CDP demos all look the same. A nice dashboard, a customer profile full of icons, an audience built in three clicks and synced to Meta before your eyes. You walk out convinced, and six months later you discover the hard part was somewhere else.
These are the eight criteria that really separate a CDP implementation that works from one that stalls halfway, and the specific questions worth asking each vendor.
1. Identity resolution
It is criterion number one and the one that gets examined least. Every CDP says it unifies profiles. What changes is how.
Deterministic matching joins records only when they share a reliable identifier: email, phone, customer or loyalty number, order ID. It is precise, and its coverage depends on how many identifiers it can use and how well it normalises them before comparing. Probabilistic matching extends coverage based on similarities, and in exchange brings the risk of false positives: two different people merged into one profile, with the privacy problem that implies. Whatever the approach, what is essential is being able to see why the system decided two records were the same person, and to undo the merge if it got it wrong.
Question for the demo: show me a unified profile and tell me which source each data point came from and which rule joined them.
2. What happens to your offline data
If you have physical stores, this point decides the project. Many CDPs were born for pure ecommerce and treat store data as an add-on: file uploads, once a day, with no decent identity resolution.
If 60% of your revenue goes through the till, a platform that only understands digital will give you a partial and expensive view.
Question: how does the store receipt come in, and how long until it shows up in the profile?
3. Real activation, not export
There is a lot of hidden nuance here. "Integrates with Meta Ads" can mean it syncs audiences automatically every hour, or that it lets you download a CSV you upload by hand. Ask to see each connector working, not a list of logos (here are our connectors, by category).
The day-to-day difference is huge. A team that has to export and import ends up not doing it.
Question: which destinations are native and bidirectional, which require development, and how often are audiences refreshed?
4. Business team autonomy
The sales pitch of almost every CDP is that marketing stops depending on IT. It is worth checking with a real case rather than the demo example.
Ask them to build, live, a segment that looks like the ones you really need, with three or four combined conditions. If someone from support has to be called in, you have your answer.
Question: can someone from my team do it during the trial, not your consultant?
5. Time to first value
The deadlines that matter are not the ones in the contract, but these three: how long until the first source is connected and validated, when the first campaign with unified data goes live, and when there are measurable results.
A serious implementation has the first source in weeks and the first active audiences around the second month. If they promise everything in two weeks, they are describing a test connection. If they talk about six months before you see anything, the project will lose internal sponsorship before it gets there.
Question: give me the plan for the first eight weeks with concrete milestones.
6. The real cost
The licence price is the easy part. What breaks budgets is usually elsewhere:
- Pricing by volume of profiles or events, which grows with you and sometimes faster than expected.
- The cost of integrating non-standard sources. Your fifteen-year-old ERP is not in the connector catalogue.
- Infrastructure cost, if the platform runs in your cloud.
- Internal time. Someone on your team will spend real time validating data. It does not appear on any invoice, and it is a cost all the same.
Question: simulate my year-three invoice with my expected growth.
7. Regulatory compliance, seriously
If you operate in Europe you need GDPR. If you operate in Mexico, the LFPDPPP. If in both, both, and they are not identical.
What to look at is not whether they say they comply, but how. Where the data is physically hosted. What happens when a customer exercises their right to erasure and that customer is replicated across five systems. Whether consent travels with the data or stays in the original form. Whether your information trains third-party models.
Question: show me the full flow of an erasure request, from start to finish.
8. Who supports you
The factor that correlates most with the success of these projects is not technical. It is whether there is someone on the other side who knows your business.
A well-implemented CDP requires business decisions, not just engineering ones: what defines an active customer, which attribution window makes sense in your sector, which segments deserve their own model. If your vendor hands over the tool and disappears, those decisions never get made and the platform ends up underused.
Question: who is my point of contact, in which language and time zone, and what happens when I want to change something six months from now?
A warning about CDP comparisons
Analyst quadrants and rankings are useful for knowing what exists, and fairly bad for deciding. They are dominated by platforms sized for companies with revenues well above average, with in-house data teams that most companies do not have.
A platform that is perfect for a multinational with fifteen analysts can be exactly the wrong one for a €50 million company with two people in marketing. The question is not which is the best CDP on the market, but which one fits the team that will use it.
How to structure the decision
What works:
- Write down the three concrete use cases you want to solve in the first year. Concrete: "exclude active customers from acquisition campaigns", not "improve customer knowledge".
- List your real data sources, including the awkward ones.
- Ask each candidate to demonstrate those three cases with data similar to yours.
- Have your team run the trial, not theirs.
- Ask for references from customers of your size and sector, and call them.
Frequently asked questions about choosing a CDP
How do I choose the best CDP for my business? Start from three concrete use cases for the first year and your real data sources. Ask each vendor to demonstrate those cases with data like yours, let your own team run the trial and call references from companies of your size and sector.
What criteria matter most when evaluating a CDP? Identity resolution, how it handles offline and store data, real activation in your channels, how autonomous your business team is, time to first value, the real cost over three years, regulatory compliance and who supports you after go-live.
What is the difference between deterministic and probabilistic identity resolution? Deterministic matching joins records only when they share a reliable identifier such as email, phone or customer number, so it is precise. Probabilistic matching relies on similarities to cover more records, at the risk of merging different people. In both cases you should be able to see why two records were merged and undo it.
How long does it take to implement a CDP? In a serious implementation, the first source is connected and validated within weeks and the first active audiences arrive around the second month. Promises of everything in two weeks usually describe a test connection; six months without results puts the project at risk.
What hidden costs does a CDP have? Pricing by volume of profiles or events, integrating non-standard sources such as an old ERP, infrastructure if it runs in your cloud, and the internal time your team spends validating data.
If you are at an earlier stage and still wondering whether you need one, start with What is a CDP and what is it for in marketing. And if your question is whether the CRM you already have covers this, it is answered in I already have a CRM, why would I need a CDP?
At FLYDE we would rather you evaluated us with these questions. Book a 15-minute demo and bring them with you.