The data model in my inbox

A 400-table airline data model landed in my inbox this week. Free. Pre-built by one of the biggest data platforms in the world, an AI agent attached to customize it.

I knew exactly what I was looking at.

I'd seen this model before. 20 years ago. With a different logo on it.

At Continental the warehouse was built by the people who lived the business—revenue management modeled bookings and tickets, loyalty modeled members, ops modeled flights. The grain was right, the relationships were right and the data taught you the business because the business had made the data.

That warehouse is where I learned this business. Not from a manual or an onboarding deck. From the tables. Why does a ticket have four coupons? What's an online OD? When would a redemption flight not show up in a member's history?

Teradata noticed what a few airlines had built and tried to bottle it. An industry logical data model—buy the canonical airline schema, skip the years of design. It was a serious effort, and they even hired someone off the team that built Continental's warehouse. But they bought the person and still couldn't buy the thing, because the thing was never the schema. It was the business owning the schema.

The pitch sounded right anyway—it always does. But in 20-some years of independent work across a lot of carriers and a lot of warehouses, I've never once seen an industry model working the way it was sold. What I have seen, twice, is my team helping a client tear out a half-built industry model and replace it with something their own people could actually use.

Here's the part people miss. These pre-built models don't fail loudly. They succeed—at the wrong goal. The industry model meets IT's need, which is completeness and a defensible architecture in one shot. It doesn't answer the business's need, which has never changed—our numbers, our way, this quarter. So we spend 18 months mapping source systems into the canonical shape while the business stands up shadow marts to get real work done, and by the time the model is installed it describes a world nobody's using anymore.

If you've been around airline data long enough you've seen one of these—the canonical model that took 2 years to implement and then got shipped into a vacuum.

Which brings me back to the model in my inbox. It's free, it's modern and the engineering is genuinely good. The packaging improved by 20 years, but the reason it doesn't work didn't move an inch.

Here's what took me too long to understand. The model was never the problem.

The same artifact means opposite things depending on who reaches for it. An airline that knows its business uses it as a checklist—a coverage list to pressure-test the model its own people are already building. An airline that doesn't know its business buys it as a solution. And buying it as a solution self-selects for the gap, because nobody who could build the right model needs the template and everybody who reaches for the template couldn't have built it.

An industry data model is the receipt for a decision that was already made—the decision not to learn your own business.

It even explains the ending we've all watched, where the business eventually gets the platform from IT and stands up its own sandbox on top of it. We read that as the business going rogue, but it isn't—it's the system finding its resting state. Knowledge beats infrastructure control on a long enough clock. IT can gate the data, but it can't gate it forever, and the day the business has its own compute it builds the thing it actually needs. The canonical model becomes one more artifact nobody opens.

So when someone hands you a 400-table model and calls it a head start, the model isn't the question. The question is which airline you are.

Ownership drifts to where the knowledge already lives. It always has.

The DOT wants to unbundle the advertised price. A proposal published July 1 would let airlines display fare, taxes and fees with equal prominence instead of one all-in price—and floats a full repeal that would allow base-fare-only advertising, with comments open until July 31. Read it as a data story, not a consumer story. The all-in number is the only thing that makes a fare comparable across channels, and the moment price becomes an assembly, every OTA, metasearch and shopping agent will assemble it differently. If your airline can't compute its own true price position by market, someone else's math becomes your brand.

Korean Air cleared the final regulatory approval for the Asiana merger—6 years and 13 competition authorities after the deal was announced—and the combination completes December 17. The hard part starts after the ceremony. Two reservation systems, two loyalty programs and two versions of revenue truth have to become one, and that integration decides whether the combined airline answers basic commercial questions in year one or year five. Regulators approve mergers. Data teams complete them.

A colleague from Toronto flew to Houston for the World Cup and sent me the text every former Houstonian waits for: where should I eat?

He got the full download. Fajitas at El Tiempo—get the bacon wrapped shrimp with whatever else you order. Breakfast tacos before the match from the most hole-in-the-wall spot you can find. Chicken fried steak at Dot's after, because the best diner you'll ever visit is open all night. Then Truth BBQ in Brenham on the drive west. In Austin, any taco truck, any time of day. Queso everywhere. And no matter what anyone tells you, DO NOT STAND IN LINE FOR BARBECUE.

This week I watched a team pull its work out of a request queue by shrinking the ask. Not the pipeline, not the curated view. Just land the raw data and give us access—one requirement too small to argue with and big enough to build everything else on.

Big asks wait in queues. Small asks ship.

—Ben