Data Rich, Insight Poor
Marketers have never had access to more first-party, third-party, commerce, and media data. The problem is turning all of it into usable insight still requires too much technical complexity, too many handoffs, and too much time.
For more than a decade, marketers have been given essentially the same advice: get closer to your customers, take control of your first-party data, connect it to the media you buy, and use it to make better decisions. Many marketers heeded that advice. Today, brands have access to more potentially valuable customer intelligence than at any point in the history of marketing. They know what people buy, what they browse, which products they own, how frequently they transact, which messages they respond to, and increasingly how those behaviors connect across digital and physical channels.
At the same time, an entire ecosystem has emerged to make that data even more useful. Retail media networks provide access to commerce signals. Publishers possess rich behavioral and contextual data. Companies like Experian, Acxiom, and TransUnion maintain enormous identity, demographic, and consumer datasets. Amazon, Google, and Meta sit on extraordinary volumes of media and behavioral information. And data clean rooms make it possible for organizations to collaborate across sensitive datasets without simply passing customer-level information back and forth.
On paper, this should be the golden age of data-driven marketing. Except it doesn’t feel like one.
The uncomfortable reality is that most marketers still struggle to turn all of this data into anything resembling a repeatable competitive advantage. The problem is not that the data doesn’t exist. But turning it into intelligence, decisions, and action remains incredibly difficult.
Amazon Marketing Cloud, or AMC, offers a useful illustration. Amazon has long made AMC broadly available to Sponsored Ads and Amazon DSP advertisers, giving them access to one of the richest combinations of advertising and commerce data in the market. Advertisers can analyze customer journeys, understand how media exposures work together, measure new-to-brand behavior, create custom audiences, combine Amazon signals with their own first-party data, and answer questions that conventional advertising reporting was never designed to address.
And yet, according to multiple independent sources—including Skai, Stratably, Intentwise, and Tinuiti—advanced adoption remains at roughly 10 percent. This gap is revealing. AMC is not an obscure or experimental technology. It is backed by one of the largest advertising platforms in the world, contains extraordinarily valuable data, and can answer exactly the kinds of questions marketers say they want answered. If adoption of its advanced capabilities remains limited, the constraint is clearly not access to data alone.
The real challenge is everything that has to happen between gaining access to the technology and getting value from it. Uploading first-party data into a clean room can require legal review, privacy approval, identity strategy, schema mapping, cloud infrastructure, data engineering, permissions management, and ongoing operational support. Once the data arrives, someone still needs to understand the underlying datasets, determine which questions are worth asking, write and validate the queries, interpret the results, and translate those findings into something a media platform or marketing team can actually use.
This takes specialist expertise, engineering capability, and significant manual coordination. The result is a process that is often complex, slow, expensive, and difficult to scale. The irony is hard to miss: marketers have spent years accumulating more data, connecting more systems, and gaining access to more sophisticated analytical environments. Yet the distance between a business question and a usable answer remains stubbornly large
This contradiction sits at the center of modern marketing data. We spent years building increasingly sophisticated technology to make customer data more powerful, secure and interoperable.
But we never made it particularly easy to use. In fact, in many cases we did the opposite.
The decline of third-party cookies, growing privacy regulations, and legitimate concerns about how customer information moves between companies have pushed the industry toward architectures that are more controlled and more privacy-conscious. Those changes were necessary, but they also introduced new layers of infrastructure, governance and technical complexity.
The result is an industry that is simultaneously data rich and insight poor. Brands possess extraordinary amounts of information about their customers. Platforms possess enormous datasets of their own, data providers can enrich those signals further, and Privacy Enhancing Technologies such data clean rooms allow those datasets to be analyzed together under tightly controlled conditions.
The raw ingredients are all there. What has been missing is a simple, scalable way to put them to work.
We Built Data Collaboration for Data Engineers
Consider what happens when a marketer wants to answer what sounds like a relatively straightforward question: Which customers were exposed to our campaign, later purchased the product, had never purchased from us before, and are now most likely to respond to a different offer?
Though this business question can be expressed in a sentence, answering it requires considerably more. Someone has to identify which datasets contain the relevant signals. Someone has to understand how those datasets are structured, determine which identifiers can connect them, establish the appropriate permissions, construct the analysis, validate the output, and decide whether the resulting audience or insight can actually be activated. If the data sits across multiple organizations or environments, another layer of identity, governance, privacy, and infrastructure enters the equation.
