CDP is one of those acronyms that got popular faster than it got understood. I’ve been in rooms where three people used it to mean three different things and nobody noticed, which is a fun way to spend a budget.
So, plainly. A customer data platform collects customer data from all your different systems, stitches it into one profile per person, and makes those profiles available to your other tools. Your website knows one thing about someone, your email platform knows another, your support desk knows a third. A CDP is the thing that works out those are all the same human being and keeps a single up-to-date record of them.
That’s the definition, and it’s not complicated. The complexity is buried in the phrase “works out those are the same human being,” which is where these projects genuinely live or die. Most of this piece is about that, plus the mistake in the title, which is buying one before you’ve decided what you’d actually do with it.
What’s in this guide
- What a CDP actually does
- Identity resolution is the whole game
- What it isn’t: CRM, DMP, and your data warehouse
- Working out whether you need one
- A readiness worksheet
- How these projects actually fail
- The stuff worth remembering
What a CDP actually does
Three jobs, in order.
It collects. The platform pulls data in from wherever your customer information lives. Website behaviour, mobile app activity, purchases, email engagement, support tickets, whatever you connect. Some of this arrives in real time, some in scheduled batches.
It unifies. This is the interesting bit. The platform takes all those separate signals and decides which ones belong to the same person, then merges them into one profile. The anonymous visitor who read three articles last Tuesday and the customer who bought something on Friday might be the same person, and the CDP’s job is to notice.
It activates. Those unified profiles get pushed back out to the tools that need them. Your email platform can now segment on behaviour it never saw directly. Your website can personalise based on purchase history it doesn’t store. Your ad platform can suppress people who already bought.
The word that matters across all three is customer. A CDP is built around people, not campaigns and not pages. That sounds obvious until you compare it to the alternatives, where the organising unit is usually something else entirely.
Identity resolution is the whole game
If you take one thing from this, take this.
The value of a CDP is proportional to how well it answers a single question: are these two records the same person? Everything downstream depends on that. Great segmentation built on bad identity resolution is confidently wrong, which is worse than being uncertain.
The problem is harder than it sounds. Someone browses on their phone on the train, buys on a laptop that evening, and emails support from a work address the next week. Three sessions, three identifiers, possibly three email addresses, one human. Meanwhile two family members share a tablet, and your system happily merges a teenager’s browsing with their parent’s purchase history.
Platforms handle this with a mix of deterministic matching, which uses hard identifiers like a login or an email address and is reliable, and probabilistic matching, which infers from signals like device and behaviour patterns and is a guess with a confidence score attached. Most real setups use both.
When you’re evaluating, this is the thing to interrogate hardest. Ask how the matching works, what happens when two profiles are merged in error, and whether you can unmerge them. Ask what the platform does with anonymous visitors before they identify themselves, and what happens retroactively when they log in. Vendors will happily show you a beautiful segmentation interface. The interface is not the product. The identity graph underneath is the product.
What it isn’t
Three comparisons clear up most of the confusion.
A CRM is built around known contacts and relationships, usually for sales. It’s excellent at recording that you spoke to someone, what stage a deal is at, and who owns the account. It’s generally poor at high-volume behavioural data, and it only knows about people who’ve already identified themselves. A CDP handles anonymous behaviour, much larger data volumes, and automatic unification. The two are complements more often than alternatives, and plenty of stacks run both with the CDP feeding the CRM.
A DMP is built around anonymous audience segments for advertising, historically leaning on third-party cookies. It deals in aggregate audiences rather than individuals and traditionally forgets people quickly. As third-party cookies have declined, a lot of what DMPs did has shifted toward first-party data, which is precisely the territory a CDP occupies.
A data warehouse stores everything and does it well. What it typically doesn’t do out of the box is resolve identity, maintain profiles in something close to real time, or push data back into your marketing tools in a form they can use. Some teams do build CDP-like capability on top of a warehouse, and that’s a legitimate route if you have serious data engineering. It’s a build-versus-buy decision, not proof that you don’t need the capability.
Working out whether you need one
Here’s how I’d think it through, in order, because the order matters.
Start with a specific question you can’t currently answer. Not “we want a single customer view,” which is a wish rather than a requirement. Something like “which customers browsed this category in the last month and haven’t bought” or “which of our email subscribers are also active app users.” If you can’t produce two or three of these, you’re not ready, and the platform will arrive without a job to do.
Then check whether your data is genuinely fragmented. Count the systems holding customer information and ask whether anyone can currently join them up. If everything already lives in one place and it works, congratulations, you’ve solved this problem without buying anything. If you have five systems and reconciling them is somebody’s manual quarterly misery, that’s the shape of a CDP problem.
