Executive Summary
Choose a customer data platform by where the customer profile will live: in the vendor’s store, in your data warehouse, or in both. Then test identity matching on your own records, and price the platform in the unit the vendor bills.
A customer data platform (CDP) collects data about customers from websites, apps and business systems, decides which records belong to the same person, keeps a profile for each customer, and sends profiles and audiences to the marketing, service and analytics tools that use them. The CDP Institute defines a customer data platform as software that creates and maintains a persistent, unified customer record that is accessible to other systems.
Three decisions separate the platforms. The first is where the profile lives. The CDP Institute describes three CDP architectures: own-store (packaged) CDPs, which keep their own copy of customer data; warehouse-native (composable) CDPs, which work directly on customer data held in a company’s data warehouse; and hybrid architectures that do both. The second is how identity is matched: exact matches on identifiers such as an email address, or statistical matches that reach more records with less certainty. The third is the pricing unit: tracked users, profiles, credits, syncs or a vendor’s own consumption unit.
This guide explains what a CDP is and is not, where the profile can live, how to test identity resolution, six approaches to buying, the eight vendors that represent them, and what each publishes about price. Two of the eight publish list prices.
What Is a Customer Data Platform, and How Is It Different from a CRM or a DMP?
The CDP Institute defines a customer data platform as software that creates and maintains a persistent, unified customer record that is accessible to other systems. The CDP Institute states that gathering customer data from source systems, linking it to customer identities and storing the results in a database available to external systems is the minimum set of functions required to meet the definition of a CDP. Most products add segmentation, activation and predictive features on top of that minimum.
The CDP Institute states that a DMP builds mostly anonymous audience segments for advertising, largely from third-party data kept for limited periods, while a CDP works mainly with a company’s own data about identified people. The CDP Institute states that an integration platform moves data between systems without keeping a permanent database, while a CDP maintains a persistent customer database. A CRM holds the sales and service records that staff work in. A CDP takes those records as one of its sources and combines them with web, app, purchase and support data.
A CDP decides which records belong to the same customer. It does not decide what counts as a customer: a household or an individual, a business account or the people in it, a buyer from five years ago or only an active one. Sales, finance, marketing and support often answer differently. Agree the definition before the project starts. The customer data platform was sold as a fix for something it cannot fix explains why projects stall on it.
Every profile a CDP builds is also a record that privacy law reaches. California’s Attorney General states that you can request that businesses delete personal information they collected from you and tell their service providers to do the same, subject to certain exceptions. California’s Attorney General states that you may request that businesses stop selling or sharing your personal information, including via a user-enabled global privacy control. California’s Attorney General states that sharing refers specifically to sharing for cross-context behavioral advertising. A CDP that has sent a profile to twenty tools has to carry a deletion or an opt-out to all twenty. Ask each vendor to show a deletion request reaching every connected destination.
Packaged, Composable or Hybrid: Where Should the Customer Profile Live?
Start with one question for your data team: is the warehouse already where customer data is cleaned, joined and governed? If it is, a composable or hybrid CDP can read from it. If it is not, a packaged CDP brings its own store, and building a customer table in the warehouse first can take longer than the CDP project itself.
A packaged CDP collects events, resolves identity and stores profiles in its own platform. Twilio states that Twilio Segment works with the sources you already have, and a warehouse is optional. Twilio states that the Customer Data Platform bundles Connections with Unify and Engage, adding identity-resolved profiles, AI-powered audiences, and journey orchestration. The trade-off is a second copy of customer data that has to stay in step with the warehouse and with every deletion request.
A composable CDP leaves the data in the warehouse and runs identity, audiences and syncs from there. Hightouch states that identity graphs and unified records stay in your data warehouse, and that Hightouch never stores your data. Hightouch states that all of its products are composable and can be purchased separately. Your data team then owns the identity rules, the data models and the warehouse compute that a packaged CDP would carry.
Hybrid products do both: real-time profiles in the vendor’s platform, and audiences built on warehouse data. Rokt states that Rokt mParticle combines the speed of a real-time CDP with the flexibility of a composable, warehouse-native architecture. Treasure AI offers a Composable CDP on your own Snowflake, Databricks, or BigQuery warehouse, or a Complete CDP on managed infrastructure. Tealium states that it streams consented events into your warehouse and activates modeled data back out, and that your warehouse stays the source of truth for modeled customer data.
