Matthew Papa and I were supposed to record a conversation in Cannes earlier this year. Unfortunately, I lost my voice, so we ended up recording in August, sitting in an air-conditioned room in Manhattan, considerably less glamorous than the original plan.
The premise came from my three-year-old son, who has entered the phase of asking “why” about absolutely everything. Every answer is met with another question, until you eventually have to admit you don’t know or reconsider something you’ve always taken for granted.
I’ve found myself thinking about how useful that exercise would be for our industry.
Why do we do things a certain way? Usually because that’s how they’ve always been done. Why were they done that way in the first place? Often because of a technical limitation or market condition that may no longer exist.
Matt and I covered quite a bit of ground, ultimately breaking our conversation into five clips. While the topics were different, I think they all point toward a similar question: What is actually scarce in adtech today, and what are we continuing to value simply because it used to be difficult to build?
When Access Was the Advantage
For much of my career in adtech, the industry rewarded companies for building connections. More supply integrations, more data partners, more measurement optionality, more ways for advertisers to execute campaigns across an increasingly fragmented ecosystem.
The prevailing approach was to give agencies as much control as possible. Connect them to everything, provide the tools, and let them choose their own adventure. Every new partnership or integration added another option to the menu.
For a long time, that made sense. Access was difficult to build and maintain, and having more of it created a competitive advantage.
Today, most major platforms can access much of the same inventory, work with many of the same data providers, and support many of the same integrations. We’ve spent fifteen years building an extraordinarily sophisticated ecosystem, but the infrastructure that once differentiated one company from another has become increasingly standardized.
The question is now what a platform can actually do with that access. What does it understand about the consumer, the content, or the opportunity that someone else doesn’t? Can it make a better decision, and can it demonstrate that decision contributed to a better outcome?
I’ve spent a considerable portion of my career building identity infrastructure, and I’ve come to believe that the utility of that infrastructure matters considerably more than the infrastructure itself.
Even the integrations are changing. Standards like MCP are making existing systems and services easier to connect to, including for a startup with just a handful of employees.
The competitive advantage is shifting from what you can connect to, toward what you can uniquely contribute once everything is connected.
The Cost of Building Is Collapsing
One of the things I’ve always loved about adtech is how accessible it is to entrepreneurs.
There are industries where participation requires enormous amounts of capital before you can even compete. You can’t exactly decide tomorrow that you’re going to launch an airline. The capital requirements alone eliminate most would-be competitors.
Adtech is different. A handful of people with a good idea, a laptop, and an understanding of a specific problem can build something that competes with companies many times their size.
I’ve experienced that firsthand, more than once.
When we were building Screen6 in 2016, a tremendous amount of engineering went into developing the machine learning models behind our cross-device identity algorithm, automating their recalibration, building client integrations, and managing the data pipelines required to ingest and process signals, construct identity graphs, and deliver the results back to clients. All of that required substantial engineering resources.
Building that same infrastructure today would be an astronomically different undertaking. Much of the surrounding technology could be accelerated through existing services, readily available tools, and AI-assisted development. The underlying identity problem would still be difficult, but the resources required to build and operate the technology would look completely different.
A three-person company can accomplish things today that would have required dozens of employees just a few years ago.
Matt predicted that we’ll see a proliferation of very small companies generating meaningful revenue over the next several years. I suspect he’s right.
But the same conditions that make it easier for a three-person company to compete also raise questions about the economics of a fifty-person company.
As the cost of building technology declines, maintaining large teams around relatively narrow capabilities becomes harder to justify. I expect we’ll see some interesting businesses emerge from this shift, and established ones forced to reconsider how they’re structured.
When Headcount Stops Measuring Work
AI is also changing the economics of how work gets done.
Historically, infrastructure costs in adtech have largely been treated as operating expenses. Servers, data warehouses, log processing, QPS capacity, and everything else required to operate at scale. These costs sit behind whatever product or service you’re selling.
AI introduces another measurable unit of consumption. As software performs more work that previously required people, compute becomes an increasingly visible input. How many tokens did a task consume? How much processing was required? And how should that factor into what a customer pays?
Consider the agency business.
For decades, much of the model has been structured around people. A client needs something accomplished, an agency determines the hours and headcount required, and pricing follows.
But when work that previously required an entire team can be accomplished through software, models, and a few employees, headcount becomes a less useful measure of what was delivered.
There’s already considerable discussion around outcomes-based pricing. I think we’ll also see experimentation with pricing tied to the computational resources required to deliver a service.
