The End of Audience

Why the atomic unit of targeting deserves a harder look.

My family goes through Kind Bars. When the box gets low, someone orders more. That someone could be any one of four people in the same household: a sixty-five-year-old female non-profit volunteer, a fifty-six-year-old investor, a twenty-nine-year-old teacher, a twenty-seven-year-old product manager. Four generations, four income profiles, four entirely different media habits. No targeting algorithm would place them in the same audience segment.

The signal is Kind Bars. The rate at which they disappear. The moment someone notices the box is low and picks up their phone.

I've been building and investing in advertising infrastructure since the late 1990s. I have spent the past year looking at that pantry and thinking the industry should consider questioning the atomic targeting unit in 2026.

Rishad Tobaccowala and Drew Ianni have been convening the Athena Project as a forum for today's leaders to push on exactly these kinds of questions. What I find unique about what they do, is something I do not encounter often: a serious willingness to question the underlying assumptions, not just the execution. This is refreshing, and the concept of pondering a question like this was the inspiration for this essay.

The essay has also prompted a topic at the Athena Project's Salon at Advertising Week New York, where Domenic Venuto of Horizon Media and I will share thoughts on this question in a fireside, moderated by Carrie Seifer. The question we are posing: The End of Audience? We are not showing up with a verdict. We are showing up in the right room of people, to get them thinking about it, and what it might mean.

Every industry is eventually humbled by what it refuses to question.

For advertising, that thing has been the Audience. Not the creative. Not the media plan. Not the measurement methodology. The Audience itself. The core assumption that the person, identified, segmented, and targeted, is the atomic unit of the entire enterprise. This assumption has survived every wave of technological change: the rise of television, the internet, the mobile phone, programmatic buying, the clean room. Cookies came and went. The Audience remained.

What is changing now is foundational.

AN APPROXIMATION THAT SERVED

For thirty years, the advertising industry built extraordinary infrastructure to find people. Demographic targeting found women twenty-five to fifty-four. Identity targeting found the same woman across her devices. Interest targeting predicted what she might want next. Purchase-based targeting used what she had already bought to find people who looked like her.

Each generation was more precise than the last. None of them questioned whether precision pointed at the right thing.

The demographic segment was always a proxy. The household ID was always a proxy. The cookie, the device graph, the RampID, the clean-room match: every atomic unit the industry named was a stand-in for the thing it was actually trying to cause. A purchase. A subscription. A behavior. The proxies improved continuously. The industry rarely asked whether proxies were the problem.

The answer to the post-cookie world has been the clean room: two datasets, one controlled environment, no raw data exchanged. Prove the sale happened. Protect the consumer. Satisfy the regulator.

Closed-loop measurement solves a real problem. If you need to prove a specific campaign moved specific product, in a specific channel, the closed loop delivers. But there is a limitation, and it is a leadership-level one: when you sit down to plan next week's budget (with a number of dollars and a blank sheet), the closed loop is silent. It cannot see outside its own walls. It cannot tell you which purchase event is most likely to happen next, or where the next intervention will work hardest. The best practitioners know that proof is not planning, and effective planning now needs speed.

WHERE THE SIGNAL NOW BEGINS

Before someone buys, they ask. That has always been true. What has changed is who answers.

A consumer today doesn't begin with a search bar. She opens an AI assistant. She asks which protein bar is worth buying, which streaming service is worth keeping, which financial advisor is worth calling. The AI answers. The answer shapes a belief. The belief shapes an intent. The intent becomes the purchase, or the competitor's purchase, or nothing at all.

Advertising has always worked downstream of belief. Look no further than the best Don Draper scenes from Mad Men, where the already held belief or desire became the jumping off point for the brand campaign. The question, the answer, the moment of conviction: those happened somewhere else, and the industry showed up to romanticize it (Don Draper), or action it (Jeff Bezos) at the moment that mattered. That upstream space is now owned by an AI answer engine. Most CMOs do not know what that engine says about their brand. They certainly do not know what it will say thousands of times today.

THE SIGNAL THAT IS ALIVE

A purchase event describes a household more precisely than any audience profile ever has.

A household that buys Kind Bars, Tide Pods, Celsius, and Zyn is not Audience #48291. It is a purchase graph. That graph knows more about what this household will do next than any combination of age, income, gender, or media behavior. The purchase history is the portrait. The identity profile was always a sketch.

The same logic holds in streaming. A household that subscribes to one service, cancels another after a season ends, and returns six months later is a subscription graph. The rate of subscription, cancellation, and return tells you precisely what caused that household to act and what will cause it to act again. The subscription is a digital SKU. The churn is a transaction.

Here is what the clean room model misses. Signal is not static. It does not live in a loyalty file waiting for a scheduled export. Real signal exists at the moment of the transaction. It is alive. It surfaces when it happens, and it tells you something precise and actionable that no audience profile could have told you in advance.

