Third in a series. Prior articles: “The Artist Already Knew” and “The AI Data Layer.“

Models respond. Agents act.

That is a small sentence with consequences, and it is the shift this article is about. The AI landscape is no longer only chat interfaces and question-and-answer flows. Gemini executes checkouts inside search. ChatGPT’s Operator capability opens browsers and fills carts. Anthropic’s Model Context Protocol lets Claude read and write to external systems. Google’s Universal Commerce Protocol standardizes how any agent can transact with any merchant. Salesforce reports that twenty percent of the 2025 holiday retail season was already influenced by AI agents. The era of the agent is not coming. It is here, it is growing quickly, and it changes what the infrastructure underneath it has to do.

Across this series, I have argued that context is the defining variable in applied AI, and that the contextual infrastructure feeding modern AI systems is a genuine next-generation data layer — the AI Data Layer — continuing a trajectory that runs from the early Analytics Data Layer through the Universal Data Layer and the Customer Data Layer that the industry now calls the CDP. In each of those transitions, Tealium has been the company building the category-defining layer. The transition to agents is the one where the stakes of that infrastructure get materially higher, because the consequences of wrong context are no longer a strange output but a strange action — a booking made, a refund issued, a preference updated, a negotiation completed.

Four shifts define what the contextual infrastructure has to do now. They build on each other, and together they describe what it means for context to keep up with an agent’s work.

The first shift is that context becomes continuous. A model answering a question receives a context window, produces an output, and the next call starts roughly fresh. An agent does not work that way. An agent operates across a session — sometimes across minutes, sometimes across days, sometimes across a long-running process that coordinates many smaller decisions — and the context it holds at step seventeen is partly a function of what it decided at steps one through sixteen. The CDP era optimized for the correctness of a profile at the moment it was read. The agent era asks for something harder: the profile has to remain coherent across the full arc of an agent’s work, update as the world changes underneath it, and deliver the right slice of that evolving picture to the agent at each step. Continuous context is not a faster version of point-in-time context. It is a different engineering problem.

The second shift is that agents generate signal, and that signal has to come home. When an agent books a flight, submits a return, updates a preference, or negotiates with another agent on a customer’s behalf, that action is itself a customer event. It has to flow back into the profile, because the next decision — whether by the same agent or a different one — depends on knowing that it happened. Agent-generated signal also carries different provenance than customer-generated signal, and the infrastructure has to capture the distinction. An action a customer took is not architecturally identical to an action an agent took on a customer’s behalf, even when the outcome is the same. Done well, this creates a virtuous loop: every agent action enriches the context that makes the next agent action better. Done poorly, it creates compounding noise. The architecture decides which of those you get.

The third shift is that permission itself becomes context that has to move. When a customer asks an agent to “book me a flight,” what they authorized is not a fixed fact stored once at the start of the session. It is a condition that applies to some actions and not others, for some data and not others, for some duration and not others. As the agent works, the permissions that govern its next action have to travel with it — available at every step, evaluable in real time, revocable the moment the customer’s intent changes. Consent in the agent era is not a flag on a record. It is a living signal that moves alongside the customer and the agent, and the AI Data Layer is what keeps it moving. Organizations that treat permission as static will find their agents making authorized decisions in one moment and unauthorized ones in the next. Organizations that treat permission as context will find that trust becomes the most durable asset their brand owns.

The fourth shift is that the architectural pattern from the last article — customer, contextual layer, AI, action — has to close into a loop. In the responsive-AI world, that pattern is roughly linear: context flows into the model, an answer flows out, and the cycle completes. In the agent world, the action the AI takes produces new state in the real world, which has to be observed, ingested, governed, and made available to the next decision. The AI Data Layer stops being a pipeline and becomes a closed loop. The pieces are mostly already there in any organization that built a CDP seriously. What changes is what the layer is for, how tightly the loop has to close, and how continuously the whole system has to keep the context current.

This is the work Tealium has been building toward across three generations of data layer. We built the Universal Data Layer and made it the standard by which a vendor-neutral description of a digital event was carried across the stack. We built the Customer Data Layer and made real-time, identity-resolved, consented customer profiles the operating foundation for a decade of marketing and experience technology. The AI Data Layer we are building now brings continuous context, bidirectional agent signal, living permission, and closed-loop architecture into one layer — and we have already stood up the connective tissue that lets agent ecosystems reach it, including a managed Model Context Protocol server that gives agents a standardized, consented way to access the customer context they need to act responsibly.

I want to be honest that no single vendor, including Tealium, has fully solved every piece of the agent context problem. What I can tell you is that the organizations we work with who are furthest along share a pattern. They treated customer data infrastructure as strategic before the agent era made it urgent. They invested in identity, consent, and orchestration when those capabilities were useful mostly for better marketing. They are now discovering that the same investments are exactly what agents need, with relatively little new work to adapt them. Three generations of data layer turn out to be a single architectural trajectory, and the organizations that walked it deliberately are the ones best positioned for what is arriving now.

Every article in this series so far has been about infrastructure — what it is, why it matters, what it has to do. Infrastructure is the prerequisite. It is not the destination.

In the final article of this series, I want to spend time on what becomes possible once the infrastructure is in place. Customer representations so complete and so live that organizations can simulate a journey before committing to it. Agents coordinating with other agents on a customer’s behalf in ways that feel, from the customer’s point of view, like a single seamless experience. A shift from managing customer records to maintaining customer representations that grow more useful with every interaction. Some of it is already starting. Some of it is further out. All of it is closer than most people think.

That is the article I want to write next.

The post The Context Has to Keep Up appeared first on Tealium.

Every era of digital has been defined by the data layer that shaped it. The next one has consequences the previous layers did not.

In my last article, I argued that context is the defining variable in applied AI, and that artists have been wrestling with the same problem for centuries. This piece takes that argument for granted and asks a harder question: if context is the constraint, what layer of infrastructure actually produces it?

The answer is not a new feature inside an existing platform. It is a new layer. And the clearest way to see it is to look at the layers that came before.

In the beginning, there was no data layer at all. Tags and tracking calls fired directly from page code. Every new tool was a new dev cycle. It worked at the scale of the early web. It did not survive contact with the first wave of serious digital measurement.

The first structured response was the Analytics Data Layer — a standardized object on the page that any analytics tool could read. Google Tag Manager’s dataLayer is the direct descendant. This was the first time anyone said out loud that the data about what was happening should be decoupled from the tools that wanted to know about it. A modest idea in retrospect. It changed everything about how digital measurement scaled.

The next transition was the Universal Data Layer, and this is where Tealium’s story begins. The UDL generalized the analytics layer into something vendor-neutral: one canonical description of what was happening on a page or in a session that could feed any downstream tool. Tealium built the category. The W3C tried to standardize it. The principle was simple and, in hindsight, obvious: the data model should be independent of the tools that consume it.

Then the surface area expanded. Customers stopped living on web pages. They lived across devices, sessions, apps, call centers, stores, and channels. The Universal Data Layer described events in a moment; it could not describe customers across time. The Customer Data Layer — what the industry now calls the CDP — was the answer. Tealium built this layer too. I have written at length elsewhere about why orchestration matters more than storage, and why the organizations that treated customer data as strategic infrastructure rather than a marketing tool have been compounding advantages ever since.

The CDP era is not over. Its infrastructure is the platform that everything after it stands on. But the primary consumer of that infrastructure is changing, and when the primary consumer of a layer changes, the layer itself has to change with it.

The next layer is the one that feeds AI systems — models, agents, and automated decisioning operating at inference-time latency with real consequences. I have gestured at this layer in passing across most of my recent writing: the context layer, the semantic layer, the AI-ready data stream. Here I want to give it a proper name, because the absence of one is what lets organizations keep treating the infrastructure AI actually needs as “CDP plus some connectors” and building accordingly. Which is roughly the architectural equivalent of calling a smartphone a calculator.

What is the AI Data Layer?

It is not a replacement for the CDP. It is what the CDP becomes when its primary consumer is no longer a marketing activation platform but a model or an agent. Most of what makes this layer necessary I have already argued in other pieces — I will not re-make the case for real-time over batch, for first-party over third-party, or for consented context over raw data. Those arguments hold, and they have held for years. What I want to spend this article on is what becomes structurally different about the layer once AI systems are its primary consumers.

Four things

The consumer is no longer predictable. The CDP era assumed the thing reading from the profile was a known set of systems: a campaign tool, an ad platform, a recommendation engine. You could shape the profile to suit those consumers. The AI Data Layer’s consumers are heterogeneous and proliferating — a fraud model, a support copilot, a recommendation system, an autonomous purchasing agent operating on a customer’s behalf. Each one needs a different slice of the customer, composed differently, at a different cadence. The layer has to compose what it serves based on who is asking and for what, and it has to be able to justify, after the fact, exactly what was sent and why. Static profile modeling cannot do this. The layer itself has to be a composer, not a container.

