How to Choose a Customer Data Platform

Updated September 2026.

A customer data platform promises one thing every marketing and data team wants: a single, trustworthy record of each customer, built from every system that touches them. What most shortlists skip is the question that should come first: whether you need a CDP at all, or whether the gap you're feeling is a reverse ETL problem, an instrumentation problem, or a marketing automation problem wearing a CDP-shaped costume.

This guide starts there, then covers the axis that actually separates modern CDPs, packaged versus composable versus warehouse-native, evaluation criteria, pricing models that don't compare on paper, and a decision framework by buyer profile. How to choose marketing automation software and how to choose product analytics software cover the two tools most often confused with a CDP.

Do you actually need a CDP?

The most useful thing this guide can do is help you conclude you don't need one. Plenty of teams buy a CDP to solve a problem a cheaper, narrower tool already solves.

One product, one database, and marketing automation that already syncs with your CRM usually means a reverse ETL problem: data is already unified in the warehouse, it just needs to get back out. If you don't trust your event data, a CDP on bad instrumentation just activates bad data faster. Fix tracking first. A CDP earns its cost only when identity is genuinely fragmented, anonymous on the website, known in the app, a different record in the point-of-sale system, and nothing already resolves that into one profile.

Land in that last row, keep reading. Land anywhere else, and the narrower tool gets you to value faster and cheaper.

Key Facts:

• CDP Institute tracked 208 active CDP vendors as of its July 2025 industry update, a sign of how fragmented and consolidation-prone this category still is, per CDP.com's industry statistics page.
• The global CDP market is projected to grow from $7.34 billion in 2026 to $14.04 billion by 2031, a 13.8% compound annual growth rate, per MarketsandMarkets.
• Marketing teams that have satisfactorily unified their data are 42% more likely to regularly respond to customers than teams that haven't, per Salesforce's State of Marketing report.
• Enterprise data teams spend 53% of their engineering capacity maintaining and troubleshooting existing data pipelines rather than building new capability, per Fivetran's 2026 enterprise data infrastructure benchmark of 500 senior data and technology leaders.
• As of January 2026, 19 US states have comprehensive consumer privacy laws in effect, with Indiana, Kentucky, and Rhode Island's laws taking effect January 1, 2026 alone, per the IAPP.

What a CDP actually does that marketing automation or a CRM doesn't

A CRM stores records your sales team enters by hand. Marketing automation triggers campaigns off known contacts. Neither resolves an anonymous visitor into the same profile as a known customer, or holds a single record every downstream tool can activate against. That's the CDP's specific job, in four pieces: identity resolution stitches anonymous and known signals (a site cookie, an app device ID, a form-fill email, a loyalty ID) into one merged record; a single profile store holds that record somewhere your other tools can query; audience building lets a marketer define a segment without an engineer writing a custom query each time; activation pushes that audience into the destinations that reach the customer, staying in sync as the profile updates.

If the gap is really just field mapping between a CRM and marketing automation you already own, see marketing automation vs CRM first. It's a shorter, cheaper fix than a full CDP evaluation.

Packaged, composable, or warehouse-native: the real decision

Almost every CDP comparison leads with a feature grid. That's the wrong first filter. The question that determines fit is architectural: does the CDP hold its own copy of your unified data, or read and write directly against the warehouse you already pay for?

A packaged CDP ingests data, resolves identity, and stores the merged profile inside the vendor's own database, a purpose-built activation UI and usually a faster path to a working profile, at the cost of a second copy of customer data to govern. A composable, warehouse-native CDP leaves the unified profile in Snowflake, BigQuery, Redshift, or Databricks, the warehouse you already operate, and layers identity resolution and activation on top without duplicating the data. You avoid a second source of truth, but the modeling a packaged CDP's UI would hide happens in your warehouse instead, as SQL a data engineer owns.

Neither is universally right. The answer follows from what you already have, not which vendor's feature list is longer.