The marketing and CRM suites now read warehouse data in place as well. Adobe states that customers can activate customer data directly from their existing warehouse without the need to move or copy underlying datasets. Salesforce states that Zero-Copy integrations in Data 360 connect directly to platforms like Snowflake and Databricks with no duplication. For every product, ask what is copied, what is queried in place, and who pays for the query compute.
| Packaged CDP | Composable CDP | Hybrid CDP | CDP inside a marketing or CRM suite | |
|---|---|---|---|---|
| Where profiles are stored | In the vendor’s platform | In your data warehouse | Real-time profiles in the vendor’s platform; other data in your warehouse | In the suite’s data platform, with warehouse data read in place |
| Who maintains identity rules | Your team, in the vendor’s tools | Your data team, on warehouse tables | Your team, in the vendor’s tools, running in either place | Your team, in the suite’s tools |
| Real-time profiles | Included | Depends on the product’s event and streaming features | Included | Included |
| Typical pricing unit | Tracked users or profiles | Syncs or usage | Profiles, events or consumption units | Profiles or credits |
| Fits when | The warehouse is not yet the customer record | The warehouse already holds clean, governed customer data | You need in-session personalization and warehouse audiences | Most activation runs in that vendor’s applications |
How do you evaluate Customer Data Platforms (CDP)?
Identity resolution decides whether two records are the same person. Exact (deterministic) matching joins records that share an identifier such as an email address, phone number or loyalty number. Statistical (probabilistic) matching links records that share patterns, which reaches more people with less certainty. Most vendors offer both and let you choose by use. Hightouch states that transactional emails work best with high-confidence deterministic matching, and ad targeting with high-reach probabilistic matching. Treasure AI states that when deterministic and probabilistic signals conflict, deterministic always wins on the action layer. Rokt states that mParticle’s Probabilistic Identity Resolution uses the Rokt Identity Graph to connect anonymous activity to known customers already in your workspace.
Test identity resolution on your own records. Give each finalist the same sample, including duplicates, shared devices and offline purchases, with an answer key your team has labeled. Count false merges, where two people become one, and missed matches, where one person stays two. Amperity states that it connects customer records across online and offline sources using rules and AI-assisted matching. Amperity states that you can inspect matching logic and review how profiles change as new data is processed.
Data quality starts at collection. A tracking plan names every event and property your sites and apps should send, and the CDP blocks or quarantines what does not conform. Rokt states that mParticle enforces quality at ingestion: you define your schema, enforce it at ingestion, and version plans as tracking evolves. Ask to see what happens to an event that breaks the plan, and who is told.
Activation sends audiences and profile fields to ad platforms, email and messaging tools, CRMs and service desks. List the destinations you use today, check that each one is supported, and measure how long an audience change takes to reach it.
Consent has to travel with the data. Tealium states that tags that require consent do not fire without it, with native CMP integrations for OneTrust, Didomi, and Usercentrics. Confirm that an opt-out recorded in one channel stops activation in every other.
Most CDPs now offer predictive scores and audiences built from a plain-language request. Ask where the model runs, which data it reads, and whether a marketer can see why a customer is in an AI-built audience.
| Capability | What it does | What to test |
|---|---|---|
| Data collection | SDKs and APIs that capture web, app and server events | Whether events are validated on arrival, and what happens to those that fail |
| Identity resolution | Matches records to people, exactly or statistically | False merges and missed matches on a labeled sample of your records |
| Profiles | One record per customer, updated as events arrive | How fast a profile reflects a purchase or a consent change |
| Audience building | Audiences built from profile, event and warehouse data | Whether marketers can build them without SQL |
| Activation | Sends audiences and fields to other tools | Coverage of your own destinations, and the delay before a change arrives |
| Consent and privacy | Records consent and carries deletions downstream | A deletion request traced to every destination |
| Predictive and AI features | Scores, recommendations and audiences from a prompt | Where models run, and whether results can be explained |
Vendors in Customer Data Platforms (CDP)
Representative vendors include Adobe Real-Time CDP, Amperity, Hightouch, mParticle (Rokt), Salesforce Data Cloud / Data 360, Segment / Twilio, Tealium and Treasure Data (Treasure AI). The list is alphabetical, is not a ranking, and is not for sale. How we choose vendors
The approaches below follow two questions: where does the profile live, and what does the vendor sell around it? Two approaches have two vendors each. Ask every vendor to name in the order form the products, editions and add-ons the quote includes.
