Are Composable CDPs Dethroning the Standalone SaaS Solution?

The CDP Buying Question Has Changed

For about a decade, choosing a customer data platform (CDP) meant picking a product. The decision might have been made on vendor preferences, feature sets, or existing stack compatibility.

That decision tree has been completely uprooted by embedded and composable CDPs.

It doesn’t mean the CDP is dead. But it does look like the standalone platform is losing its dominance to embedded and composable CDPs.

And it means that CDP implementations are as much about data architecture as they are products and features.

Composable vs Embedded: The Two Models Taking Over the Category

A composable CDP separates most CDP capabilities from the database.

Customer data lives in a warehouse or lakehouse. Different tech layered on top provides capabilities like identity resolution, segmentation, reverse ETL, and activation. In some cases, the warehouse provides compute functionality.

It means keeping one copy of customer data and composing a stack of your chosen tools.

An embedded CDP is built natively into the data platform itself.

The same platform or hub that stores data also prepares it for activation platforms. That can include identity matching, segmentation, and behaviour-based automations.

Databricks’ CustomerLake is the recent example that springs to mind. It launched in 2026 with identity, profiles and audiences, segmentation, and – crucially – campaign automation running inside the Databricks lakehouse.

A lot of that is, unsurprisingly, run by AI agents.

Gartner described its arrival as “the start of a platform-layer race that will redefine customer data management and marketing operations.”

Still, there’s a lot of work to do before we can confidently redefine anything.

Two-thirds of marketing leaders (65.7%) say data integration is their main tech stack issue. Many seem to have their hopes pinned on AI solving that problem, but AI agents can’t work with crappy data.

CDPs Aren’t a Two-Horse Race

There is a third choice. Traditional packaged (aka standalone) CDPs haven’t gone away entirely – and a lot of packaged tools ship with warehouse connectivity and real-time features.

There’s also the option to have no dedicated CDP at all. Platforms like Braze have capabilities to ingest and interpret data from a warehouse, enabling you to trigger events in-platform. Some Braze users have a CDP, some don’t.

This all gets confusing. That’s good for vendors. They’re more likely to stretch a platform’s potential just enough to close a deal than under promise and lose out to a competitor.

I don’t say that to scare you, but to make you aware that no solution is inherently better or worse. Different architectures solve different organisational problems.

That’s why, as I discussed in an earlier article, working backwards from your CDP use case is smarter than trying to mould your marketing around the latest shiny features.

At the same time, the CDP market is consolidating. The ‘best-of-breed’ idea is dissolving, because platforms are shipping features that look very similar to competitors. Marketers used to buy the best bits and build a stack. Now we’re going hub again.

Neil Hughes, Marketing Solutions Director

Traditional, Composable, Embedded: A Quick CDP Comparison

Why the Standalone CDP Is Losing Its Claim

CDPs in marketing have two core functions:

1. Pull in customer data from multiple sources to create a unified record
2. Make that customer record available for activation platforms

Embedded CDPs especially are pulling those jobs apart. Unification is moving closer to the data layer.

For enterprises or businesses with mature data infrastructure, that could be CDP capabilities layered on top of a warehouse or lakehouse (Snowflake, BigQuery, Databricks) or the CRM doubling as a de-facto CDP (Salesforce Data Cloud, Microsoft Dynamics).

This makes sense, seeing as the data is already there and governed by platform policies.

For smaller or less mature organisations, CDPs might not be named as such. The capabilities are there, just embedded into engagement platforms (Klaviyo, Zeta, Braze).

A standalone CDP holding its own copy of customer data is duplicated effort. And duplicated cost.

The Warehouse Won. Kind Of.

There’s a claim kicking around the marketing blogosphere that the warehouse “won” the battle to house customer data. That’s almost true – but it’s more nuanced.

Customer data still exists across operational martech. Engagement platforms need data close to the point of interaction, and some use cases require low-latency processing that warehouse architecture doesn’t provide.

That might happen with locally cached data, stored metadata, or engagement platforms with CDP capabilities. In other words, “zero-copy” architecture doesn’t mean zero local storage.

In reality, a lot of platforms are doing a little of everything. The warehouse might be taking on more CDP responsibilities. It hasn’t “won” every function yet.

That’s especially true in the case of a composable CDP which might use bits of a warehouse’s storage or compute functionality while also holding local data.