Governance has to move from activation to composition. I have written before about what happens when you poison a model with unconsented data, and the problem is getting worse as models retain more context and agents take more actions. The architectural response is the point. Governance cannot live at the edge of activation anymore. It has to live inside the layer, at the moment context is composed, before anything reaches the model — because once the model has it, enforcement is too late. Consent, purpose limitation, and data minimization stop being compliance layered on top of the profile. They become properties of the profile itself, enforced per signal, per consumer, per decision.

Identity has to resolve across surfaces that did not exist. The identity graph the industry built for web and app was already insufficient a year ago. Voice assistants, chat interfaces, autonomous agents acting on a customer’s behalf, human agents augmented by models — each is a new surface, and each breaks an assumption the CDP-era graph was designed around. The AI Data Layer has to resolve one human being across surfaces the CDP era never contemplated. The resolution has to be correct, and it has to be fast, because the consequences of getting it wrong compound across an agent’s chained actions in ways they never did across a marketing funnel.

The architectural pattern flips. For a decade, the dominant pattern has been consumer system → data lookup → action. In the AI era, the pattern is customer → contextual layer → AI → action. The layer becomes an input, not a query. Architectures that treat context as something to retrieve on demand will lose to architectures that treat it as infrastructure that is already composed, governed, and ready the moment the model needs it.

This is where Tealium sits, and why I think the argument matters beyond our walls. We built the Universal Data Layer. We built the Customer Data Layer. And the work we are doing now is the next layer — the one between the real customer and the AI systems that reason about them on the customer’s behalf. Not because the CDP category was wrong, but because it was early. Everything that era taught us about identity, consent, real-time assembly, and first-party signal quality turns out to be exactly what the AI era needs, with the stakes substantially raised.

One last thing, because I want to signal where this argument is going next.

Everything I have described so far is about giving AI systems the right context to respond. The harder version is giving agents the right context to act, and to keep acting, over time, in ways that remain faithful to who the customer is and what they have consented to. Agents accumulate context. They make decisions that change the context they will operate in next. They chain actions in environments where wrong context produces real-world consequences, not just bad outputs. The AI Data Layer I have described here is already necessary for chat, recommendation, fraud, and support. For agents, it is the minimum viable architecture. I will come back to that in the next piece.

The frame, as I said in the last article, is the argument. It is now also the infrastructure. Build it deliberately.

The post What is The AI Data Layer? appeared first on Tealium.

In 1917, Marcel Duchamp purchased a porcelain urinal from a plumbing supplier, rotated it ninety degrees, signed it with a pseudonym, and submitted it to an art exhibition under the title Fountain. The object did not change. The plumbing did not improve. What changed was everything surrounding it — the frame, the institution, the implied question. Duchamp understood, perhaps better than anyone in the twentieth century, that meaning is not inside an object. It is constructed around it.

I think about this often when I’m in rooms full of engineers, data scientists, and executives debating the future of AI. We talk about context constantly. Context windows. Contextual data. In-context learning. We debate how much of it to pass, how fresh it needs to be, how to structure it so a model can actually use it. It is the defining technical frontier of applied AI right now. And yet we almost never acknowledge that artists have been solving this problem for centuries. Not metaphorically. Literally. The challenges of context in AI and context in art are structurally the same — and once you see it, the solutions artists have built into their practice offer a surprisingly useful map for what comes next.

Consider what Duchamp actually demonstrated. The urinal, removed from its plumbing context and placed in a gallery, became a question: what makes something art? The answer is the context around it — the institution that frames it, the audience that interprets it, the historical moment that receives it. Raw data has exactly the same property. A customer who visited your pricing page three times in two days is, in isolation, a urinal. The signal is neither meaningful nor actionable on its own. It becomes meaningful only when surrounded by context: who this person is, what they have purchased before, what they looked at before hitting that pricing page, whether they are a new prospect or a churning customer. The same behavioral signal produces entirely different meanings — and should trigger entirely different responses — depending on what surrounds it. This is not a technical insight. It is an interpretive one. And artists have known it far longer than engineers have.

Meaning, though, is only half of what a frame provides. The other half is trust. And the art world has more to teach us about that too.

In the art market, provenance is everything. It is the documented chain of custody that traces an artwork from the artist’s studio to its current location, through every hand that held it. A Vermeer with full provenance commands a different price — and carries a different meaning — than the same painting with gaps in its history. The object may be the same. The context around it is not. We have rebuilt this concept from scratch in data, and we call it first-party data. The principle is identical. Signals with a clear, consented, directly-observed chain of custody are worth more than signals of uncertain origin. When a customer interacts with your brand directly, you have provenance. You know where the signal came from, what it means, and what you are permitted to do with it. Third-party data is art of unknown origin: it might be genuine, it might be a reproduction, and the day the market decides it cares about provenance is the day its value evaporates. We are living through that reckoning now, as third-party cookies collapse and privacy regulations tighten. The art market has always known that provenance is not a compliance formality. It is the foundation of value. We are only beginning to internalize this in data.

If provenance governs what belongs in the frame, composition governs what does not. There is a concept in visual art called negative space — the area around and between the subjects of an image. The empty space. A trained artist knows what an untrained one does not: negative space is not absence. It is composition. It is as deliberate and as important as anything that appears in the frame.

In AI, what you choose not to include in a context window is as consequential as what you do.

Anyone who has worked with large language models has encountered context poisoning — the phenomenon by which an overloaded, noisy, poorly structured context window produces degraded or hallucinated outputs. The model is not failing. The composition is. You have filled the frame with clutter and asked the model to find meaning in the noise. The best prompt engineers and context architects I know work the way painters work: through subtraction as much as addition. They ask what to remove, where negative space gives the signal room to breathe. This is not a technical instinct. It is an aesthetic one. And it is one of the rarest and most valuable skills in applied AI today.

Composition in space has a cousin: composition across time. Artists faced this one first too. In 1994, restorers cleaning Leonardo da Vinci’s The Last Supper in Milan faced a decision that has haunted restoration for decades: which version of the painting do you restore toward? The version Leonardo painted in 1498? The version it had become by 1700, after years of overpainting? The version it was in its most famous cultural moment? Each version is true. Each is a different context in time.

This is precisely the problem of temporal context in AI, and it is underappreciated. A customer who purchased from you eighteen months ago is not the same signal as a customer who purchased last week. An interest expressed two years ago may have decayed entirely, or it may represent a foundational preference that predicts everything that comes after. A behavior that was anomalous last quarter may be a new normal today. Which version of the truth are you feeding your model? As with restoration, there is no single correct answer. There is only the discipline of asking the question explicitly and making a principled choice about temporal context rather than defaulting to whatever data is cheapest to collect. Recency is not always truth. History is not always noise. The skill is in knowing which is which.

Even once you have chosen which version of the truth to work with, another problem remains: how do you describe it? The ancients had a name for this. Ekphrasis — the literary description of a visual artwork. Homer described Achilles’ shield in the Iliad. Keats immortalized a Grecian urn in forty-five lines of verse. The purpose was to bring a visual experience into language, to translate across a sensory gap. Every AI system that attempts to reason about customer behavior is engaged in an act of ekphrasis. The model never sees the customer. It sees a translation: a structured, tokenized, schema-constrained representation of a person’s actions, preferences, and history. The customer browses, clicks, purchases, churns, returns. All of that becomes a description. And something is always gained and lost in translation. What is gained is structure, scale, and comparability. What is lost is texture, ambiguity, and the kind of contextual nuance a human observer would capture instinctively. The question is not whether the translation is perfect — it never will be. The question is how faithful and how complete it is. How much of the original experience survives the conversion into data. This is a data quality question. But it is also a question writers and translators have wrestled with since antiquity.

And once the translation is made, where it lives matters as much as what it says. Imagine the Mona Lisa hanging in a dental waiting room. The painting would be identical, stroke for stroke. The experience would be entirely different — diminished, strange, disconnected from the accumulated cultural context that makes it what it is. The meaning of a great artwork is not only in the object. It is in the institution that frames it, the room that holds it, the audience that approaches it with a certain kind of attention. AI models have the same property, and we underestimate it. The same model, deployed in different products, different user bases, different system prompts and surrounding data environments, will behave in meaningfully different ways. Deployment context is not separate from model performance. It is constitutive of it. A model tuned on e-commerce data and deployed in a healthcare workflow will produce results that are, at best, strange and, at worst, harmful — not because the model is bad, but because the institutional context does not match what the object was made for. Responsible AI deployment is as much a curatorial problem as a technical one. You are not just choosing a model. You are choosing where to hang it, what to hang beside it, and what kind of attention you expect the audience to bring.

I spend a lot of time talking to enterprises about their AI strategies, and the pattern I see, over and over, is organizations deeply focused on model selection — which foundation model, which fine-tuning approach, which inference optimization — and comparatively under-invested in the context layer that feeds those models. This is the equivalent of commissioning a masterpiece and hanging it in a broom cupboard.

I should be direct about why this argument matters to me beyond the metaphor. This is the work we do at Tealium. For more than a decade, we have built the infrastructure that turns raw customer signals into something a model can actually use — a unified, real-time customer profile, assembled from every touchpoint a person has with a brand, governed by explicit consent, and made available to whatever downstream system needs it. And every concept I have described corresponds to a discipline within that work.