None of these requirements is unreasonable. In many cases, they exist for very good reasons. Customer data absolutely should be protected, and access should be controlled. Likewise, analyses should be reproducible, and sensitive datasets should not move indiscriminately between companies.
The problem is that we have historically solved those requirements by putting technical complexity between the person asking the question and the data capable of answering it.
That has created an unusually vexing dynamic in marketing. Namely, the people closest to the business problem are often several steps removed from the systems capable of solving it. Think about how it works. A marketer asks an analyst, the analyst needs data from an engineering team, and engineering needs access from another team. During this process, someone discovers the necessary field is not available in the expected schema. As a result, the privacy team needs to approve a new use case. Someone in engineering then needs to write a query, test, revise, and rerun. Eventually an answer comes back—sometimes days or weeks after the original question was asked. Then the marketer asks the obvious follow-up question. And much of the process begins again.
This is the hidden tax on modern marketing data. It is not simply the cost of storing data or licensing technology. Rather, it is the accumulated cost of translation, coordination, specialized expertise, and time required to turn a business question into an insight the business can understand and act upon.
For years, the industry responded by adding more tools and more specialists. We built better clouds, cleaner identity frameworks, more powerful analytics environments, sophisticated clean rooms, and increasingly capable activation platforms. What we did not fundamentally change was the interface between those systems and the people trying to use them.
Then the Interface Changed
For most of the history of enterprise technology, humans have had to learn how machines want to work. If you wanted information from a database, you learned SQL—or found someone who knew it. If you wanted to connect two systems, someone needed to learn how to build point-to-point integrations or understand APIs. If you wanted to analyze a large dataset, you needed the right analytical tools, technical skills, and knowledge of how the underlying data was structured. In other words, the burden of translation sat largely with the human.
Generative AI is beginning to invert that relationship. Increasingly, a business user can describe what they are trying to accomplish in ordinary language—typed or spoken—while software translates that intent into the technical steps required to execute it. A question can become a query, and a query an analysis. An analysis can then trigger another question, invoke another system, retrieve another dataset, or initiate an action.
This distinction matters. The first wave of generative AI in marketing was largely about content: writing copy, creating images, summarizing documents, generating variants, and accelerating creative production. Those use cases are valuable, and most of us leverage them daily in our work. But they only hint at what happens when AI begins interacting directly with the underlying systems that run the business.
Agents change the equation because they can do more than simply generate an answer. They can increasingly determine which tools are needed, interact with those tools, execute a sequence of steps, evaluate the results, and decide what needs to happen next.
Applied to marketing data, this creates a radically different interface. Instead of asking a marketer to know which table contains an exposure event, how Amazon structures purchase data, which identity key connects a first-party customer record to a media signal, or what SQL is required to calculate a new-to-brand conversion rate, an agent can increasingly absorb that complexity on the marketer’s behalf. As a result, the marketer can start with the thing they actually know best: the business question itself:
Which customers saw the campaign but did not purchase?
Which audiences generated the highest incremental value?
What sequence of media exposures tends to precede conversion?
Which customers should we suppress from the next campaign?
What would happen if we shifted budget toward the audiences showing the strongest new-to-brand behavior?
Behind each question may still sit complicated datasets, permissions, queries, analytical logic, and activation systems. The complexity does not disappear. But business users no longer need to personally navigate every layer of it. This is a much bigger change than giving people an easier way to write SQL. It begins to separate the ability to use sophisticated data from the technical expertise traditionally required to operate the systems containing it.
And once that happens, the bottleneck starts to move. The scarce resource is no longer simply the ability to query the data. It becomes the ability to ask the right question in the first place.
Data Collaboration to Agentic Data Orchestration
This is where the concept of agentic data orchestration begins to matter. Today, most marketing data workflows are built around the systems themselves. A marketer needs to know where the data lives, which platform can access it, what tools are required to analyze it, who has permission to use it, and how the resulting insight gets pushed into another system for activation.
Agentic data orchestration begins from a different starting point: the desired outcome. Instead of requiring the user to navigate each piece of infrastructure independently, an intelligent orchestration layer can increasingly coordinate the work across them.
The basic model looks something like this:
Ask: a user starts with a business question or objective expressed in ordinary language
Discover: the system determines which datasets, platforms, tools, and capabilities may be relevant to answering it
Connect: it establishes the appropriate connections across those environments, subject to identity, access, privacy, and permission requirements.
Analyze: agents execute the queries, transformations, models, or analytical workflows necessary to answer the question.
