The center of gravity in the customer data stack has moved to the data cloud, and it makes sense. But storage consolidation was sold to the market as stack consolidation, and marketing teams are being asked to give up something they were never supposed to hand over.
Something real happened over the last three years. The cloud data platform became the system of record for the customer. Snowflake, BigQuery, Databricks, and the lakehouse patterns around them now hold the history, the transactions, the product telemetry, and the governed truth about who a person is. That is a resolution to a problem the industry spent a decade failing to solve with point tools, and marketing benefits from it more than any other function in the company.
But along the way someone claimed that because the data settled in the warehouse, everything the industry does with customer data should settle there too. Identity resolution. Segmentation. Decisioning. Activation. The moment itself. A whole vocabulary grew up around the idea: zero-copy, composable, warehouse-native.
Storage moving is a fact. The rest is a product strategy bet, and it might be a longer shot than some of the people selling it want you to believe.
Consider what actually gets offered as evidence. Analyst coverage has expanded to include composable and warehouse-native architectures, and vendors from that lineage have moved up the rankings. That does not mean enterprises have relocated their activation layer, because relocating the activation layer is not what analyst placement measures.
And when you read that a growing share of CDPs now support a warehouse-centric architecture, notice what the number counts: vendors who shipped a capability. It does not count production deployments where a marketer, unassisted, launched a real-time campaign off warehouse-resident logic.
That number is a lot smaller than the conversation around it suggests.
I am not arguing the trend is fake. I am saying that it is narrower than the way it gets talked about, and it has been generalized past the point where it is true.
What marketing has the opportunity to gain
The wins here are real, and they are the reason the migration was the right call in the first place.
Start with a single definition of the customer. The meeting where the email platform says the segment is forty thousand people and the ad platform says thirty-two thousand and no one in the room can explain the gap is finally over because there is one place the number comes from.
That alone repairs most of marketing’s standing credibility problem with finance.
Then there is the complete history of the customer instead of just a subset. Raw events, transactions, service interactions, and product usage sit next to each other now. Segments that were impossible because the joining data lived in three systems are now just a query somebody can write.
Now governance will hold up under an audit as well. Consent state, retention, and access control can live under one enforcement model. Anyone who has ever had to prove to a regulator that a suppression actually propagated everywhere downstream knows what that is worth.
What marketing is being asked to give up, and should not
The first thing at risk is control over the segment definitions. These now live in a model owned by data engineering, and marketing's access to its own logic is often a ticket in a queue. The predictable result, and I have watched it happen at real companies, is that the number of ideas a marketing team can test in a quarter goes down in the first year after the migration.
The second thing is the issue of speed. The warehouse-first model is batch by default, and for most consumers of enterprise data it does not matter. Finance can wait until tomorrow; however, marketing cannot. An abandoned cart is worth something for about twenty minutes. In-session intent is worth something for about ninety seconds. Resolving a profile beautifully six hours after the visitor left misses the moment the campaign was built to deliver.
Third, and I don’t say this lightly, there is a significant cost, now, of curiosity. It’s important to understand that segmentation used to live under a license, and exploring it was effectively free once you had paid for the seat. Every experimental query and every rebuilt audience carries a compute cost. Teams stop asking questions because every answer shows up on an ever-increasing invoice. Good segments come from poking around, and poking around is at risk when the meter is running.
Lastly, copies of data can be a risk. Zero-copy describes a boundary around the warehouse, not around the entire stack. The instant an audience activates into Meta, or Google, or the email platform, data has been copied. However, this means that there are fewer copies than before, sharing a common parent, and that is great progress. But the promise that the customer record never leaves holds right up until the moment marketing does anything with it. Reverse ETL is a delivery mechanism, and a good one. The trouble starts when it gets bought as a nervous system.
The interface problem is not a phase
Marketers do not want tables. They want to be able to make a decision. The warehouse offers tables and a query interface, and the industry's response has been to layer a semantic model and a sync pipe on top and declare the experience solved. It is not solved. Every warehouse-native deployment I have seen struggle, struggled exactly here, even if the architecture diagram was correct in all of them.
The current answer is that AI closes the gap, and that conversational interfaces make the underlying complexity irrelevant. The concept is right but the sequencing is wrong. Natural language interfaces using stale data do not create a real-time marketing team. They just surface yesterday's data more conveniently and can create an even greater problem because the results are presented so confidently.
The position marketing should actually take
What counts as an active customer, a churn risk, a high-value household, is a commercial definition owned by the people accountable for the revenue it drives. But right now those definitions are being managed by whoever writes the model and that is almost always not marketing or anyone with a marketing background.
Marketing needs to take this ownership back. Marketers do not need to write SQL, but they need to know what a model means, which tables are certified and which are somebody's abandoned experiment, and how fresh the data they are activating actually is.
The stack should be rigid at the center and flexible at the edges, and the test of the edge is whether the person doing the job can complete their tasks without filing requests.
The warehouse is memory, and it is the best execution of memory the industry has had yet. It should hold the durable, governed, complete truth about the customer, and marketing should insist that it does. But memory is not a nervous system. The signals that determine the outcome of a marketing interaction are created, resolved, and acted on in the session, at the source, in real time, before any batch process has been scheduled. That layer is not a gap waiting for the warehouse to get around to it. It is a different function with different physics, and the teams that run both get both.
The center of gravity has moved, and marketing should be glad it did. But gravity pulls everything toward the center. It does not decide what happens next. What happens next is decided in the session, at the source, while the customer is still there, drawing on everything the warehouse remembers. Memory in one place, decisions in the moment. Marketing was never supposed to choose between them.