Engagement platforms geared towards smaller and more agile businesses increasingly manage profiles, events, segmentation, and activation. They become the customer data home almost by default.

The Market Is Voting: Consolidation and Contraction

In six weeks over the winter of 2024–25, three well-known independent CDPs disappeared into larger businesses.

1. ActionIQ was acquired by enterprise AI company Uniphore.
2. Lytics went to Contentstack, a digital experience platform.
3. Rokt, an eCommerce technology business, acquired mParticle for $300 million.

They were the first acquisitions of mid-tier CDPs in years. A couple of years on, with the benefit of hindsight, we can see that it’s just three examples of a category that’s fundamentally changing.

Neil Hughes, Marketing Solutions Director

Gartner expects this trend to accelerate. They predict that 80% of net-new enterprise CDP deployments will be embedded in or composable with data platforms rather than standalone products by 2030.

That’s not necessarily good news in the near term.

For one thing, the gradual transition means that composable CDP vendors need to fence sit, maintaining features for clients with and without zero-copy architecture.

It also introduces roadmap risk for CDP buyers.

A standalone CDP has its own product strategy (hopefully one built around customer needs). An acquired CDP tends to become a component of somebody else’s strategy.

1. ActionIQ’s capabilities now contribute to an enterprise AI proposition.
2. Lytics strengthens a content and digital experience platform.
3. mParticle supports Rokt’s ecommerce proposition.

If a platform that’s critical to your customer data architecture is acquired, whose strategy will its roadmap serve three years from now?

This is one reason pundits predict that 4 in 5 CDP renewals in 2027 will be PAYG or month-to-month, rather than multi-year contracts with standalone platforms.

And yet…

The Standalone CDP Isn’t Dead

A packaged platform remains the sensible choice for organisations that don’t run a warehouse or employ data engineers.

Real-time requirements matter too. Warehouses are built for analysis, not millisecond decisions. That’s exactly why those three independent CDPs sold for big bucks.

If you need to act in the moment, you still need a layer built for it. So don’t go dismantling your legacy CDP just yet.

The same Magic Quadrant report driving the ‘consolidation and contraction’ narrative still lists independents like Tealium, Treasure Data, Optimove and Twilio Segment among the field. They all have strong USPs for the right buyer.

Neil Hughes, Marketing Solutions Director

What This Means for Your Next CDP Decision

If you’re purchasing or renewing a CDP, or reviewing how you manage customer data, don’t start with the vendor shortlist. Define what you need first.

Don’t Automatically Renew a Standalone CDP

Map where your customer data already lives and which systems hold copies. Calculate what those copies cost to create, govern, and keep in sync. If you’ve invested in a warehouse or lakehouse, ask whether you need another persistent customer database.

Match the Model to Maturity

Embedded or composable CDP if you have the data platform and the engineers to run it. Standalone if you need speed and don’t.

Be Honest About Real-Time Requirements

“Real-time” isn’t a fixed measure. Which customer interactions genuinely require decisions in milliseconds or seconds? Which can tolerate minutes or hours? The answer could materially change the architecture you need. Plenty of teams buy speed they never use.

Price In Roadmap Risk

Don’t assess a CDP purely on what the product can do today. Look at who owns it, where it sits within that company’s wider proposition, and whether its strategic direction matches yours. An acquisition isn’t automatically bad news, but it changes the long-term equation.

Get Ahead of Agentic AI Governance

Unified customer data will soon be used by AI agents as well as people and conventional applications. That makes identity protection, permissions, governance, and decision rights more important. Decide how AI should be trusted to access and act on customer data before tools decide for you.

Treat the renewal as an architecture and ownership decision, not a binary decision about the platform.

Neil Hughes, Marketing Solutions Director

Where This Leaves You: Standalone vs Embedded vs Composable CDP

The CDP started as an answer to fragmented customer data. Now its capabilities are spreading into warehouses, lakehouses, engagement platforms, AI infrastructure, and the wider martech stack.

The standalone CDP isn’t dead. But its era of dominance-by-default is ending.

The question, then, isn’t simply which CDP to buy. It’s where your customer record should live, who should own it, and which capabilities need to sit around it.

Getting that right means looking at data, technology, and teams together. Then choosing the architecture that delivers the outcome you actually need, rather than signing a contract because that’s how the problem used to be solved.