Govern: policies, permissions, approvals, and auditability determine what the agent is allowed to access, analyze, and do.
Act: the resulting insight can be turned into an audience, recommendation, campaign change, measurement workflow, or another downstream action.
Importantly, this does not mean all of the underlying data suddenly needs to be pulled into one enormous centralized repository. In many cases, the opposite is true.
An agent can increasingly bring the question and analytical logic to the environment where the data already lives. Customer data can remain inside an enterprise data cloud. Commerce signals can remain inside Amazon Marketing Cloud. Publisher data can stay in the publisher’s native environment. Third-party attributes can be accessed for enrichment without necessarily being shipped back and forth as files. Sensitive information can remain protected behind the governance controls established by each party.
The agentic orchestration layer coordinates the work across those environments without requiring the business user to understand every technical step underneath it. This represents a monumental evolution in the way the industry has traditionally thought about data integration.
For years, the default assumption was that getting more value from data meant moving more of it: ingest it, normalize it, copy it, centralize it, and distribute it somewhere else for activation. Agentic orchestration opens another possibility. Instead of continually moving the data to the intelligence, we can increasingly move the intelligence to the data itself.
The Real Breakthrough Is Democratization
The most important implication of this shift is not that data engineers will be able to write queries faster. Instead, far more people may be able to work directly with sophisticated data in the first place. Historically, access to advanced marketing analytics has been constrained by technical literacy and the availability of engineering resources. The more complex the environment, the smaller the number of people who could realistically use it without help. A strategist might understand the customer problem perfectly but lack the technical skills required to interrogate the data. A media planner might know exactly which audience question matters but still need an analyst or engineer to translate that question into something a platform can execute.
Agentic orchestration begins to collapse that gap. A marketer does not need to become a data engineer, and a strategist does not need to learn SQL. Brand leaders don’t need to understand every table, schema, API, identity key, or permissioning model underneath the system. They just need to understand the business well enough to ask a useful question.
This is a fundamentally different model of access. The practical consequence is that sophisticated analysis can move closer to the people actually making decisions. Instead of insight being produced primarily through a centralized queue of specialists, it can increasingly become part of the normal workflow of marketers, planners, strategists, product teams, and business leaders.
Importantly, this does not eliminate the need for technical experts. If anything, their role becomes more important in different ways. Someone still needs to design the underlying architecture, establish policies, validate models, manage permissions, define trusted data products, and ensure the system behaves as intended.
What changes is where those experts spend their time. Instead of acting as the human interface between every business question and every dataset, they can increasingly focus on building the governed infrastructure that allows others to ask and answer those questions safely.
That is the real democratizing effect of agentic data orchestration. It does not make everyone a data engineer. It makes sophisticated data usable by people who should never have needed to become one.
Governance Becomes More Important, Not Less
Of course, giving more people easier access to sophisticated data introduces an obvious question: What happens when someone asks the system to do something they should not be allowed to do? This is where the conversation about agents can quickly become oversimplified. Removing technical barriers does not mean removing controls. In fact, as access becomes easier and agents become capable of taking more consequential actions, governance becomes more important, not less.
Today, far too much data governance is enforced through friction. Access to a dataset may require a ticket, and a new use case may trigger a privacy review. In many cases, someone needs to provision credentials, approve a query, or manually authorize the movement of data into another environment. These processes can be frustrating and slow, but they also serve an important and ulterior purpose: they create boundaries around who can use data, for what purpose, and under what conditions.
Agentic systems cannot simply eliminate those boundaries. But they can remove much of the friction humans face in navigating them by understanding and enforcing the rules directly.
Ironically, this means permissions, consent, data-use policies, contractual restrictions, approval requirements, and auditability increasingly need to become part of the orchestration layer itself. An agent needs to know not only what data exists, but whether a particular user is entitled to access it. It needs to understand not only how to create an audience, but whether that audience can legally and contractually be used for the intended purpose. Before an action is taken, it also needs to know whether human approval is required.
In other words, governance can no longer sit entirely outside the workflow as a separate checkpoint. Instead, it increasingly needs to travel with the data and the action. This creates the possibility of a different kind of operating model. Instead of relying on people to remember every rule and manually enforce every control, the rules themselves can increasingly become machine-readable and executable. Permissions can be checked automatically. Restricted attributes can be excluded. Certain actions can require approval. Every query and decision can be logged. Policies can be applied consistently across a growing number of data environments.
