Customer data platforms: discipline before architecture

September 25, 2026 · 5 min · Free, with full transcript

Does a marketing team with messy customer records need a customer data platform, or a governance fix it keeps refusing to fund? The position here is blunt: most CDP purchases try to solve a discipline problem with software, which is why many deployments sit unused 18 months later. Examples include a $400 million retailer paying about $300,000 a year into a platform fed with unusable data, Forrester findings that failures cluster around governance, and vendor claims from HubSpot and Semrush that deserve a crosscheck. You leave with a test for telling the two problems apart, a 90 day manual reconciliation plan, and the sentence that signals a team is ready to buy.

Listen to the podcast

4 min

Chapters

Key takeaways

  • Ask whether you can already name your best customers and their last action with the systems you own; if yes but slowly, the problem is discipline.
  • Appoint one owner for the definition of every core metric, including customer, active and churned, before evaluating any platform.
  • Reconcile existing systems by hand for 90 days until the numbers agree, which surfaces whether the problem was ever technical.
  • Discount conversion lift figures from vendors like HubSpot and Semrush that sell the tools feeding the claim, and crosscheck against independent research.
  • Pull your top hundred customers from every system and compare the lists; if they do not match, cancel the demo.
Read the full transcript

Host:You're listening to Leaders Insights. Today's subject, does your company actually need a CDP or just better discipline?

Expert:Most companies buying a customer data platform are solving a discipline problem with a software purchase. And it never works. That's the whole game, and it's why half the CDP deployments I see are shelfware 18 months in. A customer data platform, the thing that supposedly stitches all your customer records into one profile, is architecture. But the mess underneath it is usually behavior.

Host:Define the mess. What are you actually walking into?

Expert:Duplicate records. Three teams each with their own definition of an active customer. And nobody agreeing what a lead even means. I sat with a retailer last quarter, mid-size, roughly $400 million in revenue, who'd signed a three-year contract at about $300,000 a year. The CDP worked fine. The data going into it was garbage, so the profiles it built were garbage.

Host:But surely the vendor pitch is that the platform cleans that up for you?

Expert:The vendor pitch is a lovely fiction. No tool resolves a definitional argument between your sales director and your email team. That's a meeting, not a migration. What the CDP did was make the disagreement more expensive and more visible.

Host:So how does a sane operator tell the two problems apart before they sign?

Expert:One question. Can you already answer who are my best customers and what did they do last? With the systems you own? If the answer is yes, but it takes a week and three exports, you have a discipline problem. If the answer is genuinely no, the data physically lives in nine places that can't talk. Now you might have an architecture problem.

Host:Nine places sounds like every company over a certain size, though.

Expert:It does, and that's the trap. Forrester, key independent research shop, they don't sell the software, has been clear that the failures cluster around governance, not technology. Who owns the definitions? Who's allowed to change them? Who gets blamed when the numbers don't match? That's unglamorous and nobody wants to fund it.

Host:Give me the number on how big this market's gotten, because the hype is doing a lot of work here.

Expert:North of $5 billion now, and every quarter another vendor rebrands their old database as a CDP. HubSpot will tell you unified data lifts conversion by some cheerful double-digit percentage. Worth noting they sell the CRM that feeds it, so crosscheck that against someone with no dog in the fight.

Host:You mentioned that, retailer. What would have actually fixed them?

Expert:Two people and a spreadsheet for 90 days. Before you buy anything, you appoint one person who owns the definition of every core metric. Customer, active, churned. You reconcile your existing systems by hand until the numbers agree. Boring, cheap, and it surfaces whether your problem was ever technical.

Host:And if it turns out it genuinely is technical?

Expert:Then buy. But now you buy knowing what you're feeding it, so the thing actually earns its keep. The order matters. Discipline first, then architecture. Do it backwards, and you've automated your confusion at $300,000 a year.

Host:Isn't there a risk that get discipline first is just an excuse to never modernize? The stall tactic dressed up as prudence?

Expert:There's a version of that, sure. But I've watched far more money burned rushing than dawdling. Semrush, again, a vendor. They sell analytics tools. So hold their numbers loosely, and Don reckons a big chunk of MarTech spend goes unused. The people overbuying aren't the cautious ones.

Host:What's the tell that a team has actually earned the right to buy?

Expert:And the argument in the room stops being which platform and becomes, We agree on the numbers, we just can't produce them fast enough. That sentence means you've done the human work. Until you can say it out loud, you're shopping for a scapegoat, not a solution.

Host:Leave the listener with the one thing to do Monday morning.

Expert:Pull your top hundred customers from every system you own and lay them side by side. If the lists don't match, cancel the demo. Your problem was never the software.

Host:This episode draws on Forrester Research, Semrush Vendor, SEO Analytics Tools, DigiDay, HubSpot, Vendor, CRM Marketing Automation. That's it from us. The reading continues at MBA-training.com, new CMO analysis every day.

Read and go deeper