Ask a vendor whether the customer data platform has a future and the answer is always yes, usually with a slide. The honest answer is more interesting, because the question is not really about whether the work survives. Resolving identity across devices, holding consent, turning a pile of warehouse rows into a profile something can act on: that job is not going anywhere. The real question is whether it survives as a category, a line item a company evaluates, shortlists, and signs a separate contract for.
Those are different things. A category can quietly stop existing while the function it named lives on, absorbed into something else. What follows is three scenarios for where the CDP stands in 2028, the forces behind each, the signals that tell you which is winning, and the part most coverage skips: the answer is probably not the same for every buyer.
How the category got fragile
The CDP started as a clear answer to a clear problem. Through the 2010s, customer data sat scattered across a CRM, an email tool, a web analytics tag, a mobile SDK, and a dozen point products, none of which agreed on who a customer was. The CDP collected first-party, person-level data from all of it, resolved identity into one persistent profile, and let any downstream tool use that profile. It had a name, a trade body, and a tidy pitch.
Two things since then have made the category less stable than it looks.
The first is that the standalone CDP disappointed a lot of the people who bought one. The CDP Institute's own member research found the share of deployed CDPs delivering significant value had fallen to around 64 percent, and the picture for measurable return is worse: McKinsey interviewed more than 50 senior marketing leaders at large companies and none could clearly articulate the return on their martech spend. A tool buyers cannot call a clear win has an exposed budget line.
The second is architectural. When the CDP was conceived, there was no obvious other home for unified customer data. Now there is. Cloud data warehouses, Snowflake, Google BigQuery, Databricks, Amazon Redshift, became where enterprises already keep everything else, and a second master copy of the customer inside a separate CDP started to look like duplication. That is the provocation behind David Yaffe's widely shared argument that the warehouse is unbundling the customer data layer. A category is fragile when its core asset, the unified profile, can credibly live somewhere else.
Put those together and you get a market still growing in revenue but no longer growing in identity. The CDP Institute's January 2026 industry update counted 204 active vendors, with 2025 producing 19 new entrants against 7 exits, and noted that almost all the new arrivals were established firms bolting a CDP onto an existing product rather than fresh pure-play startups. New money is not betting on new standalone CDPs.
Scenario one: the CDP dissolves into the warehouse
Here the CDP loses its separate existence downward. The warehouse becomes the customer data platform, and what used to be a CDP becomes a thin set of features running inside or directly on top of it. The force is the composable, or warehouse-native, model taken to its conclusion. Today a composable CDP leaves data in the warehouse and activates it in place with reverse ETL, so the warehouse is the source of truth and the CDP is a layer of logic above it. The CDP Institute reports that more than a quarter of CDPs now support warehouse-centric architecture, and that composable vendors grew employment about six times faster than the industry average, 7.8 percent against 1.3 percent. The next step is the warehouse vendors absorbing the marketing-specific jobs themselves, identity resolution, segmentation, activation, and governance, as native features or first-party add-ons. Snowflake already markets a Customer 360 solution. Once the warehouse does enough of the work, the separate CDP contract is hard to justify.
The standard rebuttal is fair: storage and compute are not marketing logic. A warehouse does not know what identity resolution should do, how to model consent, or how to push a decision to a channel in seconds. But that rebuttal weakens every quarter the warehouse vendors close the gap. The scenario does not need the warehouse to become a great CDP, only a good enough one that a separate platform stops being worth its cost.
Signals that this scenario is winning: warehouse vendors shipping native identity resolution and audience tooling rather than leaving it to partners; reverse ETL pricing collapsing toward commodity; and buyers describing their setup as "our Snowflake stack" rather than naming a CDP at all.
Scenario two: the CDP dissolves into the suites
Here the CDP loses its separate existence sideways. It gets folded so completely into large marketing and customer experience suites that it stops being something anyone buys on its own. You do not purchase a CDP. You purchase Salesforce, Adobe, or Microsoft, and a CDP is part of what is inside.
This is already visibly underway, which is what makes the scenario credible. Salesforce renamed its CDP from Data Cloud to Data 360 and positions it not as a product you weigh against rivals but as the data foundation under its entire application portfolio and its Agentforce agents. Adobe went further and retired the Experience Cloud brand altogether, replacing it with Adobe CX Enterprise, an AI-first system with the Real-Time CDP as an internal component rather than a headline. When the vendor stops marketing the CDP as a distinct thing, the category has half dissolved on that side of the market.
Gartner calls this path platformization, and its 2026 Magic Quadrant for Customer Data Platforms frames the market as splitting between it and the agentic path. The appeal for a buyer is genuine. When the data layer and the applications come from one vendor, the seams that cause most integration pain disappear: identity is enforced once, a consent change ripples everywhere, and decisioning reaches across marketing, service, and commerce on one profile. The consolidation wave makes the same point from the supply side. Uniphore bought ActionIQ, Contentstack bought Lytics, Fivetran bought Census, and Rokt bought mParticle for around 300 million dollars. Standalone CDPs keep being absorbed into something larger.
The cost is lock-in. A CDP that is the foundation of a suite cannot be pulled out without confronting the whole building. But this scenario is about category survival, not buyer happiness, and every CDP that becomes an invisible layer of a suite is one less standalone member the category can count.