One approach starts from event collection and sells the CDP as a bundle on top of it. Twilio states that the Customer Data Platform bundles Connections with Unify and Engage, adding identity-resolved profiles, AI-powered audiences, and journey orchestration. Twilio states that Twilio Segment works with the sources you already have, and a warehouse is optional. Its prices are in the next section.
Hybrid CDPs keep real-time profiles in their own platform and also work on warehouse data. Rokt lists “Rokt merges with mParticle” under 2025. Rokt describes Rokt mParticle as a real-time customer data platform (CDP) that helps partners unify fragmented customer data, activate audiences across channels, and deliver more personalized customer experiences. Treasure AI states that before it was Treasure AI, it was Treasure Data. Treasure AI states that customer data is always persisted in your warehouse, never persistently stored in its infrastructure, and that during email, SMS and push activation passes through Treasure AI transiently for delivery, then is discarded.
Another approach starts from consented data collection across web, mobile and server. Tealium states that it collects customer data from every source, resolves identity in real time, enforces consent through every activation, and routes data to the model, channel, warehouse, or agent that needs it. Tealium states that segments can be built directly on Snowflake, Databricks, BigQuery, or Redshift and activated without loading a second copy. Tealium states that it reaches 1,300+ destinations across advertising, CRM, personalization, cloud, service, and legacy systems through pre-built connectors.
The marketing and CRM suites sell a CDP that feeds their own applications. Adobe states that Real-Time CDP, built on Adobe Experience Platform, lets you create high-value audience segments for real-time B2C and B2B use cases across the entire marketing funnel. Adobe states that Real-Time CDP comes in B2C, B2B and B2P editions of the same product, each in a Prime and an Ultimate package. Adobe states that Real-Time CDP unifies data and builds and activates audiences, Journey Optimizer orchestrates the multi-step journeys and messages those audiences receive, and Customer Journey Analytics analyzes how those journeys performed. Salesforce states that Salesforce Data 360 is the official new name for what was previously called Data Cloud.
One approach treats identity resolution as the product and runs it in the customer’s own lakehouse. Amperity states that supported batch workloads, Identity Resolution, domain tables, database generation, queries, and segmentation execute in your compute, while real time and activation stay Amperity-managed. Its matching is described in the previous section.
Composable CDPs activate data where it already sits in the warehouse, as described above. Reverse ETL, which syncs warehouse tables into business applications, is the core of this approach. Census (Fivetran) also takes this approach: Fivetran describes its agreement to acquire Census (now Fivetran Activations). Fivetran states that Census supports single second latency for streaming use cases with Live Syncs and near real-time syncs for all others. Fivetran states that Audience Hub enables marketing teams to build target audiences without SQL knowledge.
| Vendor | Approach | Where it fits |
|---|---|---|
| Adobe Real-Time CDP | CDP inside a marketing or CRM suite | B2C, B2B and B2P editions on Adobe Experience Platform |
| Salesforce Data Cloud / Data 360 | CDP inside a marketing or CRM suite | Flex Credits or per-profile pricing; zero-copy warehouse access |
| Hightouch | Composable CDP on the warehouse | Data stays in your warehouse; usage-based pricing with no MTU limits |
| Tealium | Consented collection with real-time profiles | Consent enforced at collection; segments built on the warehouse |
| Segment / Twilio | Event pipeline with a CDP bundle | Free plan to 1,000 visitors; Team from $120/month; CDP plans priced by MTU |
| mParticle (Rokt) | Hybrid CDP: real-time profiles and warehouse data | Merged with Rokt in 2025; real-time and warehouse-native in one platform |
| Treasure Data (Treasure AI) | Hybrid CDP: real-time profiles and warehouse data | Composable or Complete CDP; priced by profiles or behavioral events |
| Amperity | Identity resolution in your own lakehouse | Identity resolution runs in your compute; usage measured in Amps |
Eight vendors across six approaches; mParticle and Treasure Data share one, and Adobe and Salesforce share another. The approaches were written before any vendor was chosen, and none of these rows is for sale. Census (Fivetran) is named in the text above. Event-collection tools, customer engagement platforms and other CDPs also compete for this budget, and any CDP vendor can make its own case in the Spotlight below.
What Does a CDP Cost, and Which Vendors Publish Prices?
CDP vendors price in five different units: tracked users, profiles, credits, syncs and their own consumption units. The unit decides what you pay as traffic, customers and use grow. Convert every quote to your own volumes before you compare them.
