A Federal Model for Data Governance (part 2)
- Harper’s Ferry, West Virginia, July 2025This is the second article in a four-part series on my federal model for data governance. Find the beginning of the series here.
Friends and colleagues know I get as much mileage as possible from a good simile. You shouldn’t be surprised that data people think like this. After all, you can’t touch, taste, or weigh data. We need a good word picture to communicate our abstract work to the real world.
Last week, I wrote about one of my favorites: data should flow through a company like cars traveling on the highway. Good traffic flow hinges on balancing people’s decisions about where they want to go with the structures that make it flow for everyone:
People want to enjoy fast, safe, and relaxing drives, but that won’t happen if everyone drives at a different speed.
- Everyone has an opinion on where the roads should go, but you (the data team) know how to build highways.
- Drivers want the freedom to control their vehicles, but also need to cooperate so everyone can get where they’re going.
Just like controlling traffic, you need a plan to govern your data.
In this article, I’ll explain the two common approaches companies use today for governing data, the “democratic” model and the “traditional” model. As I explain them, I want you to notice how little these approaches require business leaders to understand how to manage their data.
Competing Models
Experts compare “data operating models” to different forms of government. That makes sense because a data solution isn’t purely technical; it’s also political. Decisions about which project to work on next (for example) give power to the decision makers.
In the traditional, centralized governance model, a team under the Chief Information Officer makes all the decisions. Anyone who needs data must submit a request and wait in line. All of a company’s data solutions depend on that one team's designs, talents, and capacity.
That operating model sounds so painful that most analytics software vendors use it as a strawman to explain why you need to buy their solution. In reality, very few companies follow a fully centralized model, but sometimes it can work. A recent case study showed how Georgia-Pacific Railways used this approach to improve data quality across all its subsidiaries. Things run smoother in a centralized model like this because everyone uses the same-sized railroad tracks. But the tracks only go to a few destinations.
Most companies I’ve worked with default to the “democratic” model, where an IT-based data team makes only technical decisions. The data platform design depends on what the users ask for. Every team asks for something different, and the system usually looks like a road to nowhere. Some of the largest companies in the world (like Amazon) use this approach. You can tell this from the skills requirements listed on their analyst job postings: they like business analysts who know SQL and create dashboards.
The democratic model rests on the premise that knowledge workers find better ideas when they work independently. Many analysts love this model because it gives them the freedom to make every decision about their data themselves. However, the democratic model has some inherent problems:
- Data duplication (multiple people recreate the same work)
- Data exfiltration (people taking data with them)
- Siloed data (custom-built, massaged sources)
- Conflicting answers to the same questions.
- Analysts spend less time analyzing.
- Unrestrained confirmation bias.
Everyone builds their own highways, but it leads to chaos.
You might notice the glaring problem these two governance models have in common: neither approach requires business and IT people to share the responsibility for managing their company’s data. That’s the problem.
You need a plan that helps everyone share the road.
A Third Way
I’ve argued that neither of these data governance models works for everyone, yet virtually every company follows one of these two approaches.
Don’t choose between these two approaches. I’ll explain my “frictionless” data governance strategy in the next part of this series, or you can read the whole story in my book. Frictionless data doesn’t just alleviate the problems of these two models. If that’s all you needed to do, you could figure out a patchwork way to lessen their problems on your own. But you’d do nothing to improve or speed up your company's decisions.
You can challenge your company’s assumptions about how it governs data. Let’s do this together.