchases resolve to one customer, so value and frequency read as one relationship.
What breaks without it. The stores and the site keep separate books on the same person, so their lifetime value shows up split across two records, and the models downstream inherit that split as if they were two different customers.
What it does. Runs the loyalty value exchange itself, the rewards, tiers, and member experience that give customers a reason to sign in on every visit.
The decision it serves. What the program should be willing to pay, in rewards, to get a customer to sign in and keep doing the things worth encouraging.
What breaks without it. One of your strongest reasons for customers to identify themselves goes unused, the known side of the base slowly thins, and personalization falls back on guessing who someone is instead of being told.
Cross-cutting. Loyalty is a customer surface here and also supplies identity, permission, value, and program signals used across the shared core. The map treats it as cross-cutting instead of drawing the same capability twice.
The layer that decides and then delivers. It chooses who is eligible, what the next best action is, what gets suppressed, and which program wins when two want the same person, then it sequences the result across every surface and hands it to the paid arm or the owned arm to execute. The part that matters here is how that choice gets made, and who is allowed to change it.
What it does. Makes the actual decision about who qualifies, what the best next action is for this person, what to hold back, and which program wins when two collide, then runs the policy checks and marks where a human still has to approve.
The decision it serves. How much the system is allowed to decide by itself before a person has to step in.
What breaks without it. With nothing in one place deciding what stands down, every channel optimizes for itself, and the customer gets the email, the push, and the retargeting ad for the same offer inside the same hour, with no one able to point to where that was decided.
What it does. Sequences the decision across surfaces and time: journeys, triggers, and conflict resolution when two programs want the same person at once.
The decision it serves. What fires next for this person, on which surface, and what stands down because something more important is in flight.
What breaks without it. The journeys exist as diagrams that were never actually wired up, so all the decisioning upstream stops short of the customer, because nothing is turning a decision into an ordered sequence of things that happen.
What it does. Moves audiences and suppression lists into the buying platforms and brings performance data back.
The decision it serves. Which segments get spent against, at what bid, with which creative.
What breaks without it. Audiences get rebuilt by hand inside each platform and start drifting the moment they are made, and suppression is the first thing to go, so the customer who just bought keeps seeing the ad to buy it for weeks.
What it does. Executes the owned side through campaign automation, journey delivery, and the channel plumbing behind messaging, site, and app experiences.
The decision it serves. How a planned journey becomes actual sends and experiences without hand-built lists.
What breaks without it. Campaigns slide back to batch-and-blast, the carefully planned journey stops short of ever becoming real sends, and the owned side ends up running on lists somebody rebuilds by hand every week.
The layer that tells the rest of the system what things mean. A machine can move data without knowing that "active customer" has one definition, that this claim is legally cleared, that this KPI is measured this way, or that the brand does not say certain things. As more of the stack runs unattended, this layer is what keeps agents inside the lines.
What it does. Holds the business meaning, the working definitions of what a customer, a segment, an active account, and each KPI actually mean, set once so people and machines read them the same way.
The decision it serves. Whether two teams arguing about a number are arguing about the business or about whose definition is right.
What breaks without it. Each function turns up to the meeting with its own numbers, and the meeting becomes an argument about whose data is right when it should have been a decision about what to do.
What it does. Encodes what the brand may say, what is legally cleared, which offers apply to whom, and where a human has to sign off, in a form the automated parts of the system can actually check.
The decision it serves. What an agent or an automated journey is allowed to say and offer without asking first.
What breaks without it. When the rules only exist in a slide and in what a few experienced people happen to remember, an automated system has nothing concrete to check against, so it either stays too cautious to be useful or acts on its own read of the brand and gets it wrong.
What it does. Supplies the structured facts about what you sell: catalog, specifications, availability, and the relationships between products, so both people and machines can answer accurately.
The decision it serves. Whether an assistant, a site, or an agent can give a correct answer about your products without a human in the loop.
What breaks without it. The machine gives a wrong answer with full confidence, the surfaces downstream carry that error forward, and the assistant that is supposed to recommend you to the customer is working from bad information.
What it does. Provides the approved content, creative, and assets the surfaces draw on, with provenance and trusted-source rules attached so the system knows what is safe to use and where it came from.
The decision it serves. Which content and sources a generative or automated system is cleared to use and trust without a human sign-off each time.
What breaks without it. Automated and generative surfaces end up drawing on whatever material is reachable, with no record of where it came from, so there is little way to tell afterward whether an output came from approved assets or from something the brand never cleared.
Where the system knows who someone is and what is true about them. Identity resolution stitches devices and accounts into people; the profile store holds who they are and what they are allowed; partner data and privacy-safe collaboration extend that safely where it is worth the cost; and analytics turns behavior into the numbers decisions run on. Shared definitions live one layer up, in context, so this layer measures rather than defines.