The practical test: no warehouse and no data team today means a packaged CDP gets you to a usable profile faster. A well-modeled warehouse already in place means a packaged CDP usually just duplicates data you already trust. Segment and RudderStack are the two most-compared vendors on opposite sides of this exact split, and the shortlist further down covers the full field across both camps.

What to look for: evaluation criteria

Score each vendor against these before a demo, not during one, since a vendor's own demo is built to hide the columns where it's weak.

The SaaS vendor evaluation scorecard turns this into a shared, scored comparison, and the integration requirements checklist is worth running before finalizing coverage.

Key questions to ask before a demo

1. Where does the merged profile actually live? The vendor's database, your warehouse, or both, and the cost to move it later.
2. What's the real billing unit, and how does it grow? Get your bill at 2x and 5x current volume, not a general "it scales" answer.
3. Who maintains the identity resolution logic after go-live? SQL in your warehouse is your data team's job indefinitely, not a one-time setup.
4. What happens to historical data and audiences if we leave? Ask specifically about exporting resolved profile history, not just raw logs.
5. What's included versus a separate contract? Identity resolution, reverse ETL, and consent management are frequently priced as add-ons after the base platform.

How to avoid software buyers' remorse is worth reading before these calls, since a CDP carries a longer implementation tail than most categories.

Pricing models: the hardest part of this purchase

This is genuinely the hardest part of buying a CDP. The market meters on at least three incompatible bases, monthly tracked users, events, and profiles or records, and a "starting price" tells you almost nothing until you know which one you're billed on.

Monthly tracked users (MTU) charges per unique user who triggered any event that month, regardless of volume. Twilio Segment's Connections: Free is $0/month for 1,000 visitors, Team starts at $120/month for 10,000 visitors, overage runs $12 per 1,000 MTUs from 10,000 to 25,000, $11 from 25,000 to 100,000, $10 above 100,000, per Twilio's pricing page. One split trips up most write-ups: Connections carries these published prices, but the CDP layer itself, Unify and Engage, is contact-sales only.

Events charges per action tracked, so a chatty product costs more than an MTU tool would for the same users. RudderStack: Free covers 250,000 events/month, Growth starts at $265/month for 1 million, saving 15% annually. Jitsu follows the same pattern smaller: 200,000 active events/month free, then $99/month for 2 million with $40 per additional million.

Profiles or records charges per customer record, independent of activity. Salesforce Data 360 prices Profiles at $240 per 1,000/year and Enterprise Profiles at $420 per 1,000/year, both drawing on Flex Credits at $500 per 100,000. Klaviyo meters active profiles: free to 250, a starter paid config around $45/month. Fivetran Activations (formerly Census) uses monthly active rows (MAR): free to 3,500, then a declining usage curve with no flat rate, a published median example running $197 to $202/month around 10,000 to 12,000 MAR.

Normalizing a comparison across metering units

There's no vendor-published exchange rate between MTUs, events, and profiles, since it depends on how your product is instrumented. Pull your own numbers first: distinct monthly users, average events per user, how many become durable profiles. Light per-user instrumentation with a large base favors MTU pricing; heavy instrumentation with a smaller base favors event pricing, which is usually the unit that grows fastest as you ship more features. Segment vs RudderStack works this exact conversion at three volume tiers.

One rule to hold onto through every vendor call: an annual commitment is essentially always cheaper than month to month. Segment and RudderStack both publish annual discounts (up to 20% and 15%). If a quote implies the opposite, the reading is wrong, not the vendor's pricing.

Total cost of ownership

The quoted subscription is rarely what you'll spend in year one. Three costs get left out of most CDP budgets: implementation (mapping sources and validating profiles, vendor project or your own engineering time), the warehouse bill for composable setups (every resolution run and sync bills compute separately), and engineering time nobody budgets for, the biggest blind spot either way. A packaged CDP still needs someone maintaining source mappings; a composable CDP needs a data engineer owning SQL models indefinitely. Per Fivetran's benchmark above, enterprise data teams already spend 53% of engineering capacity maintaining existing pipelines.

