Four tests do most of the work: run the vendor's identity resolution against a sample of your own data; have your own marketer build an audience in the product unaided; get the split of generally available, roadmap and third-party capability in writing; and model the cost at two and five times your current volume. The full evaluation is eleven questions across five stages, from before you shortlist to the week you sign — those four plus the ones buyers say they only thought to ask once the leverage was gone. For what happens when these go unasked, the companion read is what CDP buyers wish they'd known.
Where the real gate is
A formal RFP is not the constant — whether one happens tracks company size and deal size more than it tracks the category, and several buyers say plainly that they don't mandate one. What is closer to universal is hands-on validation before signature, with separate demo rounds for business and technical stakeholders. The form varies more than the word suggests — a full production-style proof-of-concept is not always practical for a platform this heavy to stand up, and some buyers run one only after the shortlist exists or substitute a controlled demo on a small segment. Treat the questions below as a script you run in whichever format you get, not a document you send.
Decides whether you should be buying this class of tool at all. No later stage can recover a wrong answer here.
Data foundation sequencing
Data residency and portability
Fit, stated honestly
The real gate, whatever form it takes — a full proof-of-concept, a shortlist trial, or a controlled demo on a small segment. Two capabilities, both testable in an afternoon, and both routinely taken on faith instead.
Identity resolution
Marketer self-service
Where the demo gets separated from the product. Both questions here are about getting something in writing while you still have leverage.
Available today vs. roadmap vs. partner
Implementation reality
Buyers who like the product still leave over price. The quote fits today's volume; the contract has to fit year two.
The cost curve
What meters, and what doesn't
Two questions that cost nothing to ask now and are expensive to raise later.
Who you actually get
The exit
Telling a real answer from a rehearsed one
Buyers rarely describe a vendor as dishonest. What they describe is a vendor answering a different question than the one asked. These are the signals they name.
Reads as credible
- Engages with the phase you are in, rather than pitching the platform
- Category depth — talks about your industry's data shape unprompted
- An AI roadmap described as what ships when, not what is possible
- Will name what it is not good at, and which customers resemble you
- Responds to a non-standard requirement instead of routing around it
Worth a second round
- Pivots to capability when you asked about sequencing
- Feature breadth pitched before your mission is understood
- Upsells a new module rather than closing the gap you raised
- "Out of the box" with no configuration estimate
- The team in the room is not the team on the account
Want this read against your own stack?
Get my read →Who is in the room changes what gets asked
Two shapes recur, and they fail differently. A named committee scoring against a written framework tests broadly and still misses usability — the people scoring rarely operate the tool daily. A lean team of two tests deeply on what it cares about and is exposed on whatever nobody owns, usually portability and the cost curve. Watch also for procurement run on a cost threshold rather than by category: it sets diligence by price tag rather than by consequence.
Worth settling internally before you ask a vendor anything: which of the eleven questions above has an owner, and which currently has none.
The honest caveat
Buyers almost never attribute their outcome to their own evaluation process — the unhappy ones describe the product, not the diligence. So treat this as a pattern rather than a promise: the buyers reporting the fewest surprises are disproportionately the ones who tested rather than accepted, and who wrote down what the platform was for before seeing what it could do. That is a correlation drawn from how they describe their processes, not a claim that good evaluation guarantees a good outcome.
Related reading
This page is the procurement moment — the tests to run while you still have leverage. For what happens when they go unrun, what CDP buyers wish they'd known is the same corpus in hindsight, with the buyer stories behind these questions. Then what buyers actually pay for a CDP for the cost picture, and real CDP implementation timelines for the schedule. Weighing specific platforms: Tealium vs. Amperity and Hightouch vs. Segment.
Common questions
What should I ask a CDP vendor before signing?
Four tests do most of the work: run their identity resolution against a sample of your own data and require an explanation of any gap against your internal numbers; have your own marketer build and activate an audience unaided; get the split of generally available, roadmap and third-party capability in writing; and model the cost at two and five times your current volume with every metered line named. The buyers reporting the fewest surprises are the ones who tested rather than took answers on faith.
Do I need a formal RFP to buy a CDP?
Not necessarily — many buyers in our interviews don't run one. Whether an RFP happens tracks company size and deal size more than the category. What is closer to universal is hands-on validation before signature, plus separate demo rounds for business and technical stakeholders — though the form varies, and buyers report that a full production-style proof-of-concept is not always practical for a CDP. In practice the validation is the real gate, not the paperwork.
What should a CDP proof-of-concept actually test?
Identity resolution and usability, whether you get a full proof-of-concept or a controlled demo on a small segment. On identity, the vendor should run against your file, and your own internal match rate is the bar — not their benchmark. On usability, the test that separates a marketing platform from an engineering one is whether a non-technical marketer can build and activate an audience unaided, and repeat it a week later without notes.
What questions do CDP buyers wish they had asked?
Two recur. What has to be true about your own data before the platform works — buyers report vendors steering to capability when the honest answer was sequencing. And data portability: only a minority raise lock-in during evaluation, and the rest meet it afterwards in mandatory replication or multi-year agreements they would leave if they could.
Get this research made for your stack
Every one of these questions matters more or less depending on your architecture, your team, and what you're replacing. Do a 15-minute interview about your stack and get the shortlist that applies to you — plus what buyers running your setup asked, and what they wish they'd asked.
Get my personalized read — 15-min interviewNo password · your interview is anonymized before it ever informs a page like this one.