Identity resolution — stitching thousands of fragmented observations across devices, sessions, and channels into a faithful representation of one human being — is the ekphrasis problem. It is the act of translation, and the fidelity of that translation determines what every model downstream is capable of understanding. Consent and data governance is the provenance problem — knowing where every signal came from, under what permissions, and what you are allowed to do with it. Real-time data quality and filtering is the negative space problem — deciding, continuously, what belongs in the frame and what is noise that should stay out. The ability to serve historical depth or recent behavior on demand is the restoration problem — letting each use case choose which version of the truth it needs. And routing the right context to the right model at the right moment, in the right deployment, is the museum effect — making sure the artwork is hung in a room where it can be seen for what it is.

The companies winning with AI today are not, in most cases, the ones with the most sophisticated models. They are the ones who have done this work first. They know who their customers are across every touchpoint. They can construct a faithful, temporally coherent representation of customer behavior in real time. They curate what goes into their context windows with the intentionality a gallery director brings to a hanging. They understand that what they don’t send to the model is as important as what they do. This is the work that happens before the model ever runs. It is contextual infrastructure. And it is the differentiating capability of this era of AI — not because the models don’t matter, but because once models reach a certain level of capability, context becomes the primary variable.

Duchamp’s urinal is still provoking conversations more than a hundred years after he submitted it. Not because of the object, but because of the frame he put around it, and the questions that frame forced into the world. The AI systems we build today will be judged not just by the models we chose, but by the contextual frames we constructed around them. By whether we fed them faithful or distorted translations of the world. By whether we curated the negative space with intention. By whether we respected the provenance of the signals we passed. By whether we understood that deploying a model in the wrong institutional context produces not just bad outputs, but a fundamentally different — and potentially damaging — thing.

Artists have always known this. The frame is not decoration. The frame is the argument. It is time we built our AI systems accordingly.

The post What Duchamp Figured Out About Your Context Window appeared first on Tealium.

Over the last few months, more and more people have asked me the same question: what is your point of view on behavioral data in a world increasingly shaped by AI agents?

It is a fair question.

A new narrative is gaining momentum. As AI assistants and autonomous agents become part of everyday buying and service journeys, some are arguing that the behavioral funnel is falling apart. The logic seems straightforward: if software increasingly makes decisions on behalf of people, then the human clickstream becomes less visible and behavioral data loses its value.

My view is a bit different.

I do believe something fundamental is changing. But I do not believe behavioral data is becoming irrelevant. I believe we are moving into a world where behavior is increasingly expressed not only through direct human interaction, but also through delegated actions, agent-led workflows, and machine-mediated decisions.

And that is exactly why this topic matters.

For years, digital teams have treated behavioral data as one of the clearest expressions of customer intent. A search. A click. A repeat visit. Time spent comparing options. A form started but not completed. These signals helped us understand where someone was in a decision journey and, just as importantly, what might happen next.

The question now is not whether those signals disappear altogether. It is whether we are interpreting the next generation of signals in the right way.

Because booking a cruise is not the same as buying groceries.

That distinction matters more than it may first appear. In fact, it may be one of the most important ways to think about the next chapter of customer data strategy.

What Is Actually Changing

Something real is happening. AI agents are moving from experimentation into production use cases across enterprises, and they are being applied to everything from customer service to churn prevention to offer optimization. At the same time, those systems only perform well when they can access unified, current customer context rather than fragmented or delayed data.

So yes, the environment is changing.

But the more useful conclusion is not that behavioral data is becoming irrelevant. It is that behavior is no longer always expressed through direct human interaction. Increasingly, it is being expressed through delegated actions, agent-led workflows, and machine-mediated decisions.

That is a very different claim.

The old model was relatively simple: behavioral data captured what a person did.
The emerging model is more nuanced: behavioral data increasingly includes what a system did on a person’s behalf.

In other words, the signal is changing shape. The interface may be shifting. The underlying intent is not disappearing.

The Mistake in the “Behavioral Funnel Is Dead” Argument

The problem with broad declarations about the death of behavioral data is that they flatten all customer journeys into one generic model.

Not all decisions are equally automatable. Not all categories carry the same emotional weight. Not all conversions rely on the same amount of exploration, reassurance, comparison, or validation. And that means not all funnels will evolve in the same way.

Some journeys will compress dramatically.
Others will remain rich in behavioral context, even if AI becomes an active participant.

This is why industry matters. This is why conversion type matters. And this is why the future of behavioral data is not one story, but many.

Booking a Cruise Is Not Buying Groceries

Imagine two consumers.

The first is planning a cruise for their family.
The second is buying groceries for the week.

Both may use digital tools. Both may be supported by AI. But the role of behavior in each journey is fundamentally different.

A cruise booking is a high-consideration decision. It is infrequent, financially meaningful, emotionally loaded, and often influenced by multiple people. The customer may spend days or weeks exploring destinations, comparing dates, reviewing cabin types, checking cancellation terms, looking at excursions, returning to shortlisted options, and weighing trade-offs between convenience, experience, and cost.

That journey produces a dense layer of behavioral context. It reveals preferences. It reveals hesitation. It reveals momentum. It reveals what matters before the transaction happens.

This is precisely why behavioral data remains so valuable. Transactional data tells you that a booking happened. Behavioral data helps explain what led up to that decision and what the customer was moving toward before conversion.

Now compare that to groceries.

A grocery purchase is often routine, repeatable, low risk, and heavily convenience-driven. In many cases, the customer is not looking for a journey. They are looking for completion. Reorder my usual basket. Optimise for price. Avoid products I rejected last time. Deliver tomorrow morning.

Here, the classic funnel can compress significantly. There may be fewer visible product views, fewer comparison loops, fewer signals of active exploration. A customer may move from intent to fulfilment with minimal interaction because the decision has effectively been delegated.

But that does not mean behavioural data has disappeared. It means the relevant behaviour has changed.

In one category, behavior is exploration.
In the other, behavior is delegation.

That is not the death of behavioral data. It is the evolution of behavioral data.

From Direct Behavior to Delegated Intent

This is the shift I believe many organizations still need to internalize.

Historically, we have treated behavioral data as a record of direct digital actions: pages viewed, buttons clicked, products browsed, content consumed, searches performed.

Going forward, we need a broader model.

We need to understand:

• what the individual did directly
• what an agent did on their behalf
• what a system inferred from past behavior
• what was automated based on routines, rules, or learned preferences

If a customer asks an assistant to shortlist cruise options for August, that is still an expression of intent.

If a household uses an agent to reorder the same grocery basket every Friday, that is still an expression of intent.

If a system automatically filters choices based on known budget, delivery windows, or brand preferences, that is still part of the customer context.

The actor may increasingly be a machine. The intent still belongs to a person.

And that distinction is critical.

Because if organizations treat agent activity as separate from customer behavior, they risk misreading the market entirely. They may conclude that signals are weakening when, in reality, the signals are simply being expressed through a different interface.

Why Identity Becomes More Important, Not Less

As more actions are mediated by software, identity becomes even more central.

AI systems need access to a complete and current view of the customer, not a fragmented collection of IDs and disconnected touchpoints. If the same individual appears as a cookie in one system, a mobile user in another, a CRM contact somewhere else, and an agent-generated session in yet another, then the organisation is not seeing behavior clearly. It is seeing broken context.

This is where many future discussions about behavioral data will be won or lost.

The question is not simply, “Are we collecting enough signals?”
The better question is, “Can we correctly assign those signals to the person, household, or account they actually represent?”

In the age of agentic experiences, that becomes the real challenge:

• not just data capture
• but data interpretation
• not just event collection
• but identity resolution
• not just automation
• but trusted context

What This Means for Brands

For brands, the implication is not to move away from behavioral data. It is to mature the way they interpret it.

They need a model that distinguishes between:

• direct behavior — what a customer did themselves
• assisted behavior — what happened with AI support
• delegated behavior — what a system did on the customer’s behalf
• automated behavior — what happened based on stored rules, preferences, or routines

They also need the infrastructure to act on these signals in real time. Batch-oriented models were built for campaign cycles and delayed analysis. They are poorly suited to environments where AI-driven decisions happen inside live workflows and where even short delays can make context stale.

This is why trusted, consented, real-time customer data becomes more strategic in an agent-led world, not less.

The winning organizations will not be the ones that declare behavioral data obsolete. They will be the ones that learn how to connect human behavior, delegated action, and identity into one coherent view of the customer.

The Funnel Is Not Dead. It Is Becoming More Contextual.

The behavioral funnel is not collapsing in a uniform way. It is becoming more uneven, more category-specific, and more mediated by machines.

In routine purchases, visible human exploration may shrink dramatically.

In high-consideration purchases, behavioral context will remain indispensable.

And in both cases, the real strategic task is the same: understand the individual behind the interaction, even when the interaction is no longer purely human.

That is the shift in front of us.

Not from behavioral data to no behavioral data.

But from behavioral data as a record of clicks to behavioral data as a richer layer of human intent, delegated choice, and machine-mediated action.

The interface may change. The customer does not.

And that is why the behavioral funnel is not dead. It is being rewritten.

