Most CDP RFP processes fail before the first vendor response arrives. Procurement teams send questionnaires full of feature checkboxes, vendors return documents with every box ticked, and the buying team is left trying to read between lines that were written to obscure rather than inform. If you're running a CDP evaluation, the quality of your questions determines the quality of your decision. These CDP RFP questions to ask vendors are built around one principle: make it hard to give a vague answer.
This guide walks through five evaluation dimensions, with specific questions for each. It also covers what a strong answer looks like versus a deflection, so you can score responses with some objectivity.
Why Most CDP RFPs Produce the Wrong Shortlist
The standard CDP RFP asks about integrations, support SLAs, and pricing tiers. Those things matter. But they don't tell you whether the platform will work with the data architecture you already have, whether the vendor's "AI features" are productized or in private beta for one customer, or whether you'll need to rip out your data warehouse to get the product to function.
The CDP market has also changed substantially. Several years ago, the dominant model was a vendor-managed customer database that ingested your data and became the system of record. Today, a growing segment of enterprise buyers wants to keep data in their own cloud warehouse — Amazon Redshift, Google BigQuery, Snowflake, Databricks — and run CDP functionality on top of that data without copying it into a vendor silo.
These two architectural models produce very different answers to the same RFP question. A vendor that manages its own proprietary database will answer "how do you handle data residency?" differently than one built on a Composable CDP model. Neither answer is automatically right for every buyer. But you need to know which type of architecture you're evaluating so you can interpret the response correctly.
With that context established, here are the questions worth asking.
Architecture and Data Ownership Questions
These questions expose the fundamental infrastructure model and where your data actually lives.
- 1. Where does our customer data reside after ingestion — in your infrastructure, ours, or both?
A strong answer describes the storage model precisely. The vendor should be able to tell you whether data is copied into a proprietary database, kept in your warehouse via a zero-copy approach, or split across both. A deflection sounds like: "We offer flexible deployment options depending on your needs."
- 2. If we terminate our contract, how do we retrieve our data, in what format, and within what timeframe?
This is a data portability question with real operational stakes. Strong answers give you a specific file format (CSV, Parquet, JSON), a defined export process, and a contractual timeframe — typically 30 to 90 days. Vague answers that reference "our offboarding process" without specifics are a flag.
- 3. Can our data science and analytics teams query the same customer data the CDP uses, without a separate data export?
This question gets at whether the CDP creates a data silo. If the vendor hosts your data in its own system, your BI tools and data science workflows typically need a separate export or API call to access it. A warehouse-native or composable approach lets your existing tools query the same underlying data directly.
- 4. How do you handle schema changes in our source systems?
Production databases change. Tables get renamed, columns get added, data types shift. Ask the vendor to walk you through what happens to existing segments, syncs, and audience definitions when your upstream schema changes. This is a practical stress test of system resilience.
Identity Resolution Questions
Identity resolution is one of the most technically complex functions a CDP performs, and it's also one of the most frequently oversold.- 5. Walk us through your identity graph model — deterministic, probabilistic, or both?
Deterministic matching links records via known identifiers (email, phone, loyalty ID). Probabilistic matching uses statistical inference when deterministic signals are absent. Both have use cases. Ask the vendor to explain which method they use by default, how they handle conflicts, and how they surface match confidence scores.
- 6. Can we define and override identity merge rules ourselves, or is matching handled entirely by your system?
Some vendors offer configurable merge rules. Others apply a fixed proprietary algorithm you can't inspect or adjust. If your business has specific compliance requirements or unusual data structures — multiple accounts per household, B2B account hierarchies — configurability matters significantly.
- 7. How do you handle identity across anonymous, known, and offline profiles?
This question tests breadth. A strong answer describes how the system stitches together web sessions, CRM records, point-of-sale transactions, and call center data into a unified profile. Ask for a specific example using a customer journey with at least three touchpoints.
- 8. What happens to historical event data when two profiles merge?
Profile merges should carry historical events forward, not discard them. Ask whether a merged profile retains the full behavioral history of both constituent profiles, or only data from the surviving record.
Segmentation and Activation Questions
This dimension tests whether marketers can actually use the platform without filing tickets to the data team every time they need a new audience.
- 9. Show us how a non-technical marketer builds a segment using data that lives in our warehouse.
This is a live demo request embedded in the RFP. Ask for a recorded walkthrough. The vendor should show a marketer-friendly interface that queries real warehouse data without requiring SQL. If the demo requires a data engineer to set up the segment first, that tells you something about the true persona for the tool.
- 10. How many destinations do you support natively, and how do you handle custom destinations?
Native connectors to downstream tools — ad platforms, email ESPs, CRMs, data warehouses — determine how much custom engineering your team will need to do after implementation. Ask for a current list of supported destinations, not a "300+ integrations" claim without specifics.