Next, ask whether you have enough data to matter. This one goes unasked and it should be first. A CDP earns its cost when there’s volume and variety to unify. If you have a few thousand customers and one channel, a well-run CRM and a decent spreadsheet habit will get you further for a fraction of the money.
Then be honest about who will own it. A CDP needs someone accountable for data quality, someone defining the segments, and technical resource to build and maintain the connections. This is not a set-and-forget purchase. Without an owner, profiles drift, connections silently break, and within a year you’re paying for a very expensive pipe.
Finally, decide what you’d activate. Unified profiles are only worth what you do with them. Write down the two or three campaigns or experiences you’d run in the first quarter. If that list is empty, you’re buying infrastructure and hoping strategy shows up later. It usually doesn’t.
A readiness worksheet
Those five questions are easy to nod along to and much harder to answer as a group, which is exactly why they’re worth doing properly.
We’ve put them into a short readiness worksheet: the five areas scored one to five, space to write the specific questions you can’t currently answer, an inventory of the systems holding customer data, and a section for the activations you’d run first. Fill it in with your marketing lead, your data or technical lead, and whoever owns the customer relationship. Then compare. The gaps between those three answers tend to be the real project.
What this looks like on a real customer
Let me make the unification concrete, because it’s the part people nod along to without picturing.
Take one fairly ordinary person. On a Tuesday they read two articles on your site from their phone, on mobile data, never logging in. On Thursday they click a link in a marketing email on the same phone. The following week they buy something from a laptop at work, using a different email address to the one they’re subscribed with. A month later they email support about the order.
Without a CDP, your systems hold four disconnected fragments. Analytics has an anonymous mobile session. Your email platform has a click. Your commerce system has an order under a work address. Your helpdesk has a ticket. Nobody’s wrong, and nobody has the picture. Reporting can tell you what happened in each channel, which is why marketing attribution arguments are so unwinnable.
With one, those fragments start collapsing into a single record. The email click is the moment of truth, because clicking a tracked link ties the known subscriber to that anonymous device. Everything the device did before, including Tuesday’s browsing, can now be attributed backwards. The purchase attaches through the shared device even though the email address differs. The support ticket attaches through the order number.
Now the useful things become possible. You can suppress that person from the acquisition campaign they no longer belong in. You can tell your support team the customer read two articles about the exact issue before they bought. You can stop emailing them about the product sitting in their hallway.
Notice how much of that hinged on one email click. Take it away and the chain breaks, the browsing stays anonymous, and the profile is thinner. This is why identification moments matter so much, and why teams with lots of anonymous traffic and few login prompts get less from a CDP than the demo suggested. The platform can only join up what you give it something to join on.
How these projects actually fail
Worth naming the common ones, because they repeat with remarkable consistency.
Buying before defining activation. The platform gets implemented, profiles get built, and then nobody’s quite sure what to do with them. Twelve months later it’s an expensive data pipe feeding a couple of email segments that a simpler tool could have produced.
Garbage in, unified garbage out. A CDP doesn’t fix data quality, it consolidates it. If three systems hold conflicting addresses, you now have one profile with a confidently wrong address. Cleaning up sources is unglamorous work that has to happen anyway.
Underestimating the connections. Every integration is real engineering, and every one needs maintaining as the systems on either end change. Teams budget for the licence and forget the pipework, which is where the time actually goes.
Treating identity resolution as a solved detail. Merge two people who aren’t the same and you’ll show someone else’s data to a customer. That’s not a data quality issue, that’s a privacy incident. Ask hard questions about match confidence before you sign.
Forgetting consent travels with the data. If someone opted out of marketing in one channel, that has to be respected everywhere the profile goes. Building the platform without mapping consent alongside it creates a compliance problem that’s much harder to retrofit than to design in.
The stuff worth remembering
- A CDP collects customer data from your systems, unifies it into one profile per person, and pushes those profiles back out to the tools that need them.
- Identity resolution is the actual product. A beautiful segmentation interface built on weak matching produces confident nonsense.
- It’s not a CRM, which is built around known contacts and sales relationships. It’s not a DMP, which deals in anonymous ad audiences. It’s not a warehouse, though you can build similar capability on one if you have the engineering.
- Start from questions you can’t currently answer and activations you’d actually run. If both lists are empty, you’re not ready.
- Volume matters. Small customer base and one channel usually means a CRM will serve you better.
- The failure mode is rarely the technology. It’s buying before defining what it’s for, and having nobody who owns it afterwards.
Not sure whether the problem is your data or your tools?
A CDP makes more sense next to the rest: how it differs from a CRM, the CDP vs DMP vs CRM picture, and how to choose one.
Plenty of the conversations worth having here end with “your data is fine, the issue is that nobody has defined what you’d do with it.” If you want a second opinion on whether a CDP is the right answer for your situation, or whether something smaller and cheaper would do the job, get in touch and we’ll work through it with you.