The post The Behavioral Funnel Is Not Dead appeared first on Tealium.

DV NYC was a room full of people who own data layers, CDPs, warehouses, and models. This recap is for you: a technical view of where the market is going and five concrete moves you can execute now.

The Moment: AI Agents as a New Channel

We’ve all lived through channel shifts: IVR, web, mobile, apps. AI agents are the next one—and they’re different:

• Agents are decision-makers, not just referrers. They select products, flows, and even vendors on behalf of customers.
• Agentic traffic is high-intent by default. When an agent hits your site or APIs, it’s usually mid–late funnel.
• By 2030, agent-originated transactions will be material. If your stack can’t serve rich, policy-aware context, you’re effectively invisible.

The pattern from every prior shift holds: if you miss the new channel, you lose the next generation of customers.

At DV NYC, we framed the question this way:

Are you building the context supply chain that both humans and AI agents will depend on?

Tealium as the Independent Context Layer for AI

Tealium’s role is straightforward: we sit between your digital properties, data clouds, AI platforms, and engagement systems as the real-time context layer.

Concretely, that means:

• Collection at the edge:
  Client-side and server-side capture across web, mobile, call center, brick & mortar, and more, with low-latency streaming.
• Identity + profiles in motion:
  Unified profiles and audiences that recompute as events arrive, not hours later.
• Data labeling, enrichment, and consent at collection time:
  PII classification, purpose flags, region/policy tags, and business semantics applied before data leaves your environment, so it’s AI-ready and policy-aware.
• Real-time APIs for humans and agents:
  Services like MomentsAPI, server-side connectors, and Functions expose the full customer signal in a single call—identity, traits, predictions, entitlements, and history.
• Independence by design:
  We integrate deeply with AWS, Snowflake, Databricks, and major marketing clouds without tying you to any one of them. In an AI world where models and platforms will change, vendor independence is a feature, not a bug.

Think of Tealium as the trusted context bus your LLMs, agents, and apps can plug into.

How Tealium Works with AWS, Snowflake, and Databricks

As you think about next steps after Digital Velocity, we’re here to help you make the most of the investments you’ve already made in your data stack. Tealium is designed to work natively with the platforms you rely on most, including AWS, Snowflake, and Databricks.

• Tealium + AWS: Capture rich, real-time customer data from web, mobile, and offline channels and stream it directly into AWS services (like Amazon S3, Redshift, Kinesis, and EventBridge) to power analytics, AI/ML, and personalization across your AWS environment.
• Tealium + Snowflake (including the new Native App): Create a bi-directional bridge between your behavioral data and Snowflake. Tealium can feed high-quality, consented customer data into Snowflake and activate Snowflake audiences in real time across channels. With the new Tealium Native App in Snowflake, you can manage key CDP capabilities natively inside Snowflake—reducing data movement while improving governance and time-to-value.
• Tealium + Databricks: Unite real-time customer behavior with your lakehouse data in Databricks to build more accurate features and models, then push high-value segments and predictions back through Tealium for activation across marketing, product, and customer experience use cases.

From Slides to Systems: What Leading Brands Are Doing

Patterns we showcased in NYC weren’t prototypes—they’re live:

• Operationalizing ML and traditional models in real time
  Models trained in AWS/Snowflake/Databricks; scores land back in Tealium; downstream actions (offers, journeys, suppressions) fire within seconds, not after a batch cycle.
• Feeding LLMs and RAG pipelines with governed customer data
  Tealium’s labeling and consent controls determine which attributes are legal and appropriate for LLM context and retrieval.
• Equipping agents with full customer state
  Not just “who is this?” but value tier, churn risk, current intent, sentiment, and recent actions—all available in a single context call.
• Closing the loop on agent outcomes
  Every agent decision (offer, answer, route) and outcome (accept, escalate, churn) is written back into Tealium, so your next prediction and next interaction—human or AI—is smarter.

This is what “from collection to action” actually looks like in 2026.

Five Actions to Take Post–Digital Velocity NYC

You don’t need a 12‑month transformation plan to start. Here are five moves you can make now.

1. Make Your Website Agent-Ready with Tealium iQ

LLM-based agents (ChatGPT, Gemini, Claude, Perplexity, etc.) already crawl your site. They’re not just ranking you—they’re deciding whether to recommend you.

Do this:

• Use Tealium iQ to inject JSON-LD / schema (products, pricing, inventory, FAQs, policies) via your existing data layer.
• Ship it as an iQ extension—no engineering sprint, no code deploy.
• Optionally, personalize schema per visitor (e.g., offers, availability) using MomentsAPI or Data Layer Enrichment, so agents see context that matches the actual user.

Why it matters:
Agents strongly prefer clean, machine-readable, trustworthy content. If your competitor is easier to parse and reason over, they will win the recommendation.

2. Treat AI-Referral Traffic as a First-Class Channel

Most teams are flying blind here. AI referrals show up today as “direct” or generic “referral,” even though they often convert far better.

Do this:

• In Tealium, define a single rule that classifies requests from known AI user agents, referrers, or tracking params into an “AI Referral” channel.
• Write that classification into events and profiles so it appears in your warehouse/BI, experimentation tools, and downstream activations.
• Build journeys and audiences specifically for AI-referred visitors (e.g., shorter education path, more aggressive offers, fewer basic explainer steps).

Why it matters:
You can’t optimize what you don’t measure. AI referral will be one of your highest-intent inbound sources; you should treat it with the same rigor as paid search.

3. Close the Loop with AWS, Snowflake, and Databricks

If you use any of these platforms, you’re probably sitting on models that aren’t fully operationalized.

Do this:

• Stream clean, consented events and profile traits from Tealium into your data cloud (S3/Redshift, Snowflake, Databricks Delta) as the canonical behavioral feed.
• Standardize one or two production models (e.g., churn, upsell propensity, risk score) and publish outputs back into Tealium as profile attributes.
• Use Tealium’s audiences and connectors to wire those predictions into email, mobile, web personalization, ad platforms, and AI agents.

Why it matters:
This is how you turn “we have a model in a notebook” into lift in conversion, retention, and CSAT, not just nicer dashboards.

4. Pilot an “Agentic Front Door” with Tealium as the Context Supply Chain

As agents move from browsing to transacting, you’ll need authenticated, consent-aware endpoints that:

• Accept calls from customer agents (MCP, A2A, A2C),
• Resolve identity,
• Enforce consent and policy,
• Serve the right context and actions back.

Do this:

• Pick one high-value use case (e.g., high-value returns, loan pre-qualification, complex B2B renewals).
• Use MomentsAPI (and Tealium’s emerging MCP flows) to expose a single, composable context payload: identity, traits, ML scores, recent events, and allowlist of permitted actions.
• Log every agent call and result back into Tealium as events, so you can score, QA, and iterate on prompts, policies, and journeys.

Why it matters:
Your “agentic front door” should be above any specific LLM or framework. Tealium supplies a stable, governed context contract while you experiment with different AI providers.

5. Double Down on Consent, Privacy, and On-Device AI

As you expose more data to more models and agents, trust becomes the hard constraint.

Do this:

• Make sure every new AI initiative (on AWS, Snowflake, Databricks, OpenAI, Vertex, Bedrock, etc.) is wired through Tealium’s consent and labeling model from day zero.
• Use Tealium’s Consent Manager and privacy tooling to:
  • Classify attributes (PII, PHI, sensitive vs non-sensitive),
  • Tag allowed purposes (analytics, personalization, ML training, agent context),
  • Enforce filters on what can be sent where.
• Explore on-device AI (mobile SDKs, browser-side models) for use cases where you want personalization without exporting raw sensitive data to the cloud.

Why it matters:
Data quality and latency get you better AI; governance and consent keep you out of the news.

Looking Ahead

When every customer has an agent, every enterprise needs a context supply chain.

Tealium’s job is to be that trusted, real-time, independent context layer—one that works with your investments in AWS, Snowflake, Databricks, and the broader AI ecosystem rather than competing with them.

DV NYC was about ideas. The next 90 days should be about implementation:

• Make your site agent-ready.
• Start measuring AI referral as its own channel.
• Wire Tealium into your data clouds for closed-loop ML.
• Pilot an agentic front door.
• Tighten your consent and privacy posture for AI.

We’re building Tealium for exactly this moment. Now is the time to turn that architecture into durable growth.

The post Digital Velocity NYC: From Collection to Action in the Age of AI Agents appeared first on Tealium.

No time to read? You can also listen to this blog.

In computing, a cache is fast and efficient, but the moment the power cuts, the data is gone. Acquisition-heavy marketing works the same way: it’s fast to spin up but expensive to maintain, and the moment a campaign ends, the connection to the customer vanishes.

Without a persistent data layer, and a unified first-party profile, every new campaign starts from zero. Instead of compounding the value of your previous wins, you are simply paying to reload the cache every time a prospect interacts with your brand.

The Problem: Why Your Growth Strategy Needs a Persistent Memory.

