Part 1 of this series left the data management platform at the height of its run. The DMP was the audience engine of programmatic advertising: it pooled mostly anonymous, third-party, cookie-based data, kept it for short windows, built lookalike audiences, and fed them into the buying machinery around it. For about a decade that was how large advertisers thought about audiences. Rented data. Cookie identifiers. Campaign-shaped thinking.
This part is about what replaced it, and why the replacement was not really a technology upgrade. It was a change of mind. Marketing stopped wanting a pile of rented strangers and started wanting a customer database it actually owned. The customer data platform is the name that change took.
Origin: a category that got named before it had a market
The phrase "customer data platform" did not come from a vendor launch. It came from an analyst noticing a pattern. In 2013, the marketing technology consultant David Raab saw a cluster of new applications that all did something the older tools did not: they assembled a unified customer database from many sources and made it usable by other systems. He gave the pattern a name. That September he published the first industry report on the category, profiling eleven systems. In 2016 he founded the CDP Institute, the same year CDPs first appeared on Gartner's Hype Cycle.
The definition the Institute settled on is worth holding onto, because it is precise and most marketing chatter is not. A CDP is packaged software that builds a persistent, unified customer database accessible to other systems. Three parts, each load-bearing. Packaged software, so a marketing team can run it without a data-engineering project for every change. Persistent and unified, so the customer record survives across sessions and devices instead of expiring with a cookie. Accessible to other systems, so the profile is not a dead end but a source other tools can read and act on.
Set that against the DMP and the contrast is sharp. A DMP held anonymous identifiers for a short window and pushed them at ad platforms. A CDP holds named, person-level data, including the personally identifiable kind a DMP deliberately avoided, keeps it for the customer's lifetime, and serves it to any channel a company owns. The DMP thought in campaigns. The CDP thinks in relationships.
The clearest single story of the category is Segment. Four founders started it in 2011, and the first idea was not a data platform at all. It was a classroom tool that let students flag confusing moments in a lecture. It failed. A web analytics product followed. That failed too. With the money nearly gone in December 2012, the team open-sourced the small JavaScript tracking library they had built along the way, posted it to GitHub, and it spread fast among developers who were tired of wiring up a new tracking snippet for every analytics tool. That library, the plumbing nobody set out to build, became the foundation of the company. Segment grew into the vendor most people credit with making the CDP a category buyers asked for by name. In 2020, Twilio acquired it for about 3.2 billion dollars. A side project had become one of the most valuable assets in martech.
Present: what broke the DMP, and what the CDP absorbed
The CDP did not win on charm. The DMP's foundation was pulled out from under it, and four forces did the pulling.
The first was browser policy. Safari and Firefox blocked third-party cookies years ago. Chrome spent years promising to follow and reshaped its plans more than once, but the direction never reversed. The signal a DMP ran on was decaying browser by browser.
The second was regulation. The General Data Protection Regulation in Europe and the California Consumer Privacy Act in the United States, followed by a wave of similar laws, put real constraints on collecting and sharing third-party data. Rented audience data became a legal exposure, not just a line item.
The third was Apple. In April 2021, iOS 14.5 introduced App Tracking Transparency, which forced every app to ask, in a blunt system prompt, before tracking a user across other companies' apps and sites. Most people said no. The economists who studied the change for an FTC filing found the share of trackable Apple traffic in the United States fell from 73 percent to 18 percent. A 55 percentage point collapse in cross-app identifiers, by regulator-cited research, decided by a one-line dialog box.
The fourth force was not a restriction at all. It was a strategy. As third-party signal decayed, first-party data, the information a company collects directly from its own customers with their consent, became the obvious thing to build on. It was durable, it was legally cleaner, and crucially it was owned. A company could not be cut off from it by a browser update.
Put those four together and the DMP's core asset, rented anonymous identifiers, lost most of its value at once. The end was not gradual. Part 1 covered the closures in detail; the short version is that two of the biggest DMPs were shut down within months of each other in 2024, one of them after advertising revenue had fallen from a multi-billion-dollar peak to a few hundred million.
Here is the part that gets missed. The CDP did not just sit next to the DMP's grave. It picked up the DMP's actual jobs and kept them. Audience building, lookalike modeling, suppression of people you do not want to pay to reach again: all of that survived the DMP and now runs inside the CDP, except aimed at known first-party profiles instead of anonymous cookie pools. Then the CDP added the things a DMP never had. Persistent identity that ties a person's behavior together over years. A unified profile any owned channel can read. Activation across email, app, site and service, not only into ad auctions.
That last point is what the category's whole pitch rests on, and it deserves a plain explanation. The promise is unification and identity resolution. In most companies, a single customer is scattered: an email address in the marketing tool, a purchase history in the commerce platform, a support ticket in the service desk, an app login on a phone, an anonymous web session from before they ever signed up. Identity resolution is the work of recognizing that all of those fragments are one person and merging them into one record, often called a golden record or a Customer 360. It runs two ways. Deterministic matching joins records on exact shared identifiers, an email or a phone number or a customer ID, and is highly reliable. Probabilistic matching estimates, from signals like device and IP and behavior, that two records are probably the same person when no hard identifier links them. Most CDPs do both and store the result as an identity graph: a living map of every identifier and device that belongs to one customer. Get that right and a marketer can finally segment and message a whole person instead of a handful of disconnected sessions.