Matt mentioned that OpenAP had already been approached by a vendor exploring compute-based pricing for data usage.
Tokens aren’t a perfect measure of value. Two models can consume vastly different amounts of compute to accomplish the same task, and output quality matters far more than processing volume.
Still, when the cost structure of delivering a service changes this dramatically, pricing models will eventually follow.
And knowing this industry, I suspect there will be no shortage of companies looking for arbitrage opportunities along the way.
The Cost of Being in the Middle
Matt estimated that roughly 60 to 70 percent of companies in adtech sell primarily to other companies in adtech. The exact percentage is debatable, but anyone who’s worked in this industry knows what he’s talking about.
Think about how many companies sit between an advertiser deciding to spend a dollar and a consumer ultimately seeing an advertisement.
An entire ecosystem exists to sell data, identity, measurement, verification, optimization, and infrastructure to one another.
And for good reason. The open internet became as sophisticated as it is because thousands of companies solved problems that needed solving. Many were extraordinarily difficult, and the businesses that solved them created real value.
But how much of that complexity is still necessary?
If a capability that once required years of engineering and substantial capital investment can now be accessed and experimented with through an LLM and an MCP connection, the economics of an entire category can change remarkably quickly.
Many intermediaries perform essential functions, and some possess proprietary assets that are extremely difficult to replicate. But simply existing as another step in an already complicated transaction becomes harder to defend.
Technology businesses should ultimately be judged by the value they create, rather than the complexity of what they’ve built. As more of that complexity becomes accessible to everyone, that distinction becomes harder to ignore.
We Mistook Impressions for Attention
This was probably my favorite part of the conversation, particularly as more advertising investment moves into CTV.
In most markets, scarcity is fundamental to pricing. Whether we’re talking about oil, real estate, or virtually any other commodity, supply and demand help determine what something is worth.
Digital advertising developed somewhat differently.
The internet created what appeared to be an almost unlimited supply of advertising opportunities. More pages, more content, more ad slots, and more impressions available for purchase.
We built increasingly sophisticated mechanisms to price and transact against that supply. But somewhere along the way, I think we conflated the abundance of advertising impressions with the availability of human attention.
There have always been a finite number of people with a finite amount of time to consume content, and an even smaller number of moments when someone is genuinely paying attention.
That distinction becomes particularly important in environments like CTV.
There are a finite number of households, a finite amount of premium programming, and a finite number of hours people spend consuming it. Publishers can adjust advertising loads, but there are practical limits before the consumer experience begins to suffer.
And even within those advertising opportunities, delivery doesn’t necessarily mean attention.
An advertisement can be served, rendered, and completed without someone meaningfully engaging with it. Yet we’ve spent decades optimizing around whether something was delivered rather than whether it had a genuine opportunity to influence someone.
As we improve our ability to measure attention, understand content, and connect exposure to outcomes, I think we’ll see a more meaningful distinction between impressions that are technically available and those that are genuinely valuable.
Scarcity never disappeared from advertising. We just became very good at manufacturing supply without necessarily measuring the attention behind it.
The New Basis of Competition
Looking back across these five discussions, I keep coming back to the same conclusion.
Access to inventory is increasingly commoditized. Infrastructure that once required years to build can now be accessed through existing services. Integrations that previously demanded dedicated engineering resources are becoming far simpler through standards like MCP. And AI is accelerating the pace at which new businesses can build and compete.
As those barriers fall, the value of what cannot easily be replicated becomes more important.
Direct relationships with advertisers and publishers. Unique data and signals. An understanding of consumers, content, and behavior that can’t simply be purchased from another vendor. And the ability to demonstrate that those advantages produce measurably better outcomes.
For a long time, technical complexity itself functioned as a competitive advantage. If something was sufficiently difficult or expensive to build, relatively few companies could compete. That barrier justified entire businesses.
Those barriers are disappearing. And as they do, companies whose differentiation depended primarily on the difficulty of building their technology will have to answer a question they may not have faced in years: What value are we creating that someone else cannot now deliver at a fraction of the cost?
I think that’s a healthy development for the industry. It will challenge established businesses, create opportunities for new entrants, and force us to be more deliberate about what we’re actually building.
Which brings me back to my three-year-old.
He hasn’t learned to accept “because that’s how we’ve always done it” as an answer. I’m starting to think that’s a pretty useful way to approach an industry that’s spent fifteen years building things we may no longer need to build.