Consider Sarah. On Tuesday, she clicks a paid ad on her laptop; by Thursday, she is browsing your site on her phone. In a traditional acquisition-heavy stack, these are treated as two entirely unrelated events because Sarah on her laptop is seen only as a cookie ID, while Sarah on her phone is seen as a device ID. These fragments of her identity never meet, leaving you with no unified view of her journey.

This lack of persistence is more than an architectural flaw, it is a mounting financial drain. It means your acquisition team might spend their budget retargeting Sarah as a “prospect” even after your retention team has already converted her into a loyal customer. You are essentially paying to reload the same customer into your memory over and over again.

This inefficiency is why acquisition costs continue to climb. As cookie deprecation, GDPR, and browser restrictions make traditional third-party tracking less effective, filling your funnel becomes more expensive every year. The only way to bridge this gap is through a first-party data strategy. According to Tealium’s research, these strategies convert at 4x the rate of third-party guesswork, and that performance gap is only widening.

The Solution: Building a Unified Memory Architecture

To stop the cycle of reloading customers like Sarah, organizations need to move beyond volatile memory and build a foundation on ‘disk’ – a unified, persistent first-party data layer.

A Customer Data Orchestration Platform bridges this gap by acting as a persistent layer. It captures every real-time event and writes it to a unified profile that stays with the customer across every device and interaction. Tealium manages this through three distinct scopes that provide a complete picture of the customer journey:

• Event Scope: Captures exactly what is happening in the current moment.
• Visit Scope: Tracks the context of what has happened throughout the day.
• Visitor Scope: Records what is always true about that person, such as their lifetime value, churn probability, and long-term preferences.

When Sarah returns to your app after her third purchase, the platform doesn’t see a “new device ID”. It recognizes a high-value customer, allowing you to trigger real-time actions based on her actual behavior, rather than wasting your budget on another “prospecting” ad.

The Hard Math of Retention

While the logic of a unified memory architecture is sound, the financial results are even more compelling. Most marketing budgets are heavily weighted toward acquisition, yet the data suggests this is an inefficient use of capital. In reality, it is 5 to 25 times more cost-effective to retain an existing customer than it is to acquire a new one from scratch.

When organizations stop treating every interaction as a “volatile” event and start writing to “disk,” the return on investment becomes exponential:

• Compounding Returns: Danone Nutricia saw a 418% ROI on their Customer Data Orchestration Platform investment within just two years.
• Predictive Churn Prevention: By using an AI-powered loyalty strategy to identify at-risk signals, one energy provider slashed their churn by 50%.
• Deepening Engagement: Sportsbet generated 9 million additional sessions annually by focusing on the behaviors of their existing customer base.

These are the result of a fundamental shift in how data is stored and utilized. By building lookalike audiences from your best “visitor” profiles rather than third-party guesswork, you’re finding the right people.

The gap between those using first-party data and those relying on the “cache” is widening. How long will your organization be able to afford paying the “reload tax” on your current acquisition stack.

The Loop That Compounds 

Acquisition without retention is a leaky bucket: you fill it fast and at great expense, but the value keeps pouring out. The answer isn’t to choose between the two, but to build a unified memory architecture where the ‘disk’ informs the ‘cache’. Sarah is already generating the data you need to acquire the next customer. The only question is, are you writing it to disk?

The post Retention is your Hard Disk. Acquisition is your Cache. appeared first on Tealium.

That is exactly what many automotive organizations are dealing with today. Dealer groups depend on a growing mix of websites, digital retailing tools, chat platforms, trade-in applications, and agency-managed implementations.

That’s the gap the Automotive Standards Council (ASC) was created to close, standardizing how digital behavior is tracked across an otherwise fragmented ecosystem.

Each system captures behavior a little differently, which creates inconsistent event names, missing parameters, and fragmented reporting. The result is not just messy analytics, it’s a weak data foundation that makes it harder to compare vendor performance, trust your reporting, and activate customer data across the rest of your stack.

What Does ASC Actually Solve?

ASC exists to bring consistency to automotive measurement by defining a common, event-based framework for digital tracking, especially in the context of Google Analytics 4 (GA4). Its purpose is to ensure key interactions and business-critical actions are named and structured consistently across vendors and platforms.

What Does This Look Like

ASC specifications define a standardized data layer for automotive, ensuring there’s a consistent foundation for capturing and tracking what actually matters. Here’s what that might look like:

• Standardized Event Names: High-value actions like asc_form_submission, asc_cta_interaction, and asc_purchase.
• Granular Conversion Tracking: Separating sales and service submissions at the source.
• Engagement Signals: Tracking asc_form_engagement and asc_comm_engagement to understand behavior leading up to the conversion.
• Vehicle-Level Detail: Tying interactions to item_id, item_price, and item_make.

ASC brings order to the flexibility of GA4. The goal is simple: make sure the same action means the same thing everywhere, ensuring your entire ecosystem is finally speaking the same language.

Where Do Things Break

ASC does a good job defining what good data should look like. It does not guarantee that your entire ecosystem will produce it consistently on its own. Even in organizations that support ASC, gaps show up quickly:

1. Inconsistent Vendor Adoption: Not every vendor moves at the same speed. Some platforms align well to the specification, while others still send incomplete or inconsistent data.
2. Implementation Fragility: Hard-coded implementations create risk. Manual tagging and vendor-specific logic make tracking harder to maintain every time a site changes or a new tool is introduced.
3. Events Lack Context: ASC captures what happened, but it does not tell you who the customer is, how their behavior connects across sessions, or what should happen next.

This is Where Tealium Fits

Tealium does not replace the ASC. It gives organizations a practical way to implement the standard consistently, govern it centrally, and actually put it to work across the business.

1. Standardization at the Point of Collection

Tealium acts as a central “translation layer.” Instead of hoping every vendor implements ASC perfectly in isolation, Tealium allows you to ingest different event formats and map them to ASC-compliant parameters in real time.

If one vendor calls a lead event form_submit and another uses lead_gen_success, Tealium resolves that conflict and automatically aligns both to the ASC standard in real-time.

2. Enrichment With Customer Context

ASC standardizes the “what,” but Tealium serves as the identity layer that identifies the “who.” By tying standardized events to a unified customer profile, you move from knowing “what happened” to knowing “who it happened to.”

That is what turns a simple vehicle detail page view into a high-intent signal. When those events are tied to a shopper’s history and behavioral context, you can respond across sessions and devices in real time.

3. Activation Across the Stack

Once your data is standardized and enriched, it becomes far more useful. Tealium can distribute ASC-aligned data to your entire stack, GA4, Meta, Google Ads, and your CRM simultaneously. This ensures that your reporting, targeting, and follow-up are all driven by the same consistent signal, rather than fragmented guesses.

4. Vendor Agility Without the Risk

Automotive stacks change constantly. When your data strategy depends on hard-coded, point-to-point implementations, every site redesign or vendor swap creates risk. Tealium acts as an abstraction layer. You can add or replace tools without having to rebuild your tracking from scratch every time the ecosystem shifts.

Making the Standard Actually Work

ASC is a great step forward for the automotive industry. It gives dealers, agencies, and OEMs a shared framework for how digital interactions should be measured.

But a standard is only valuable when it is implemented consistently and connected to results. The real opportunity isn’t ASC compliance, it’s building a governed, flexible data layer that can support reporting and customer intelligence without forcing you to rebuild your tracking every time a new vendor enters the mix.

This is where Tealium fits.

It gives dealer groups a practical way to implement ASC consistently and actually use the data it produces.

The post ASC Is the Standard. Tealium Makes It Work appeared first on Tealium.

We are living in an era where the interface is no longer the only product: the data is the product. Every touchpoint, from a mobile app push notification to a real-time support interaction, relies on a constant stream of high-fidelity information. I see organizations spending thousands on sophisticated activation tools while neglecting the very plumbing that makes those tools effective.

The Complication: The Invisible CX Killer

As investments in AI, real-time orchestration, and hyper-personalization accelerate, a critical gap has emerged between having data and understanding its health. This is the “Visibility Gap.” Most organizations are flying blind, operating on the dangerous assumption that if the data exists in a table, it must be correct.

This assumption is a strategic liability. Data corruption, schema drift, and ingestion latency are silent killers of the customer journey. When a personalization engine fails, it rarely “breaks” in a way that triggers a traditional IT alert. Instead, it fails quietly by serving a “Hello [NULL]” greeting or recommending a product the customer bought ten minutes ago.

The complication is compounded by the rise of “Agentic AI” and autonomous systems. These models do not just report on reality, they act on it. If your data supply chain is polluted, your AI will not just make mistakes; it will automate them at scale. The conversation is no longer just about “collecting” data. It is about the reliability of the entire customer data supply chain that fuels campaigns, journeys, and AI models.

The Question: Why Move Beyond Monitoring?

If we agree that data is a strategic asset, why are we still treating its health as an IT ticket? Historically, organizations relied on data monitoring. It is reactive, using predefined metrics or static dashboards to tell you when something is “down.” If a server fails or a tag stops firing, monitoring generates an alert.

But for a CX leader, “up” is not good enough. A system can be “up” while serving garbage data. This leads us to the pivotal question: how do we transition from reactive monitoring to proactive observability to protect the brand and the bottom line?

