Latest / MBA Training Data / Internal data products: a field guide to the teams and companies worth studying
Transcript
- 0:00Welcome to the Leaders Insights Podcast. Today's episode, Internal Data Products, a field guide
- 0:11to the teams and companies worth studying. There's a new field guide out that ranks
- 0:15the companies worth studying for internal data products, and it doesn't read like the
- 0:19usual vendor puff. What's the one name at the top that should make everyone else nervous?
- 0:24Airbnb. And not because they're loud about it, because of Minerva, their metrics layer,
- 0:30which they've been documenting publicly since 2021. The idea is dead simple, oh well, and almost
- 0:37nobody executes it. One place where active user or revenue is defined once, computed once,
- 0:43and everyone downstream drinks from the same tap. That sounds like plumbing. Why does the field guide
- 0:49treat it like the crown jewel? Because the alternative is chaos with a straight face. When
- 0:55finance says revenue was 100 and the growth team says 94, you don't have a data problem,
- 1:01you have a definitions problem. Minerva means the number is the number. Airbnb reported it killed
- 1:07off a swamp of conflicting dashboards. That's the whole game. Fine, but Airbnb has thousands
- 1:13of engineers. What's a company with 100 people supposed to take from that? The principle scales
- 1:18down. You don't need Minerva the software. You need Minerva the discipline. Pick your 15 metrics
- 1:24that actually run the business, define them in one file, and refuse to let anyone redefine churn.
- 1:30The share of customers who cancel in their own spreadsheet. That costs you a week, not a platform
- 1:36team. The guide ranks by documented influence. That's a polite way of saying most companies
- 1:42just copy the ones who blog. Isn't this survivorship bias dressed up as research?
- 1:47Partly, yes, and the author more or less admits it. We study the companies that write things down,
- 1:53and the ones who write are disproportionately the ones who succeeded. But there's signal in the
- 1:58pattern. When Uber, Spotify, and Netflix independently land on the same shape,
- 2:03a self-serve platform, an internal catalog, treating data sets like products with owners,
- 2:09that convergence isn't marketing. That's physics. Netflix gets name checked constantly. What did
- 2:15they actually build that's worth stealing versus admiring? Their data mesh work, and I mean the
- 2:20real definition, decentralized ownership, where the team that produces the data, also owns its quality,
- 2:26instead of a central data team playing janitor for everyone. Netflix's move was making that
- 2:32ownership non-optional. If your team ships a data set, your name's on it. And when it breaks
- 2:38at 2 a.m., it's your pager. That sounds like a way to make engineers hate you. It is. Until they
- 2:44realize the central team model made everyone hate each other. The dirty secret is most data
- 2:49quality failures are accountability failures. Nobody owned the pipeline, so nobody fixed it.
- 2:56Spotify's internal platform, they call it a paved road, the supported easy default path,
- 3:02works because ignoring it is harder than using it. That's the trick. You don't mandate good
- 3:07behavior. You make it the lazy option. Let's talk monetization, since that's supposedly the
- 3:13point. Internal data products don't make money. How do you justify the spend to a CFO in 2026
- 3:19who's cutting everything? You reframe it. Internal data products don't earn revenue.
- 3:24They reduce the tax on every decision. Airbnb's own framing. Analysts were spending the majority
- 3:31of their time finding and reconciling data, not analyzing it. Cut that in half, and you've
- 3:37effectively doubled your analytics headcount without hiring. That's a number a CFO understands.
- 3:43Effectively doubled is the kind of soft math that gets people fired. Which is why you measure it
- 3:48directly. Time to first query for a new hire. Number of duplicate pipelines you deleted.
- 3:53Tickets to the data team per month. Those are hard numbers. The vendors selling catalogs will
- 3:59quote you a 400% return. Ignore that. It's their slide. Track your own three metrics before and
- 4:06after. If they don't move, you built a museum, not a platform. So what's the one thing a listener
- 4:12should do Monday morning? Open your dashboards and find two that claim to show the same metric
- 4:17with different numbers. There's always a pair. Fix that one definition in one place this week.
- 4:23That's the smallest possible version of everything Airbnb built. And it's the only version that ever
- 4:29earns you the right to build the big one. That's it. The CDO decision tools are live at
- 4:34MBA dash training dot com.