This does not make governance disappear. Instead, it makes governance part of the infrastructure itself. And this may be one of the most important prerequisites for democratizing data access safely and at scale.
The Data Stack Looks More Like a Network
For years, the dominant philosophy of enterprise data architecture was centralization. For those who are interested, we covered this issue at length during an excellent podcast with Kyle Csik titled, "Machines of Loving Grace? Kyle Csik on AI, Human Judgment, and the Future of Work."
The conventional wisdom was straightforward: bring the data together, normalize it, resolve the identities, and store it in a warehouse, lake, CDP, or data cloud. Then push the resulting audiences and insights back out to the systems where they can be used. There are many good reasons why this model became dominant. Centralization can create consistency, make governance easier in many cases, and give organizations a common foundation for analytics and activation.
But it also creates another problem: data does not naturally live in one place. Customer records may sit inside an enterprise data cloud, while advertising exposure data lives inside media platforms. Commerce data sits inside retailers and marketplaces, while publishers maintain their own behavioral and contextual signals. Identity providers, measurement companies, and third-party data partners each control and safeguard additional datasets of their own.
Trying to continuously copy all of this information into a single environment is expensive, technically difficult, and in many cases undesirable or impossible. Agentic orchestration suggests a different model.
Instead of thinking about the data stack primarily as a collection of systems feeding one central destination, we can begin to think about it as a network of governed environments that intelligence can operate across.
In this model, the data can remain where it makes the most sense for it to live. The customer record can stay in the enterprise data cloud. Commerce signals can remain with the retailer. Media exposure data can stay inside the media platform. Publisher data can remain with the publisher. Sensitive information can remain inside clean rooms or other controlled environments.
What moves across this network is increasingly not the raw data itself, but the question, the analytical logic, the permissioned request, and ultimately the resulting insight or action. This dynamic changes the architecture in an important way. Yesterday’s data stack was designed primarily to move data to applications.
The emerging model increasingly allows intelligence to move to the data. Once intelligence can move across a network of governed environments, the value of the data is no longer determined solely by how much of it an organization can centralize. Instead, it is determined by how effectively the organization can discover, access, analyze, and act across the data it already has permission to use.
The Missing Layer Was Always Usability
Looking back, the marketing industry has solved an extraordinary number of difficult data problems. We figured out how to collect enormous volumes of customer information. We built cloud infrastructure capable of storing and processing it at unprecedented scale. We developed identity frameworks to connect signals across systems and channels. We created clean rooms and other privacy-enhancing technologies that allow organizations to collaborate without indiscriminately exchanging customer-level information. And we built increasingly sophisticated platforms for measurement, analytics, audience creation, and activation.
What we never really solved was data usability. We built an ecosystem in which the data became extraordinarily powerful while the process required to use it became increasingly complex. The result was not surprising: the organizations with the most sophisticated technology did not necessarily generate the most sophisticated insights. Often, the advantage went to the organizations with the engineering resources, specialized talent, operating processes, and patience required to navigate all of that complexity.
Agentic data orchestration has the potential to change this equation. It does not eliminate the underlying infrastructure. The databases still exist, and identity systems still matter. Clean rooms still play an important role, permissions still need to be enforced, and privacy requirements certainly still apply. Moreover, data engineers, architects, analysts, and governance teams remain essential.
What changes is the interface itself. The business user can increasingly begin with intent rather than infrastructure. A question expressed in ordinary language can initiate a governed workflow that discovers the appropriate data, accesses the right systems, performs the analysis, respects the required policies, and ultimately produces an insight or action. This not only collapses the distance between a question and an answer, but also—and perhaps more importantly—changes who is capable of asking the question in the first place.
This is why agentic data orchestration matters. Its promise is not simply that existing analytics workflows become faster or cheaper. More importantly, it has the potential to unlock the intrinsic value of the extraordinary data infrastructure we spent the last decade building, making it usable by the people who understand the business problems it was supposed to solve.
For years, marketers were told that more data would create better marketing. We got the data, and now we need to make it usable.
Looking ahead, the next competitive advantage will not come from simply possessing more customer information than the competition. Instead, it will come from being able to turn that information into intelligence—and intelligence into action—faster than everyone else.
------------------------------------------------
Rio is an executive with 20+ years at the intersection of strategy consulting, AdTech, data, and media. He's a trusted advisor on customer experience, digital strategy, and marketing transformation. He's a partner at Credera, Omnicom's consulting arm. He's also a podcast host, writer, and public speaker focused on the future of advertising and AI-driven infrastructure.
Comments