The Answer: Data Observability as a Strategic Imperative

Data observability is a proactive and holistic discipline. It focuses on understanding the internal state and behavior of your data systems by looking at the data they produce over time. Instead of simply asking “Is the system up?” observability asks “Is the data that powers our customer experience complete, correct, and timely?”

Common data observability frameworks describe five core pillars that are especially relevant for marketing and CX:

• Freshness Is the data up to date? In CX, “yesterday’s data” is often not useful. If a customer abandons a cart at 10:00 AM, a recovery email at 10:05 AM is helpful: one at 10:00 AM the next day is a nuisance and a risk to the brand.
• Volume Is the data stream complete? If event-level data suddenly drops by 40 percent, observability tools should flag the anomaly before you waste ad spend on a broken audience or a misfiring suppression rule.
• Distribution Is the data within expected patterns? If a “Loyalty Tier” field that is usually well distributed suddenly becomes 99 percent “Null,” your personalization programs and decisioning logic will fail.
• Schema Has the structure changed? If a developer renames a field in a mobile app without telling marketing, downstream triggers like loyalty point updates or AI features can silently break.
• Lineage Where did the data come from and where is it going? Understanding the journey from the initial click to the final CRM record is vital for troubleshooting, attribution, and compliance.

The Visibility Gap: A CX and Trust Problem

To understand why this matters for CX, we must look at foundational UX principles. The Nielsen Norman Group (NN/g) identifies “Visibility of System Status” as the first of ten usability heuristics. The principle is simple: a system should keep users informed about what is going on through timely and appropriate feedback.

In a marketing context, the “system” is the entire brand experience. When data observability is missing, that system becomes a black box. The brand makes decisions based on data that may be incomplete or corrupted. The customer feels this as “journey-level pain points.”

If a customer updates their preferences in an app but the pipeline fails to sync that to the email platform, the CX is broken regardless of how beautiful the app looks. As NN/g notes, “Communication creates trust.” When a system behaves unpredictably because of bad data, the brand relationship is no longer on equal footing. Customers experience inconsistency as indifference.

Why Observability is the Autopilot for CX

In high-stakes environments like AI-powered contact centers or next-best-action engines, the need for observability is even more acute. Traditional IT monitoring might tell you the model endpoint is responding, but CX-oriented observability asks different questions:

• Are there latency spikes causing AI responses to arrive several seconds too late in a live conversation?
• Are key context fields, such as open cases or consent flags, being passed correctly to the AI layer?
• Is the sentiment analysis model misreading customer frustration because of missing data fields?

Consider a cockpit analogy. Monitoring is the gauge that tells you the engine is running. Observability is the autopilot system that correlates wind speed, altitude, fuel, and navigation data to keep the plane on course. For marketing and CX leaders, observability provides the control plane for the customer experience. It shifts teams from reactive firefighting to proactive management of journeys.

The Strategic Impact on Marketing ROI

From my perspective, the most compelling argument for data observability is the impact on the bottom line. Marketing budgets are under more scrutiny than ever. Impersonal or irrelevant marketing is one of the fastest ways to burn through spend and damage brand equity. Data observability directly influences ROI in three ways:

1. Eliminating Waste Without reliable volume checks, marketers retarget “ghost” visitors or target customers who have already converted because the “Purchase” event was never recorded. Observability surfaces these anomalies before significant spend is wasted.
2. Safeguarding AI Investments AI is only as good as the data that feeds it. Data observability acts as guardrails for AI by monitoring the freshness of feature data and detecting schema drift in model inputs.
3. Accelerating Time-to-Insight Analysts often spend 80 percent of their time “wrangling” data. When observability is built into the data layer, organizations reduce the “time-to-trust.” This translates into faster, more confident decision-making.

The Tealium Perspective: Trust in the Data Supply Chain

At Tealium, we view customer data as a supply chain. It must be captured accurately at the edge, unified into a consistent profile, and activated responsibly in real time. Data observability is the quality control discipline that operates across this entire chain.

With Tealium, brands can establish a Universal Data Layer that standardizes events at the point of collection. Tools such as Event Specs and Trace help teams validate event quality and debug issues before they impact audiences. This enables “experience-centric” observability. It is not only about knowing if the database is running: it is about knowing that the “Gold” customer segment is populated with accurate data before a major campaign launch.

Implementing a Proactive Strategy

For leaders looking to bridge the gap between data engineering and CX, three strategic shifts are essential:

1. Shift Left on Data Quality Do not wait for data to reach the warehouse to check its health. Implement observability at the point of collection. Treat data quality issues as CX incidents, not just technical defects.
2. Break Down Ownership Silos Data health is not just an IT responsibility. Marketing and CX leaders must participate in defining requirements. Establish data SLAs aligned to experience needs. For example, “Homepage personalization requires data within 50 milliseconds.”
3. Invest in Real-Time Feedback Loops Static weekly dashboards are not enough. Use tools that provide immediate alerts to the cross-functional teams who can actually take action.

Conclusion: Making the Invisible Visible

Every brand today claims to be “customer-centric.” The real differentiator is whether the data powering those experiences is trustworthy. Data observability matters because it turns data from a fragile liability into a resilient strategic asset.

When you have clear visibility into the health of your customer data, you can innovate with confidence. You can deploy AI with a stronger risk posture, personalize at scale with higher precision, and build long-term relationships based on reliable, contextually relevant interactions.

In the world of CX, what you cannot see can absolutely hurt you. It is time to make the invisible visible and put observability at the heart of your marketing and experience stack.

The post Why Data Observability Matters for Marketing and CX appeared first on Tealium.

An AI model that does not know your customer is just a calculator. It is only when you feed that model the rich, nuanced, and immediate context of a specific human being that it transforms from a generic language engine into a revenue-generating asset. There is no shortage of smart models that are easy to access. The market is flooded with them. But there is a real challenge that marketers, data scientists, and IT realize, and that is the challenge of the pipeline. How do you extract data from a mobile app, ensure the user has consented to its use, format it so a machine can understand it, send it to an inference endpoint, and return a personalized action to the user’s screen–all before they blink?

This is the outermost edge of customer data orchestration. Whether you are leveraging commoditized large language models or deploying highly secure, proprietary algorithms built by your internal data science team, the bridge between your data and your AI dictates your success. What follows is a comprehensive guide to building this bridge, exploring how modern orchestration platforms like Tealium execute this seamlessly, and examining the alternative architectural paths available in the wider ecosystem.

Your Organization Must Have Consent, Context, and Real-Time Streaming Data

Before discussing the mechanical connections to any AI model, we must address the data that travels through those connections. If you attempt to connect a Large Language Model directly to a raw, unfiltered database, the project will fail. The data must be conditioned for intelligence.

First, the data must be deeply contextual. AI models do not possess innate memory; they are stateless. If you send a model a raw clickstream log indicating that a user clicked a specific SKU, the model lacks the reasoning to know if that user is a highly valued loyalty member or a first-time guest. Modern orchestration solves this by transforming those raw, chaotic events into a structured JSON payload–a lightweight, universal text format that both web browsers and AI models can instantly read. This contextual payload travels with the user’s signal. This contextual payload travels with the user’s profile, carrying zero-party data (what they explicitly told you) and first-party data (their historical behaviors, lifetime value, and current session intent). When the AI receives this JSON payload, it does not have to guess; it has the complete story.

Second, the data must stream in real-time. In the world of AI personalization, latency destroys relevance; if you are slow, you are invisible. If a customer adds an item to their cart, hesitates, and begins to navigate toward the exit button, your site and marketing technology must react in milliseconds. Batch processing–syncing data overnight–is useless for in-session personalization. The data pipeline must be built on a real-time event streaming architecture. Crucially, this is not just about the speed of the action; it is about the freshness of the intelligence. Your AI must react to the user’s immediate signals, not to stale behavioral logs pulled from a warehouse weeks or months later. Think of a platform like Netflix: if its recommendation engine suggested what to watch next based on your viewing habits from 15 years ago, rather than the series you finished last night, the suggestions would be completely disconnected from who you are today. When the data is out of date, the personalization fails.

Finally, and perhaps most importantly, this foundation must be governed by strict consent. Feeding unconsented Personally Identifiable Information (PII) into an AI model is a catastrophic regulatory risk. Once an LLM ingests and trains on forbidden data, it is incredibly difficult to force the model to “unlearn” it. A robust orchestration layer acts as a safety valve, evaluating user privacy preferences at the edge and redacting restricted data before it ever reaches the AI endpoint. Consent must be the gatekeeper of the intelligence pipeline.

Path 1: The Fast Track to Generative Intelligence via LLM Connectors

For many organizations, the fastest path to AI ROI does not involve building a model from scratch. The democratization of AI means that foundational models–such as OpenAI’s ChatGPT, Google’s Gemini, Anthropic’s Claude, and Amazon Bedrock–are available as commoditized utility APIs. The challenge is safely marrying your real-time customer data with these public or private cloud endpoints to generate in-session responses.