The promise was strong enough to build an industry. The CDP Institute counted 208 vendors in its July 2025 update, with employment across the sector at 18,361 people and total funding at about 9.4 billion dollars. Third-party market sizings vary so widely they are almost a warning in themselves, from roughly 4 billion dollars to around 10 billion for 2026 depending on which firm you ask, because every analyst draws the category's boundary in a different place. The looseness of the number is itself a fact about the category: nobody fully agrees on what counts.
The names worth knowing split into a few groups. The packaged enterprise platforms include Salesforce Data 360 (formerly Data Cloud), Adobe Real-Time CDP, Oracle Unity, Treasure Data, Tealium and Twilio Segment. The composable or warehouse-native challengers, which take a different architecture this series returns to in Part 3, include Hightouch and the reverse-ETL vendors. Amperity, BlueConic, Microsoft Dynamics 365 Customer Insights and SAP Emarsys fill out the field. In the 2026 Gartner Magic Quadrant for the category, Salesforce was joined in the Leaders quadrant by Hightouch, Oracle and Uniphore, while four vendors, ActionIQ, Redpoint Global, mParticle and Zeta Global, dropped off the quadrant entirely for failing tougher inclusion criteria. A maturing market raises the bar and thins the field.
Future and impact: the gaps the boom papered over
A category this hyped accumulates expectations it cannot meet, and the honest picture of the CDP in 2026 is that several of its promises came with asterisks.
The first gap: a CDP does not fix bad data. Many buyers assumed that pouring messy records into a CDP would somehow produce clean ones. It does not. Identity resolution can merge records, but if the underlying data is wrong, duplicated or stale, the unified profile is a tidy presentation of wrong, duplicated, stale data. Garbage in, garbage out survived the upgrade. The work of cleaning and governing data still has to happen, and it has to happen before the CDP, not inside it.
The second gap: real-time is uneven, and the word is abused. Vendors call almost anything real-time, and practitioners have stopped trusting the label. A MarTech guide to checking the claim is blunt about the pattern: data may arrive quickly, but enrichment runs on a schedule, segments can rebuild overnight, and re-qualifying a person into an existing segment can take hours. A CDP can update a profile in well under a second for one job and lag by hours for another. "Real-time CDP" describes an aspiration more than a guarantee, and the only fix is to ask, per use case, exactly which step is fast and which is batched.
The third gap is the one that reshapes the category's future: the CDP competes with the data warehouse. The cloud warehouses, Snowflake, BigQuery, Databricks, Redshift, have become the place where many companies already keep their customer data. That raises an awkward question. If the warehouse already holds the unified data, why pay a CDP to hold a second copy? The provocation has a famous form: a widely shared post arguing "Snowflake is killing the CDP". The honest answer is that a warehouse is storage and compute, not marketing logic. It does not, on its own, do identity resolution, audience management, consent handling or activation to channels. But the question forced a real architectural response, the composable approach that leaves data in the warehouse and adds only the missing layer, which Part 3 takes up directly.
And the verdict from buyers has been mixed. eMarketer, citing the CDP Institute's 2024 member survey, reports that only 64 percent of deployed CDPs deliver significant value, a figure that has slipped over time. The same piece relays a McKinsey finding that is starker still: of roughly 50 senior marketing leaders at Fortune 500 companies interviewed, not one could clearly measure the return on their martech investment. Hightouch, a vendor with an interest in the point but citing it openly, has called traditional CDPs marketing's most disappointing tool. The recurring causes are not mysterious: poor integration, underskilled teams, and platforms bought before anyone defined what they were for.
The consolidation tells the same story in deal terms. mParticle, an independent CDP that had raised about 272 million dollars in venture funding, sold to the commerce company Rokt for 300 million in early 2025, barely above the capital invested. In the span of a little over a month, Uniphore acquired ActionIQ and Contentstack acquired Lytics. Months later, Fivetran agreed to acquire Census. Standalone CDPs are increasingly being folded into larger platforms rather than thriving alone.
None of that means the CDP failed. It means the CDP did the genuinely hard and genuinely useful thing, building a persistent, owned, unified customer profile, and the market is now arguing about where that capability should live. The shift it represents is real and not reversible: from rented data to owned data, from anonymous identifiers to known people, from campaigns to relationships. That argument about where the capability belongs leads straight into Part 3, which is about the next force pulling on the customer profile: the agent layer, and what it means when the main thing reading the profile is no longer a person at all.
Council summary
This post argues that the CDP replaced the DMP not as a technology upgrade but as a change of intent: marketing chose owned, named, first-party profiles over rented anonymous identifiers. It traces the category from David Raab naming it in 2013, through Segment's accidental rise and 3.2 billion dollar sale to Twilio, to a 2026 market of 208 vendors now consolidating and arguing with the data warehouse over where the customer profile belongs. The council verified every figure against primary sources, including the CDP Institute's July 2025 update, the Skiera FTC filing on App Tracking Transparency, and the 2026 Gartner Magic Quadrant, and corrected two attributions: the 64 percent value figure belongs to the CDP Institute's 2024 survey, and the unmeasured-ROI finding to McKinsey's Fortune 500 interviews. The takeaway: the CDP did the hard work of building a persistent owned profile, but a warehouse copy, abused real-time claims, and dirty source data mean buyers should ask what each capability actually does before paying for it.
Comments