Twilio states that Free covers 1,000 visitors at $0, Team starts at $120/month for 10,000, and Business is custom volume. Twilio lists each additional 1,000 monthly tracked users at $12 in the 10K - 25K range and $11 in the 25K - 100K range. Twilio also states that Team starts at $120/month for 10,000 visitors, then adds $10 per 1,000 visitors beyond that. The plan table and the FAQ give different rates above 10,000 visitors, so get the rate in writing. Twilio defines an MTU, or monthly tracked user, as any unique user Twilio Segment collects data on in a billing month, identified or anonymous. Anonymous visitors count, so a high-traffic site with few signed-in customers pays for its traffic.
A request for mParticle’s pricing page (mparticle.com/pricing) is answered by Rokt’s mParticle platform page (rokt.com/rokt-mparticle/platform). That page lists no prices. Ask for a quote in a unit you can set beside tracked users, such as monthly users or events at your current volume, and ask what happens to the price at twice that volume.
Salesforce lists Flex Credits at $500 per 100k Flex Credits, Profiles at $240 per 1k profiles/year, and Enterprise Profiles at $420 per 1k profiles/year. Salesforce states that bringing data into Data 360 is free, and that customers pay when they use it to drive results. Salesforce states that Salesforce customers can provision Data 360 for free with a limited amount of storage and consumption credits. Adobe states that a Real-Time CDP license includes the platform services it depends on rather than being a separate purchase.
Treasure AI states that it prices by customer profiles or behavioral events, not by query volume or processing power. Amperity states that it measures platform usage in Amps. Amperity states that activities such as processing customer records, building segments, running campaigns, or activating insights are calculated in Amps. Hightouch states that it charges based on actual usage, with no MTU (Monthly Tracked Users) limits and no caps on sources, destinations, or user seats. Hightouch’s documentation states that self-serve pricing is based on the number of active syncs: 2 active syncs per month on the Free plan and 10 on the Self-serve plan.
Warehouse compute is a separate cost. Composable and hybrid products run queries on your warehouse, and the warehouse bill grows with every audience, sync and refresh. Ask each vendor to estimate the compute its workload will use each month.
| Pricing unit | You are charged for | Grows with | Where it goes wrong |
|---|---|---|---|
| Monthly tracked users (MTUs) | Each identified or anonymous user seen in a month | Traffic, including anonymous visitors | A traffic spike or a wave of bots counted as users |
| Profiles | Each unified customer profile per year | Customer count | Duplicate and inactive profiles kept and billed |
| Credits or consumption units | Actions such as processing, segmenting, querying and activating | Use | Real-time and AI features using credits faster than planned |
| Profiles and behavioral events | Profiles plus the events attached to them | Customers and their activity | Event volume growing faster than the customer count |
| Active syncs or usage | Syncs or operations that run each month | Audiences and destinations | Many small syncs created for one-off campaigns |
| Warehouse compute | Queries the CDP runs in your warehouse | Audiences, syncs and refresh frequency | Billed by the cloud provider, outside the CDP quote |
How long does implementation take for Customer Data Platforms (CDP)?
A CDP rollout is mostly data work: the tracking plan, identity rules and consent. Start with one use case that has a measurable result, such as a suppression list or an abandoned-cart audience, and add sources after it works.
Pick two or three use cases and the data each needs. Write the tracking plan, list the identifiers each source carries, and decide where consent is recorded and how it reaches each destination.
Configure exact matching first, then add statistical matching where reach matters more than certainty. Check the results against a labeled sample and write down the merge rules.
Send the first audiences to one or two destinations. Compare audience counts with your own SQL, measure the delay before changes arrive, and confirm that suppression and opt-outs work.
Add sources and destinations, set alerts on tracking-plan violations, test deletion requests end to end, and compare actual use with the pricing unit before renewal.
What should you ask vendors about Customer Data Platforms (CDP)?
Send these questions with your proof-of-concept data and ask for written answers.
-
Does your data warehouse already hold a clean, governed customer table?Yes Shortlist composable and hybrid products, and price warehouse compute with each.No Shortlist packaged and hybrid products that store profiles themselves.
-
Do you need to personalize within the same session?Yes Require real-time profiles and test the delay from event to audience.No Audiences refreshed hourly or daily may be enough, which widens the field.
-
Does most of your marketing run in one vendor’s suite?Yes Compare that suite’s CDP with independent CDPs that connect to it.No Weight destination coverage and identity matching over suite integration.
Customer Data Platform (CDP) RFP package: 128 questions, 10 demo scenarios and a vendor scorecard
RFP packages are a separate paid product. Vendors do not pay to appear in buyer guides.