Tealium addresses this by providing native AI connectors that securely route your structured JSON payloads directly to these commoditized LLMs. This happens entirely in real-time. And Tealium also only sends the data that is entirely necessary, increasing speed and token efficiency. Imagine a user browsing a travel website, looking at family resorts in Hawaii, but they continuously check the cancellation policy. The orchestration layer recognizes this specific behavioral pattern–high intent, high hesitation. Instead of relying on the marketing team to manually predict and hardcode every possible point of friction, Tealium shifts the cognitive load to the AI.

Now suppose the same user’s behavior becomes more erratic–rapidly toggling between resort pages, dwelling on the checkout screen, opening a new tab. Tealium detects this session anomaly. To protect token efficiency, minimize cloud costs, and ensure millisecond latency, Tealium does not send the user’s entire multi-year database record. Instead, it extracts a surgical micro-payload containing only the essential context: their loyalty tier, their current cart value, and their last five event actions.

Tealium routes this lean payload to an LLM like OpenAI via a secure connector, accompanied by a dynamic prompt engineered by the marketing team: “You are a high-end travel concierge. Analyze this user’s recent session events and loyalty status. Identify their likely emotional state–are they frustrated, comparing prices, or searching for reassurance? Based on the specific friction you deduce from their session data, write a two-sentence, highly personalized intervention to assist them.”

OpenAI processes the micro-payload, infers that the rapid toggling between the booking page and the FAQ indicates anxiety over hidden fees, and generates a tailored response guaranteeing price transparency for VIP members. Tealium then instantly injects that exact messaging into the live web session. The customer feels understood, the conversion is saved, and the enterprise leverages the true reasoning power of the LLM without overwhelming its token limits. This is the power of bringing commoditized intelligence directly into the live customer experience.

Path 2: Powering the Enterprise Engine with Data Clouds and “Invoke Your Own Model”

While generative LLMs are transformative for content, true enterprise differentiation lies in proprietary predictive models. A financial institution’s custom fraud detection algorithm or a retailer’s bespoke churn-prediction model represents their unique intellectual property. These models are typically built, trained, and hosted by internal data science teams within massive Data Clouds.

To support this, the data architecture must be bi-directional. Tealium provides deep Data Cloud connectors to industry giants like Databricks, AWS, Snowflake, and Google Cloud. On the outbound side, Tealium streams the pristine, consented, and structured event data directly into these environments. This provides the data science teams with a continuous, clean supply of high-fidelity data to train their models, eliminating the need for them to spend months doing data engineering work..

But training the model is only half the equation; the business must then use that model to influence the customer. When a model predicts that a user has a 90% propensity to buy a premium product, that insight is useless if it sits in a data cloud until the next day.

This is where the architecture pivots from training to inference, utilizing capabilities like Tealium Functions to ‘Invoke Your Own Model.’ When a user takes an action on the website, Tealium triggers a serverless computing function at the edge. For the cloud architects in the room, this isn’t a clunky container that takes seconds to spin up. It is a globally distributed execution environment optimized for sub-millisecond invocation, mitigating the cold-start latency that traditionally plagues serverless architectures during high-traffic spikes.

Because the execution environment is highly performant, developers can write a few lines of JavaScript within the Function to aggressively parse the data in flight. Instead of sending a massive, slow payload of raw logs to the enterprise’s cloud inference endpoint, the function strips the payload down to include only the exact, minimal features the custom model requires–perhaps just recency, frequency, and lifetime value. This surgical micro-payload is fired to the Databricks or AWS endpoint via a RESTful API call–a standardized, high-speed digital handshake between systems. The proprietary model evaluates the data, scores the user, and returns the prediction to Tealium instantly. Because the payload was highly optimized and the edge execution avoided a cold start, the entire round-trip takes milliseconds.

Tealium can then use that score to immediately alter the customer’s live session, perhaps by upgrading their shipping options to secure the high-value conversion. The data science team protects their cloud compute costs, and the marketing team activates proprietary AI in real-time.

Path 3: The Frontier of Agentic Configuration

As the technology evolves, enterprises needs are moving beyond isolated predictive scores and generative text. We are entering a time where AI does not just answer questions, but autonomously executes multi-step workflows to achieve a business goal.

To connect AI to customer data in an agentic framework, the orchestration platform should act as both the brain and the safety rails. These agents are tied to custom prompts and are granted governed access to act on your behalf using the rich zero-party and first-party data residing in the Tealium customer profile.

Unlike a simple predictive model, an agent would be goal-oriented. A marketer could configure an agent with a specific directive: “Maximize margin while preventing cart abandonment for VIP users.” As a VIP user shops, the agent evaluates the live data stream. If the user shows signs of abandoning a cart, the agent autonomously decides the next best action. It doesn’t just pull a pre-written email; it might choose to dynamically generate an offer for loyalty points instead of a blunt 20% discount, recognizing through the user’s profile that points are a stronger, more cost-effective motivator for this specific individual.

Crucially, this agentic configuration requires a strict separation of policy and optimization. The marketer sets absolute boundaries (the policy layer)–for instance, dictating that the agent must never contact users who have opted out, and must never exceed a certain discount threshold. The AI agent operates safely within these deterministic guardrails, optimizing the outcome without ever posing a risk to brand safety or regulatory compliance. Everything occurs in the same session, turning the digital experience into a fluid, hyper-personalized negotiation between the user and the brand’s autonomous agent.

There is a Need for True Observability to Keep the AI Honest

As organizations hand over more decision-making power to AI models and autonomous agents, a critical question arises from enterprise architects and compliance officers: When the AI misfires–and it eventually will–how do we trace the error? Black-box decisioning is a massive liability. If a generative model hallucinates an inaccurate return policy in a chat window, or a custom model calculates an illogical discount, the engineering team cannot spend weeks digging through disparate server logs to find the root cause. When an AI-driven personalization fails, you must be able to debug it instantly.

This is where the orchestration layer doubles as your system of record for observability. By routing your AI pipelines through a centralized hub like Tealium, you maintain absolute auditability. Every interaction leaves a deterministic footprint. If an anomaly occurs, architects and compliance teams can inspect the exact lifecycle of the decision. They can see the precise JSON decision payload that was sent at the exact millisecond of invocation, the specific prompt or configuration parameters that accompanied it, and the raw response returned by the model.

When you can instantly replay and audit the exact inputs and outputs of your intelligence pipeline, you remove the fear of the unknown. You transform AI from an unpredictable, unmonitorable black box into a measurable, highly governed enterprise asset.

Alternative Architectural Approaches: Connecting Models Without a CDP

While a real-time customer data orchestration platform like Tealium provides the most streamlined and secure bridge between data and AI, it is not the only architectural pattern in the ecosystem. It is vital to understand how the broader market approaches this challenge, as many organizations attempt to build this pipeline using disparate tools.

Composable CDP –  Over the last few years, this movement has gained significant traction. In this architecture, the central Data Warehouse (like Snowflake or BigQuery) acts as the absolute center of gravity. Data science teams build and train their models directly where the data lives, generating predictive scores as new columns in the warehouse tables, and using Reverse ETL tools to query the warehouse and push those scores out to marketing platforms.

• Advantage: This is objectively the superior architecture for deep, retroactive analytics and highly complex, batch-based audience segmentation. If your primary goal is to run computationally heavy models that require joining massive historical datasets across months of behavior (like calculating 12-month churn probabilities or multi-year Lifetime Value), doing this directly where the data rests makes perfect sense. It maximizes your existing data warehouse investment and keeps data science workflows centralized in a familiar SQL environment.
• Drawback: The limitation emerges when the business requirement shifts from insight to instant action. Because data must be ingested into the warehouse, processed, scored by the model, and then queried to be pushed outward, the latency floor is typically measured in minutes or hours. It is fundamentally incompatible with in-session personalization. Additionally, governing real-time consent signals as they traverse these disconnected batch pipelines remains an immense engineering hurdle. Another significant drawback is the total cost of ownership. This model requires significant engineering resources and very high compute costs in the data clouds.

Custom Event Streaming (Pub/Sub Microservices) – Highly mature engineering organizations sometimes choose to build the entire real-time pipeline from scratch. They deploy open-source event streaming platforms like Apache Kafka or cloud-native equivalents like Amazon Kinesis. They capture data at the edge and build custom “pub/sub” (publish and subscribe) microservices–independent, highly specialized pieces of code that act like operators on a switchboard, subscribing to specific data streams, passing them to internal AI models, and publishing the results back to the frontend.

• Advantage: Let us be clear: if you have an elite engineering organization, this approach offers a level of architectural purity and absolute, unbounded flexibility that no vendor can match. You own every line of code. For organizations where high-speed data streaming is the literal core of the product (e.g., ride-sharing applications, high-frequency trading platforms, or global streaming services), this bespoke architecture is genuinely superior. Your microservices subscribe to the exact data they need and process it exactly how you dictate without any vendor-imposed limits.
• Drawback: Drawback is a staggering Total Cost of Ownership (TCO). For most enterprise brands, building and maintaining a bespoke real-time orchestration platform diverts top-tier engineering talent away from core product innovation. Every time a new global privacy regulation is passed, or the marketing team wants to test a new commoditized LLM, those highly paid software engineers must pause feature development to rewrite data pipelines, update JSON schemas, and manually enforce new governance rules. It redirects senior engineering talent toward ongoing pipeline maintenance.