The SaaS total cost of ownership guide walks through an honest 24-month model, and the software security checklist for buyers covers the security-review costs a CDP tends to trigger.

Migration and lock-in

Your event schema, the taxonomy your team spent months making consistent, is worth more than most buyers realize until they try to leave a vendor. A packaged CDP's merge rules typically live inside its own product; raw events export easily, but the resolved logic that turned events into profiles often doesn't. A composable CDP has an advantage: the logic is SQL you own, so leaving means losing only the activation UI, not the model.

Ask every vendor what a full export looks like, not just raw data but resolved profile history and the rules behind it. A vague answer is part of the price, paid later in re-implementation time.

CDPs by architecture: a shortlist

A reference shortlist grouped by the packaged-versus-composable split above, not a ranked list. For scored, head-to-head rankings across the full 15-platform field, see the best customer data platforms 2026 roundup.

Packaged CDPs (the vendor stores the unified profile):

Composable / warehouse-native CDPs (the warehouse stays the system of record):

"No published price" isn't "expensive," it's a sales-process choice, so get a written quote before ruling a vendor out on opacity alone. Several of these vendors changed names or owners in the past two years (Census became Fivetran Activations, mParticle joined Rokt, Salesforce Data Cloud became Data 360), so confirm the current entity before an old RFP misleads your shortlist.

How to choose: a decision framework

Match your situation to a direction, not a specific vendor.

How to build a software shortlist covers narrowing a wide field like this one down to a runnable evaluation without losing a month to vendor calls.

Frequently asked questions

Do I need a CDP or is reverse ETL enough?

If your data is already unified in a warehouse and the problem is getting it back out, you need reverse ETL, not a full CDP. Composable tools like Hightouch and RudderStack technically do both, but if identity resolution isn't your actual gap, buy the narrower capability alone.

What's the real difference between packaged and composable CDPs?

A packaged CDP stores its own copy of the unified profile inside the vendor's database. A composable, warehouse-native CDP leaves that profile in your own warehouse and only adds identity resolution and activation on top. The choice depends on whether you have a modeled warehouse and a data team, not which feature list is longer.

How much does a customer data platform actually cost?

It ranges from genuinely free (RudderStack, Jitsu, Fivetran Activations all publish real free tiers) to entirely custom-quoted (Adobe, Tealium, Amperity, mParticle publish no price at all). Vendors meter on at least three incompatible units, so a direct dollar comparison is meaningless until you convert to your own usage pattern.

Can I switch CDP vendors later without starting over?

Partially. Raw event data is usually portable. Harder to move is the resolved identity logic, the merge history behind trusted profiles. That's more portable on a composable tool, where it lives as SQL you own, than on a packaged CDP, where it lives inside the vendor's product.

Is marketing automation the same thing as a CDP?

No. Marketing automation triggers campaigns for contacts you already know. A CDP resolves anonymous and known signals into one profile, then makes it available to every downstream tool, marketing automation included, a distinction covered in more depth earlier in this guide.

Making the call

Start by ruling the purchase out, not in. If a narrower tool closes your gap, buy that instead. In the fragmented-identity case, let your warehouse situation decide packaged versus composable before comparing a feature grid, then price two finalists on your own usage numbers, since the metering unit alone can flip which one is cheaper.

The shortlist above covers both camps across company stages. Narrow to two, run both against your real customer data rather than a demo dataset, and confirm export terms before you sign anything.

Related reading

• How to Choose Marketing Automation Software: the tool most often confused with a CDP, and when you need it instead
• How to Choose Product Analytics Software: a different job than a CDP, covered here in detail
• Marketing Automation vs CRM: untangling overlap before you add a third system
• Segment vs RudderStack: a full worked cost comparison between the two most-compared CDPs
• Best Customer Data Platforms 2026: scored rankings across all 15 platforms in this category
• Software Total Cost of Ownership Guide: building an honest 24-month cost model for any software purchase
• How to Avoid Software Buyers' Remorse: the discipline that catches most CDP-purchase regret before it happens