Latest / MBA Training Data / Clean rooms in practice: a CDO's playbook for data collaboration without the risk
Transcript
- 0:00Welcome to the Leaders Insights podcast.
- 0:08Today's episode, Clean Rooms in Practice, a CDO playbook for data collaboration that
- 0:13actually works.
- 0:14Last week, a mid-market retailer told me they'd spent 18 months and roughly $2 million
- 0:19standing up a data clean room with a big media partner and got exactly one usable report
- 0:24out of it.
- 0:25What went wrong?
- 0:26Governance, almost certainly.
- 0:28A clean room, a shared environment where two companies run joint analysis without either
- 0:34seeing the other's raw customer records, only works if both sides agree in advance what
- 0:39questions are allowed.
- 0:41They didn't.
- 0:42So every query became a legal negotiation.
- 0:44So the technology wasn't the problem?
- 0:47The technology is almost never the problem.
- 0:49Snowflake, Google's Ads Data Hub, AWS Clean Rooms, LiveRamp—they all do the encryption
- 0:55and the aggregation fine.
- 0:58What kills these projects is that the two parties want different things and never say
- 1:02so out loud.
- 1:03Give me a real example.
- 1:05A consumer packaged goods brand wants to know which of its buyers also shop at a specific
- 1:09grocery chain.
- 1:11The grocer wants to sell ads.
- 1:13Those aren't the same goal.
- 1:15The brand wants insight.
- 1:16The retailer wants revenue.
- 1:18If you don't reconcile that on day one, you build a room nobody wants to enter.
- 1:22You've published this as a playbook with a specific sequence.
- 1:26That's step one and don't tell me, align stakeholders.
- 1:29Step one is you write the questions down.
- 1:32Literally.
- 1:33Before you buy anything, both sides list the five business questions the clean room must
- 1:37answer.
- 1:39If you can't agree on five, you don't have a partnership, you have a press release.
- 1:43That sounds almost insultingly simple.
- 1:46It is simple and almost nobody does it.
- 1:49I've seen teams sign a vendor contract before they've written a single question.
- 1:53Then they spend a year reverse engineering a purpose.
- 1:56The tools charge you by compute.
- 1:58Every query you run costs money, so you're literally paying to figure out what you wanted.
- 2:04Let's talk about that overlap problem.
- 2:06What if the two data sets barely intersect?
- 2:08Then you find out fast and cheap.
- 2:10Step two is a match rate test.
- 2:12How many of your customers actually appear in their data?
- 2:15Run it before the full build.
- 2:17If your overlap is 8%, no governance framework on earth makes that partnership worth it.
- 2:23I've seen deals collapse on that number alone.
- 2:26And thank God they did before the spend.
- 2:28What's a good overlap?
- 2:30Depends on the use case.
- 2:31But for advertising measurement, you generally want north of 30% to trust the results.
- 2:37Below that, you're extrapolating from a rounding error and calling it strategy.
- 2:41The article mentions differential privacy and aggregation thresholds.
- 2:45Translate.
- 2:47Aggregation threshold means the room refuses to show you any result covering fewer than,
- 2:51say, 50 people, so you can't back into identifying one individual.
- 2:57Differential privacy goes further.
- 2:59It adds a bit of statistical noise to the output so no single person's data can be reverse
- 3:03engineered.
- 3:05Both protect you legally, and they protect the customer.
- 3:08Set those numbers in the contract, not in a slack message.
- 3:12You mentioned misaligned incentives.
- 3:14How do you fix a partner who just wants to sell you ads?
- 3:17You make the value flow both directions.
- 3:19If they only monetize and you only spend, they'll milk it and you'll resent it.
- 3:24So step three, agree on a shared metric you both improve.
- 3:28Incremental sales lift, say, the extra revenue that came specifically from the joint activity.
- 3:34Not the sales you'd have gotten anyway.
- 3:37When both teams are measured on the same number, the negotiations stop.
- 3:41Vendors love pitching clean rooms as privacy safe.
- 3:44Are they actually or is that marketing?
- 3:47It's mostly real, but read the fine print.
- 3:49The vendor's own benchmark saying 90% faster insights is a sales deck, not a fact.
- 3:56The privacy protection is genuine if you configure the thresholds correctly.
- 4:01The speed claims are whatever they measured in a lab with clean data you'll never have.
- 4:06One thing our listener can do tomorrow morning.
- 4:08Open a blank document, write the five business questions your clean room must answer, and
- 4:13email them to your prospective partner.
- 4:16If they can't send five back within a week, walk.
- 4:19You just saved yourself $2 million in 18 months.
- 4:22That's it.
- 4:23The CDO Decision Tools are live at MBA-training.com.