Monolithic Marketing Cloud – The legacy approach relies on monolithic marketing clouds (like Salesforce or Adobe). These vendors are aggressively building their own AI capabilities (like Einstein or Sensei) directly into their suites. To connect data to these models, organizations simply implement the vendor’s proprietary tracking tags and use the suite’s native features.

• Advantage: For organizations that have consolidated their entire digital operations into a single vendor’s ecosystem, utilizing these native, built-in AI capabilities is the path of least resistance. It requires virtually no custom data architecture. The marketing team can activate predictive scores natively within the interfaces they already use every day. If your use cases are heavily localized to email marketing or standard web personalization within that specific suite, this unified approach is incredibly efficient, marketer-friendly, and easy to train staff on.
• Drawback: There is an inevitable “black box” limitation and vendor lock-in. These native AI models are often constrained by the data that exists purely within their specific cloud, leading to fragmented intelligence if you use outside tools for point-of-sale, customer service, or mobile apps. Furthermore, if your internal data science team builds a brilliant, proprietary fraud or propensity model in Google Vertex AI or AWS, piping that external intelligence into the legacy marketing cloud’s native decision engine in real-time is notoriously rigid and difficult.

Infrastructure is Destiny

The evolution of artificial intelligence is moving faster than any enterprise can adapt to on an algorithm-by-algorithm basis. If you build your digital strategy around the specific capabilities of today’s models, your architecture will be obsolete in a matter of months.

The true differentiator for the modern enterprise is not the model itself, but the infrastructure that supports it. To truly capitalize on the AI revolution, organizations must build an intelligence pipeline that is agnostic to the model but fiercely protective of the data. By leveraging real-time, consented, and contextually rich data, and utilizing a centralized orchestration layer to route that data to commoditized LLMs, enterprise data clouds, and autonomous agents, businesses can finally close the gap between artificial intelligence and actual revenue.

You can always swap out the AI engine when a faster one comes along. But the organization that owns the most robust, real-time data refinery controls the future of the customer experience.

The post How to Connect AI Models to Your Customer Data appeared first on Tealium.

The agent reasoned correctly with the data it had. The data was only 90 seconds old.

Ninety seconds is nothing for a weekly analysis.

But for an agent acting inside a live workflow, it’s a lifetime.

The Shift from Reporting to Acting

Enterprise data systems were built to answer questions.

What happened last quarter? How did the campaign perform? Which segments converted?

Reports get read, decisions get made, and the data’s freshness is measured in hours or days. That’s fine, because a human is in the loop.

Agents collapse that loop. They observe, decide, and act inside the interaction, not after it. A churn-prevention agent has to catch the signal before the cancellation click. A service agent has to know a delivery failed before the customer complains. A personalization agent has to recognize intent while the session is still open.

When the agent is the decision-maker, latency stops being a performance metric and becomes a capability boundary. Data your agent can’t see in time is data that might as well not exist.

Why Traditional Data Pipelines Buckle

Most enterprise data architectures look something like this: events are collected at the edge, queued for batch ingestion, loaded into a warehouse, transformed, modeled, and eventually made queryable by operational systems. Each hop introduces delay.

In batch-first environments, those delays stack up in predictable ways:

• Ingestion on scheduled loads every 15 minutes to 24 hours
• Transformation jobs (dbt, SQL) that run on a cadence, not on-demand
• Identity stitching and profile updates that depend on upstream freshness
• Reverse ETL syncs pushing profiles back out to activation channels

End to end, many enterprises are operating with customer data that’s 30 minutes to 24 hours old by the time any downstream system can touch it.

Most of this infrastructure works exactly as designed, and that’s the problem.

The Latency Math Agents Actually Require

Humans tolerate a few hundred milliseconds of lag before a system feels unresponsive. The Doherty threshold, established in the 1980s and reaffirmed in modern UX research, puts that number around 400 milliseconds. Slower than that and engagement falls off.

Agents working inside those interactions have to respect the same physics. That means the full round trip fits inside a budget that looks more like this:

• Event captured and enriched: under 500ms
• Profile updated with the new signal: under 1 second
• Agent queries context and decides: under 1 second
• Action dispatched to the activation channel: under 500ms

Total budget: roughly 2 to 3 seconds from event to action, with sub-second targets at every hop. That’s the floor for anything that feels like real-time decisioning, not the ceiling.

If your data platform takes 15 minutes to reflect a new event in a customer profile, your agent is reasoning about a customer who no longer exists.

Analytical vs. Operational Latency

Two kinds of latency matter here, and it’s important to not confuse them or you’ll end up focusing on the wrong upgrades.

Analytical latency is the delay between an event happening and the data being available for analysis. This is what gets measured when someone asks “how fresh is our warehouse.” It feeds dashboards, training datasets, and forecasting models. Improving it is valuable. It’s also largely irrelevant to agents.

Operational latency is the delay between an event happening and a system being able to act on it. This is the number that governs whether an agent’s decision is still useful when it lands.

You can have best-in-class analytical latency and still fail operationally. A streaming warehouse that refreshes every 90 seconds is a miracle for analysts.

In agent-based systems, latency compounds across the pipeline. Each component can look fast in isolation and still add up to something unusable in aggregate.

A typical flow looks like:

Customer event → Event collection → Enrichment →
Identity resolution → Profile update → Context retrieval →
Agent reasoning → Action execution

Each hop has its own latency budget. Most teams only measure the ends. They know how long the event took to hit their pipeline and how long the agent took to respond. They have no visibility into the 4 to 6 stages in between.

This is where systemic latency hides. Individual components meet their SLAs. The end-to-end experience still misses the window. Optimizing any single stage won’t close the gap. Collapsing the pipeline will, because fewer stages means fewer places for latency to compound.

Architectural Patterns That Actually Work

Teams building for agent-based workloads are converging on a handful of patterns that treat data flow as continuous rather than periodic. None of them are optional.

Event-driven collection at the source. Events enter the system the moment they occur, not on an ingestion schedule. This is where Tealium lives: capture the event, normalize it, and emit it downstream in a single motion measured in milliseconds. The entire architecture downstream depends on this point being fast and clean.

Streaming pipelines with no hidden batch re-entry. Moving to Kafka, Kinesis, or equivalent streaming infrastructure only helps if nothing downstream quietly reintroduces batch. A single nightly reverse ETL between your warehouse and your activation layer re-imposes all the latency you just paid to eliminate. Audit the full path. If any stage runs on a cron, you still don’t have a streaming pipeline.

Profiles that update in real time, not on a schedule. This is where most stacks break. The warehouse gets fresh events, but the customer profile the agent queries was snapshotted from a job that ran three hours ago. Tealium handles this by updating profiles continuously as new signals arrive, so the profile the agent retrieves reflects the session that’s still happening.

Low-latency context retrieval. Agents querying profile data need sub-100ms response times. That means the serving layer has to be purpose-built for that access pattern, not borrowed from the analytics warehouse that was never designed for point lookups at scale. Your warehouse is not your agent runtime.

Incremental computation over full recomputation. When one attribute changed, don’t re-run the full model. Update what moved. This sounds obvious. It’s not how most enterprise data stacks actually work today.

The Role of the Customer Data Layer

Agent decisions are only as good as the context they can reach. In customer-facing workflows, that context lives in the customer data layer: behaviors, transactions, profile attributes, consent signals, and interaction history across channels.

For agents to operate reliably, the customer data layer has to support four things at operational speed:

1. Real-time ingestion of behavioral events across every channel
2. Continuous identity resolution so profiles stay stitched as signals arrive
3. Sub-second profile availability at the moment of query
4. Governed, consented access so agent actions stay compliant with privacy commitments

Tealium is built around exactly this premise: give agents a profile they can trust, delivered fast enough to act on, with the consent and governance plumbing already in place. Without that foundation, agents work with fragmented, delayed, or non-compliant data. Output quality falls off accordingly.

What to Measure

If you’re operationalizing agents, you need latency metrics that match the job. A warehouse freshness dashboard won’t tell you whether the agent has what it needs.

Instrument these four:

• Event-to-availability latency: time from event occurrence to data being queryable by operational systems
• Profile update latency: time from new signal to updated profile attribute
• Agent decision latency: time from agent invocation to action selected
• End-to-end response time: time from the originating event to the action landing in the customer’s experience

Track the 95th and 99th percentile, not just the average.

What This Means for Your Roadmap

Model accuracy won’t determine which organizations get agent-based systems into production. Data latency will.

Audit your pipeline for batch dependencies that quietly reintroduce delay. Measure operational latency separately from analytical latency. Invest in real-time profile serving, not just real-time ingestion. And treat the customer data layer as the foundation of your agent strategy.

Agents will keep getting smarter. That’s the easy part. The hard part is getting them fresh data fast enough to matter. The organizations that solve that problem will run agents that feel useful. The ones that don’t will run agents that feel like they’re always a step or two behind.

The post Agents Don’t Wait: How Agent-Based Systems Change Data Latency Requirements appeared first on Tealium.