Composable CDP: build it on your data warehouse

September 1, 2026 · 4 min · Free, with full transcript

Can a marketing team get one customer profile without shipping its data into a vendor's black box and renting it back? The position here is yes, if you stop trying to own a database and ask your data team for a schema inside the warehouse you already pay for, whether that is Snowflake or BigQuery, then add a reverse ETL layer like Census or Hightouch for activation. You get the cost comparison against a packaged Segment or Adobe deal, including a move from a quoted $240,000 a year to roughly $40,000, the pricing shift from records stored to syncs run, the identity resolution logic you now write in SQL yourself, and the metric that proves it worked: time from idea to live audience, cut from 11 days to under one.

Listen to the podcast

4 min

Chapters

Key takeaways

  • Ask your data team for a schema inside the existing warehouse instead of standing up a separate marketing database.
  • Add a reverse ETL tool such as Census or Hightouch so marketers can build audiences without filing an engineering ticket.
  • Compare pricing models directly: packaged CDPs scale with records stored, reverse ETL tools charge per sync.
  • Write and review your identity resolution logic in SQL, because a bad match sends a welcome offer to someone who just spent $900.
  • Track time from idea to live audience as the proof metric, not just the size of the vendor bill.
Read the full transcript

Host:Welcome to the Leaders Insights podcast. Today's episode, Building a Composable CDP on Your Data Warehouse, a CMO's playbook. The Composable CDP just crossed into boring and mainstream territory this year, and that's exactly when I get suspicious. Salesforce paid roughly a billion dollars for a data integration startup called OWN back in 2024. Because the package thing quietly bills you for a copy of data you already own.

Expert:A CDP, customer data platform, the system that stitches every customer touchpoint into one profile, traditionally means you ship your data into a vendor's black box and rent it back.

Host:Composable means the profile lives in your warehouse, the snowflake or BigQuery you already pay for, and the tools just read from it.

Expert:Correct, and that's the real fight. The playbook's whole premise is that you're not building anything new. You're borrowing infrastructure that finance and product already trust. When I did this at a mid-size retailer, the win wasn't technical. It was walking into the data team and saying, I don't want my own database, I want a schema in yours. That sentence turned a nine-month turf war into a three-week project.

Host:Fine, but the old sales pitch for a package CDP was speed. Marketers can't wait on a data engineering backlog to launch a campaign. Doesn't Composable just move the bottleneck?

Expert:It would if you stopped at the warehouse. You need what's called reverse ETL, a tool that pushes data out of the warehouse back into your ad platforms and email, the opposite of the usual pull-it-in direction. Census and high touch are the two names here. That layer is what lets a marketer build an audience without filing a ticket. The engineers own the data. Marketing owns the activation.

Host:You just named two vendors, so we're back to paying somebody.

Expert:Sure, but the bill is different in kind. A traditional CDP like a segment or an Adobe deal starts around six figures a year, and the price scales with how many customer records you store. You get punished for growing. Reverse ETL tools charge on how many times you sync, not how much data you hoard. One company I advised went from a quoted $240,000 a year to about $40,000, and that pricing's from the vendors, so take the exact numbers with salt, but the shape of it is real.

Host:What breaks? Nobody builds this in sales off happily. Where's the two-in-the-morning phone call?

Expert:CD resolution. Deciding that the person who bought on the app and the person who abandoned a cart on the website are the same human. Packaged CDPs hide that logic. Composable means you write it yourself in SQL, the query language your data team already speaks. Get it wrong, and you email a customer a welcome offer the day after they spent $900. That's the failure that gets a CMO fired, not the tech bill.

Host:So the honest version is, you trade a vendor's black box for your own homework.

Expert:Yes, and I'd argue that's the point. When the matching logic lives in your warehouse, you can actually see why two profiles merged. In the black box, when it's wrong, you file a support ticket and pray. I'll take homework I can inspect over magic I can't.

Host:Give me the number that tells a CMO this was worth it, not just cheaper.

Expert:Time from idea to live audience. Before, at that retailer, a new segment took 11 days because it round tripped through the vendor. After, under a day. Cost went down and speed went up, which almost never happens together. So I don't blame anyone for being suspicious.

Host:Last thing, one move a CMO makes Monday morning.

Expert:Book 30 minutes with whoever owns your warehouse and ask one question. Where do our customer profiles actually live today? And who can read them? If nobody can answer cleanly, you don't have a CDP problem. You have a plumbing problem and no vendor will fix that for you.

Host:That's your five. The full CMO track, structured no fluff, is at MBA-training.com.

Read and go deeper