Customer Data Platform

Nobody has ever run marketing out of a single tool. Ads live in the ad platform, visit records in the analytics tool, signups and payments in the product database, send history in the email tool. Each tool also recognizes people by a different key: a cookie value in the browser, a device identifier in the app, an address in the email tool. One person ends up existing as four people in four places.

That turned an ordinary request into a two-day job. Someone would say "let's email last month's signups who haven't created a single document yet," and the signup list had to come out of the product side, the document activity out of the analytics tool, and the addresses out of the email tool, then get reconciled in a spreadsheet. Because the keys didn't match, rows got joined by name and address approximately, and roughly 200 people ended up attached to nothing.

The worse part was that the list was good for one day. Sending on the same condition the next week meant pulling everything again, and anyone who created a document in the meantime stayed on the old list — so people already using the product happily received a note saying they hadn't started yet. A customer data platform (CDP) takes that joining work out of human hands and keeps it running continuously.

It joins scattered records into one person and keeps them that way. The work splits into three steps.

1. Collecting — records from each tool are pulled into one place. What arrives is decided by the tracking code planted through a tag manager, the tool that lets you add and remove those small site snippets from one screen.
2. Joining — records that arrived under different keys get tied to one person. That processing is called identity resolution. An anonymous browser that clicked an ad yesterday and an account that logged in today are linked as one person on the evidence of the single moment when both keys appeared together.
3. Exporting — people matching a condition are selected from the joined result and handed to execution tools. Set a condition like "14 days since signup, zero documents, opted in to email" and the list refreshes itself daily, dropping anyone who no longer matches.

Once joined, a person becomes a single storyline: arrived through a search ad on March 2, signed up a week later, installed the app, opened two onboarding emails, still hasn't invited a teammate. Viewed tool by tool, those five fragments looked like records belonging to five different people.

In practice the value usually shows up in exclusions. Retargeting ads that keep chasing someone who already paid, a second email landing on someone who got one yesterday — those stop here. Deciding "this person already received it" requires that the records be joined under one person in the first place.

There are three ways to do the same job without buying the tool.

1. Reconciling by hand — download a list from each tool and match them in a spreadsheet. With one or two conditions it's quick and free. But it has to be redone every time, and since the matching rule lives in whoever did it, last month's number and this month's number stop agreeing.
2. A data warehouse — collect every record into one database and let an analyst pull lists with queries. It's flexible and the history runs deep. In exchange, the plumbing that pushes a finished list into ad and email tools has to be built, and a marketer who wants to change a condition goes through the analyst.
3. Audience features inside execution tools — marketing automation tools can also filter people by condition. Skipping the joining step makes this the fastest route. But such a tool only sees the records it holds. An email tool knows whether a message was opened; it has no idea whether a document was created in the app.

The deciding factor is not company size but how many keys are in play. Web only, with login required, means one key, and option 3 covers it. Once web, app, and offline event lists mix and pre-login behavior has to connect to post-login behavior, option 1 stops holding up and option 2 gets hard for marketers to operate. A web analytics tool doesn't substitute either — that side aggregates visit flows onto a screen rather than exporting lists of people.

"Pull that segment out of the CDP and push it straight to the ad account."

A request to export everyone matching a condition into the ad platform.

"Paying customers should be excluded from retargeting. Is the condition set in the CDP?"

Checking whether ad budget is still being spent on people who already bought.

"App and web still aren't resolving to the same person. Let's look at the merge rules."

The setting that ties the two keys together isn't finished.

At Pagedeck, Ravi sent an onboarding email to free users and Sofia came back with complaints. About 30 accounts that had already bought the team plan received a message inviting them to try the free plan. The email tool held no payment status, so the list had been built on signup date alone.

Ravi builds his lists in the customer data platform now rather than in the email tool. Payment status comes in from the product side and email consent from the email tool, and because both attach to one person, the exclusion is a single line in the condition. The half day that list-building used to take went away with it.

It's tempting to expect the data to clean itself up once the tool is connected. The word platform suggests something inside is tidying up. But this system only joins incoming records by the rules it's given, so if one tool calls a completed signup "signup" and another calls a started signup the same thing, the joined result carries that mismatch through. Without first agreeing on what each event name means across the team, an expensive tool gets installed and the same spreadsheet work continues.

Treating it as an analytics tool is just as common. It handles customer data, so opening it feels like it should show performance. What it produces is a list, not a chart. Why conversion dropped is a question for the analytics tool; who to contact about the drop is this side's job. Expect both from one tool and three months after rollout someone will say there isn't much to look at.