The industry built infrastructure to freeze signal into lists. What it needs is infrastructure to reason with signal while it is still moving.

THE MACHINE THAT DOES NOT NEED THE APPROXIMATION

Humans invented audiences because humans needed simplification. No analyst can reason over billions of individual transactions, accounting for substitution patterns, price elasticity, seasonality, household composition, purchase velocity, and competitive activity simultaneously. So the industry collapsed all of that complexity into demographic segments and audience IDs. It was a reasonable approximation given the tools available.

AI does not need the approximation.

An AI reasons directly over the transaction graph. It doesn't need a suburban-parent media persona. It needs the knowledge that specific products disappear from a specific household at predictable rates, and that a specific intervention at a specific moment will cause the next purchase. There is no audience in that equation. There is a predicted event and a causal signal.

Advertising built systems around: who should see this ad? The next generation of systems will be built around: which purchase event is most likely to happen next, and what causes it? That is a big change. We need everyone to be prepared for this, from the C-Suite on down.

QUESTIONS THE C-SUITE AND BOARD CANNOT DEFER

Whether you are a CMO, a board director, or a CEO watching a marketing budget allocation that no longer quite makes sense, the transition from audience to signal is a leadership problem before it's a technology problem.

Start with measurement. Most companies are paying for evidence of the past. The closed-loop study confirms a campaign worked. That's useful. It's not nothing. But the planning question, which purchase event will happen next and where should the next dollar intervene, is going unanswered. Boards have accepted proof as a substitute for planning, and in a world where signal is moving in real time, that's a trade-off that keeps getting more expensive.

The question I hear least often in boardrooms is: what does AI say about our brand? Before a consumer acts, she asks. If your organization has no view into how AI answer engines respond to questions in your category, you have no view into the belief formation that precedes purchase. That's not a future gap. It's today.

Signal ownership is a governance question most companies haven't answered. Purchase signals sit in operations. Churn signals sit in finance. Subscription signals sit in a vendor's dashboard. The leader who can't consolidate those into something a planning team can reason over has already fallen behind. Not because of technology. Because of organizational structure.

The last question is hardest: is your team actually built for this? The skills that built audience-based systems are the skills for a different problem. I've watched talented people struggle to retool because the problem changed faster than the organization acknowledged it. Signal-based planning lives closer to causal inference than creative strategy. The companies closing that gap are not waiting for the industry to define the job description.

THE QUESTION THAT EXTENDS BEYOND ADVERTISING

The audience was advertising's core assumption. Every industry has one. And AI is putting pressure on all of them.

In healthcare, the physician was the unit. Now a patient arrives at the appointment having already consulted an AI that read every clinical study and patient forum. The belief formation happened before the appointment.

In financial services, the relationship was everything. When the first answer to "should I rebalance my portfolio" comes from an AI at two in the morning, the consideration set forms before the advisor enters the conversation.

In insurance, the underwriter was the unit. Risk assessment moved through professional judgment, people who had spent careers learning to read a claim, a property, a life history. Distribution moved through the broker who knew the client. What happens when a business owner asks an AI to explain which coverage she needs, compare carriers, and identify exclusions before a broker enters the conversation? And what happens when the system on the receiving end can assess that same risk and bind a policy directly? The underwriter and broker are still in many transactions. But the belief formation, and in some cases the decision, has moved upstream.

These are not analogies to advertising. They are the same type of transition, though. In every case, the unit the industry built its infrastructure around was a proxy for something more fundamental: a decision, a belief, an intent. And in every case, the upstream moment, where the belief forms before the decision is made, now belongs, in no small way, to an AI answer engine.

The leader's question is not whether their industry is like advertising. It is whether their industry has an analog to advertising's core concept of audience. And whether that analog is about to become a signal. Most do. Most will.

THE QUESTION THAT REMAINS OPEN

Is this the end of Audience?

Not entirely. Not overnight. The infrastructure is too embedded, the budgets too large, the habits too ingrained. Audiences will not vanish from the planning conversation. But the atomic unit of advertising is shifting. The proxy is becoming unnecessary. The signal is alive and the systems reasoning over it are being built now, not in some future cycle.

I spent twenty-five years inside the systems this piece argues against. Not as a critic. As a builder: founding companies, ringing the NYSE bell, investing and acquiring companies and teams. I was fortunate to work alongside people who understood the business in ways I did not, and to be present as each wave changed the terms. I watched the cookie replace the panel. I watched mobile identity replace the cookie. I watched retail media close the loop and produce more proof than the industry had ever had. Each era produced better tools pointed at the same unit. None of us, at the time, thought to question the unit.

Perhaps it is time.