Signals that this scenario is winning: more suite vendors dropping "CDP" from product names; analyst inclusion criteria tightening until few independent CDPs qualify, as the 2026 Magic Quadrant did by dropping four; and the continued thinning of pure-play independents through acquisition.
Scenario three: the CDP is reborn as the agent context layer
This is the only scenario in which the category does not just survive but gets a sharper reason to exist. The CDP is reborn as the agent context layer: the governed place AI agents read customer data and act on it. The force is the rise of agents as the primary consumer of that data. Agents do the reading and deciding a marketer once did by hand, and Gartner expects 40 percent of enterprise applications to embed task-specific AI agents by the end of 2026, up from under 5 percent in 2025. An agent deciding what to do for a customer needs to know who that customer is, what just happened, and what it is permitted to do. That is context, and it is what a CDP produces.
Why this is a distinct layer rather than just a warehouse table is worth being precise about. A raw warehouse gives an agent rows. It does not give a resolved identity, a current consent state, a freshness guarantee, or a curated set of decision-relevant traits. Twilio, which owns Segment, makes the argument plainly: a CDP is well suited to be the contextual layer for agentic systems because it validates events before they reach an agent, keeps one consent-aware profile, refreshes in seconds, and hands an agent a focused handful of traits rather than an unreadable dump. Databricks frames the same need as a customer context layer separating who the customer is from what they are doing now. The CDP's job is the closest fit for that service.
The catch is that this rebirth is conditional. An agent context layer is only useful if the context is trustworthy, which means the unglamorous work has to be done first. Agents have no tribal knowledge. If "active customer" means one thing to finance and another to growth, a human disambiguates and an agent does not. That is why Snowflake, with dbt Labs, Salesforce, and others, launched the Open Semantic Interchange to standardize machine-readable definitions. A CDP that does not solve identity, consent, freshness, and shared definitions just becomes a faster way to feed agents bad data.
Signals that this scenario is winning: CDPs exposing profiles through agent-facing interfaces rather than dashboards; governance, lineage, and consent becoming the features vendors compete on; semantic-layer standards getting real adoption; and buyers evaluating a CDP by what agents can safely do through it.
The answer is probably segment by segment
Most writing on this treats the three scenarios as a single race with one winner. The more defensible view is that they are not competing for one outcome. They are each winning, in different parts of the market, at the same time.
For large enterprises already standardized on Salesforce, Adobe, or Microsoft, scenario two is effectively the present, and the standalone category is mostly gone for them. For digitally mature companies with real data engineering talent and a warehouse they trust, scenario one is the natural path, and the line between their warehouse and their CDP is already blurred. For organizations serious about putting agents in front of customers, scenario three is the live question, and it cuts across the other two, because a suite CDP and a warehouse-native CDP can each try to become the context layer.
The mid-market is shaping up as its own case. Analysis of the 2026 market argues that as enterprise CDP adoption plateaus, the mid-market becomes the segment driving net-new growth, and that those buyers want neither a heavyweight suite nor a build-it-yourself composable stack. For them the standalone, packaged CDP, lighter and activation-focused, may stay a real and separate category precisely because the two dissolving forces do not reach them cleanly.
What to do now, whichever way it breaks
The useful response to this uncertainty is not to guess the winner. It is to make decisions that hold up across every scenario.
Invest in the data foundation, not the destination. Identity resolution, consent management, and shared metric definitions are load-bearing in all three scenarios. The warehouse needs them, the suite needs them, the agent context layer needs them most. That work is never wasted, and no vendor switch fixes it for you. Getting "revenue," "active customer," and "churn" defined once takes longer than buying software, and agents act on whatever those definitions say.
Keep the customer data portable. The single biggest risk across these scenarios is being locked into a layer that turns out to be the wrong one. Favor architectures where your customer data and your identity graph stay yours and movable, data you can point at a warehouse, a suite, or an agent layer without re-resolving every profile. Portability is the hedge that pays off whichever way it lands.
Buy for the architecture roadmap, not today's feature list. A CDP contract typically runs three to five years, so a platform signed now will live straight into 2028. Weigh how a vendor sits against these scenarios: where its data physically lives, how it exposes profiles to agents, and how it governs autonomous action.
This returns to the question underneath the whole category. The CDP was never really a product. It was an answer to the problem of a fragmented customer, and that problem is permanent. The answer keeps changing shape: it was a database, it is becoming a layer, it may become a context service for machines. Whether the three letters survive on a vendor's homepage in 2028 matters far less than whether your organization has done the work they stood for. Resolve the identity. Govern the consent. Agree the definitions. Keep the data yours. Do that, and no scenario can hurt you. Skip it, and no scenario saves you.
Council summary
This post argues that the CDP's function is permanent but its status as a separate category is not, and it lays out three concurrent fates: dissolving into the warehouse, into suites, or into a governed context layer for AI agents. The council verified every figure against primary sources, including the CDP Institute's January 2026 count of 204 vendors, the 7.8 percent versus 1.3 percent composable employment gap, the Rokt mParticle deal, and Gartner's 40 percent agent-adoption forecast. One claim was corrected: the McKinsey research interviewed more than 50 marketing leaders at large companies, not Fortune 500 firms specifically. The takeaway is the strongest part of the piece: stop betting on a winning architecture and do the scenario-proof work instead, resolving identity, governing consent, agreeing definitions, and keeping customer data portable.
Comments