Latest / MBA Training Data / Data clean rooms explained: what they actually do and when they're worth the effort
Transcript
- 0:00Welcome to the Leaders Insights Podcast.
- 0:07Today's episode, Data Clean Rooms explained what they actually do and when they're worth
- 0:12the effort.
- 0:13Data Clean Rooms are the darling of every privacy-conscious marketing deck right now.
- 0:18So tell me honestly, are most of them a solution looking for a problem?
- 0:22Most of them, yes.
- 0:24About eight in 10 clean room projects I've seen get sold on privacy theater.
- 0:29The idea that if two companies never see each other's raw data, everything is fine.
- 0:34That's not why you'd build one and pretending otherwise is how you waste a year.
- 0:38Strong start.
- 0:39So what actually is a clean room in plain terms?
- 0:42It's a locked room where two parties put their data, run agreed queries against the combined
- 0:46set and only take out aggregate answers, never the individual records.
- 0:52Think of it as a shared kitchen where you both bring ingredients, cook one dish, and
- 0:57neither of you gets to walk off with the other's groceries.
- 1:00Fine, but if the whole selling point is privacy, why do you say that's the wrong reason?
- 1:05Because privacy is table stakes, not value.
- 1:08A clean room that protects data but tells you nothing new is an expensive vault.
- 1:13The reason to build one is overlap.
- 1:16You, finding the customers you share with a partner that neither of you could identify
- 1:20alone.
- 1:21If the overlap is thin, you've built infrastructure for a rounding error.
- 1:26Give me a number.
- 1:27What counts as thin?
- 1:28Google's own ads data hub materials, and that's a vendor pitching its own clean room, so read
- 1:33it with a pinch of salt.
- 1:35Lean on match rates north of 50% to make the maths work.
- 1:38In practice, I've watched two retailers discover their shared customer base was 6%.
- 1:44Six.
- 1:45You don't need a privacy-preserving compute environment for 6%.
- 1:48You need a coffee and an honest conversation.
- 1:51So who's it genuinely for?
- 1:53Retail media is the honest use case.
- 1:56A supermarket like Tesco, through its Dunhumbi arm, sits on loyalty data.
- 2:01Who actually bought the yogurt?
- 2:03A brand like Denone wants to know if its ads drove those purchases.
- 2:07Neither will hand over raw customer lists, and legally, they can't.
- 2:12The clean room lets Denone measure the sales lift without ever seeing a single
- 2:16Tesco shopper's name.
- 2:18That's real money.
- 2:19Retail media is a fast-growing ad category precisely because of this.
- 2:24You said everyone gets something wrong.
- 2:25Name it.
- 2:26People think the clean room removes the legal risk.
- 2:29It doesn't.
- 2:30It removes the raw data exposure, which is a different thing.
- 2:33You still need a lawful basis to combine the data in the first place.
- 2:37Consent, contract, whatever your regulator demands.
- 2:42I've seen companies treat the clean room like a legal laundromat,
- 2:45as if running personal data through a fancy query engine washes the
- 2:49compliance out of it.
- 2:51Regulators don't grade on cleverness.
- 2:53So the technology gives false comfort.
- 2:55Exactly.
- 2:56And there's a subtler trap.
- 2:58Differential privacy, the maths that adds statistical noise,
- 3:02so no single person can be reverse engineered from the results.
- 3:05It's genuinely good, but every vendor tunes how much noise to add,
- 3:10and more noise means fuzzier answers.
- 3:12Ask them where that dial is set.
- 3:15If they can't tell you, they're selling you a black box with a privacy sticker.
- 3:19What does one of these cost to run, realistically?
- 3:22The license is the cheap part.
- 3:24The expensive part is the people.
- 3:26You need someone who can write the queries,
- 3:28someone who understands the privacy budget,
- 3:31and a data team on both sides keeping the inputs clean.
- 3:35Budget six to nine months before you get a decision you'd act on.
- 3:39Snowflake and AWS will tell you it's faster.
- 3:42Their sales targets require them too.
- 3:44So when is it actually worth the effort?
- 3:46Three conditions, all of them.
- 3:48High overlap with the partner, measure it first with a cheap test.
- 3:53A decision worth real money attached to the answer.
- 3:56And a partner who will still be at the table in a year,
- 3:59because clean rooms reward repeated use, not one-off flings.
- 4:03Give me the one thing a CDO should do Monday morning.
- 4:05Before you sign anything, run a match rate estimate,
- 4:08a rough count of shared customers, on a sample.
- 4:12If the overlap's under 20%, walk away and save yourself the platform fee.
- 4:17Done.
- 4:18Want to know where you actually stand?
- 4:20Take the CDO self-assessment at MBA-training.com.