- 11. Can segments update in real time, or is there a minimum refresh interval?
Some use cases — cart abandonment, fraud signals, live event triggers — require segments that reflect events from seconds ago. Others work fine with daily batch refreshes. Know which cadences the platform supports natively and at what cost tier.
- 12. How do you handle suppression lists and audience exclusions at scale?
Suppression is often treated as an afterthought in CDPs, but it's operationally critical — particularly for compliance with unsubscribe lists or for preventing ad spend waste on existing customers. Ask what happens when a suppression list has five million records.
AI and Decisioning Questions
AI claims are everywhere in CDP marketing. These questions help distinguish productized capability from vaporware.
- 13. Which AI features are generally available today versus in beta or on a roadmap?
Request a written list. Anything framed as "coming soon" or "available to select customers" should be discounted from your evaluation. You're buying what exists, not what the vendor plans to build.
- 14. Where does AI-driven decisioning happen — inside the CDP or in a connected model layer?
Some platforms run their own proprietary ML models. Others let you bring your own models from SageMaker, Vertex AI, or Databricks and apply them in the CDP layer. Neither is universally superior, but the architecture determines how much your data science team can customize and audit the models in use.
- 15. How do AI-generated recommendations get reviewed or overridden by a marketer?
This question tests whether the system keeps humans in the loop. Responsible AI in marketing requires that a person can inspect a recommendation, understand why it was made, and override it when needed. Any answer that describes AI decisions as fully automated without a review mechanism warrants a follow-up.
- 16. Can you show us a case where AI decisioning improved a specific metric — open rate, conversion rate, revenue per customer — for a customer similar to us?
Ask for a named customer reference or a case study with specific numbers. Directional claims like "significantly improved engagement" without a metric or baseline are not sufficient.
Implementation, Support, and Total Cost Questions
The purchase price of a CDP is rarely the full cost. These questions surface what you're actually committing to.
- 17. What does a typical implementation look like for a company of our size and stack? Who does the work?
Some vendors rely heavily on system integrators, which adds cost and timeline risk. Others have in-house implementation teams. Ask for a typical project plan with milestones, and ask explicitly which tasks fall to your internal team versus the vendor.
- 18. What's the average time to first value — meaning the first production audience sync — for a new customer?
This is a benchmarking question. Strong answers give you a range ("six to twelve weeks for a mid-market customer") with the assumptions baked in. Non-answers reference "it depends" without offering any reference point.
- 19. How is pricing structured — by MAUs, events, connectors, rows processed, or something else?
CDP pricing models vary widely and can create unpredictable cost scaling. Understand the unit of billing and model out what happens to your costs if your data volume doubles, if you add three new destinations, or if you onboard a new brand.
- 20. What does your escalation path look like when there's a production issue at 2 a.m. on a Saturday?
SLA documents cover response times. This question tests whether there's a real escalation path or just a ticket queue. Ask for the specific process, not the target resolution time.
What to Look for in an Evaluation-Ready Vendor
A CDP vendor worth shortlisting will give you direct answers, offer reference customers with contact information, and be able to demonstrate their product against your actual data or a realistic proxy. They'll also be honest about limitations.
One architectural approach worth understanding: customer data stays in your warehouse, and the CDP layer sits on top of it without making a copy. The Agentic Marketing Platform from Hightouch connects that data foundation to marketer workflows, including audience building in Customer Studio, campaign orchestration in Lifecycle Marketing Studio, and paid media management through Hightouch Ad Studio. Identity Resolution is part of the Composable CDP layer, and AI Decisioning operates within Lifecycle Marketing Studio rather than as a disconnected module.
This architecture gives direct, verifiable answers to most of the questions above. Data lives in your warehouse. Schema changes in source systems don't break audiences because the query runs against the live warehouse. AI Decisioning recommendations are visible and overridable by marketers. None of those claims require you to take the vendor's word for it — they're testable in a proof of concept.
That's the standard to hold every vendor to: not polished answers, but answers you can verify.
Running the Evaluation After Responses Come In
Once you have written responses, score them against a rubric before you watch demos. It's easy to get impressed by a slick product walkthrough and forget that the vendor gave a non-answer to your data residency question.
For each question, rate the response on specificity (did they give a real answer or a deflection?), evidence (did they support claims with reference customers or concrete examples?), and fit (does the answer match your actual requirements?).
Also run a reference check against the questions in section four — specifically the AI section. Ask a reference customer whether the AI features they're using were generally available at purchase or were delivered later, and whether they were able to override AI recommendations when they disagreed.
The CDP market has no shortage of capable vendors. What it does have is a gap between what vendors demonstrate and what they deliver. Asking sharper questions is the most reliable way to close that gap before you sign anything.