What it does. Stitches cookies, devices, emails, and accounts into one person, and maintains the graph as signals decay and reappear.
The decision it serves. Whether the person on the site, the person in the store, and the person in the email list get treated as one customer.
What breaks without it. Without it, the systems downstream double-count the same person, so frequency caps miss, personalization contradicts itself, and measurement splits one person's journey across three.
What it does. Holds the working customer profile: attributes, consent, segments, loyalty status, and the audience definitions every delivery arm pulls from.
The decision it serves. Which customers belong in which audiences, settled once against one shared profile that every tool reads from.
What breaks without it. Audience definitions fork channel by channel, so "lapsed customer" comes to mean something slightly different in each tool, and afterward it is hard to say which version actually fired.
What it does. Matches your first-party data against partners, publishers, and retailers using privacy-safe methods (clean rooms are one) where analysis happens without either side exposing raw records.
The decision it serves. Whether a given partnership is worth the legal and engineering cost it takes to stand up.
What breaks without it. Partner measurement comes down to trusting the partner's own screenshots, retail media is left to vouch for its own results, CTV stays unproven, and most second-party opportunities stall out in procurement before they begin.
What it does. Runs attribution, mix models, experiments, and the machine learning that scores propensity and feeds decisioning, all reading from the same foundation and the same shared definitions.
The decision it serves. What worked, what to spend next, and which model outputs are trustworthy enough to act on automatically.
What breaks without it. Spend gets allocated on whoever argues hardest in the room, and because nothing tells you which model outputs are safe to act on without a person checking, the org either sends everything for review or trusts all of it blindly.
The governed base the layers above read from. A warehouse or lakehouse holds the analytical and customer data; event streams carry what has to move in real time; operational and profile stores serve the decisions that cannot wait for a batch. The point is not one physical place. It is governed, shared access, so every layer reads from a common source rather than keeping its own copy.
What it does. Stores the customer, campaign, catalog, and outcome data the analytical and batch layers read, on a cloud platform the organization already runs.
The decision it serves. What "one customer, one truth" physically sits on for the data that can be read in batch.
What breaks without it. Copies pile up until every tool is holding its own slightly stale extract, the sync jobs moving data between them quietly turn into the thing you actually maintain, and reconciling which version is right becomes a standing job.
What it does. Carries the signals that have to move now: a purchase, a site event, a service interaction, delivered to the decisions that act on them in the moment.
The decision it serves. Which decisions cannot wait until tonight's batch and therefore need their data live.
What breaks without it. The decisions that need to happen in the moment end up running on yesterday's data, so the suppression that should fire the second someone buys instead fires tomorrow, long after it would have mattered.
What it does. Serves the low-latency reads that decisioning depends on: the profile lookup, the feature a model needs, the API a surface calls mid-interaction.
The decision it serves. What a live surface can know about a customer fast enough to act on it inside a single visit.
What breaks without it. The decision layer can have exactly the right logic and still no way to run it quickly enough to matter, so personalization falls back to static rules because the real-time read is too slow to use.
The layer underneath everything, spanning both sides. It reads as overhead until an agent acts without asking. Then it is the only record of what happened and who allowed it.
Consent. What each person permitted, enforced everywhere a signal moves.
Data governance. Who can use which data for what, written down and machine-enforceable.
AI governance. What models and agents may decide alone, and what needs a human.
Security. Who gets in, what they can touch, what gets logged.
Policy. Retention, purpose limits, and regional variation, kept current.
Measurement standards. One set of definitions for what counts, so success is not negotiable per channel.
Observability. Whether anyone can see what the models and agents actually did: data quality, lineage, decision logs, evaluation, and what to do when one goes wrong.
This is a point of view, not the universal 2026 stack. It is drawn for large consumer-facing enterprises; a B2B, publisher, retailer, marketplace, or service business would keep the shared-core idea and redraw several of the surfaces and cells. Product names on the map (ad formats, protocols, platforms) are dated examples that will age first, well before the layers underneath them do. The claims are checked against primary and official sources and reviewed on a schedule; where a source is a vendor, an analyst, or a forecast, it is treated as a lead to verify before anything relies on it. Reviewed August 2026.
I have spent years walking marketing, data, and technology leaders through a map like this one. It was a working tool for surfacing the real issues and aligning on solutions, the map that organized the thinking and made the needs and the opportunities plain. This is the version I reach for first, before the vendor names and the org chart, because the shape holds across most of them. If a trace made you picture your own stack and the place it thins out, that is the map doing its job.
The intelligence layer holds the customer profile this architecture runs on. To watch those signals operate (velocity, permission, and what actually fires a message), open the Customer Signal Console.
Open the ConsoleYou just walked one trace through the map. The worksheet runs the same walk on an event of your own.
Take it into a room