You Are Already Doing Data Governance. You Just Don't Know It Yet.

You Are Already Doing Data Governance. You Just Don't Know It Yet.
Photo by Nick Fewings / Unsplash

In 1940, Ludwig Mies van der Rohe drew up a master plan for the Illinois Institute of Technology in Chicago. It was among the most rigorously ordered campus plans in America, laid out on a 24-by-24-foot grid that governed where every building and column would sit. Mies explained the grid simply: "Orderliness was the real reason."

Over fifty years later, the students had drawn a plan of their own.

By the 1990s, Chicago's elevated train line split the campus in two, with residence halls on one side and academic buildings on the other. Between them sat a heavily used parking lot, and students crossed it on whatever diagonal got them where they were going, grid or no grid.

When the university held an international competition to design a campus center on that lot, Rem Koolhaas and his firm, OMA, did something unusual before putting pen to paper. In 1997, they had a team of students track how people actually moved across the site, then mapped the most heavily traveled routes. Planners call those routes desire lines, or desire paths: the way people really walk, which is often in contrast to the orderly 90-degree sidewalk paths of a typical urban environment.

Koolhaas won the commission in 1998, and the map became the building. The McCormick Tribune Campus Center opened in September 2003, organized around diagonal interior walkways that extend the paths students were already using, with the bookstore, food court, and meeting spaces set in the islands between them. Not everyone loved it. Linking the new center to one of Mies's original buildings meant a fight with purists who wanted his work left standing alone. The diagonals stayed anyway, because students had been walking them for years.

Your company has its own version of that parking lot. Let's call it the "corporate lawn." A sidewalk runs around the grass at a crisp right angle, and within a month a brown stripe cuts across it, corner to corner. The usual institutional response is easy to predict: a polite sign asking people to use the sidewalk, then a low chain, then a row of shrubs, and finally a dirt path with a bend in it where everyone walks around the shrubs. The rarer response, now referred to as the Koolhaas response, is to study the stripe and pour concrete right on top of it.

Most organizations are in a similar position with data governance. People hear "Data Governance" and think about endless meetings, mountains of data dictionary work, onerous processes and budget line-items that are perpetually up for approval next year. All the while, the analysts walking the halls, trying to get the job done are quietly laying down desire paths, attempting to hold everything together without the official sanctioned structure. You'll find most of the standard structures: validation, documentation, reference data, and access control. All of it is happening in spreadsheets, on sticky notes, and in the heads of people who don't realize that "data governance" is something they can put on a résumé.

Every example that follows describes a desire path. Someone needed a governance outcome, and the official route either didn't exist or didn't go where they needed. So they wore their own track into the grass. The instinct is right, and the work is real. What's missing is the pavement.


Path One: The CEO Who Checks the Math

Right action, no home.

Picture a CEO with a monthly ritual. On the first Sunday of every month, coffee in hand, he opens a spreadsheet with a single tab. There's a column of months, a column of revenue figures typed in by hand from the finance close package, and a line chart that looks like it was built in 2011 and never touched since. Beside it sits a second line pulled from the company's BI platform, with every gap between the two shaded in yellow.

Ask him why he does it, and he'll tell you (without any heat) that he has no reason to distrust the numbers. He just doesn't have a reason to trust them yet.

By any textbook definition, what he built is a data quality control. He compares two sources of the same figure and flags variance before it travels further. An auditor would call it a reconciliation. A data engineer would call it a test. He calls it his Sunday check.

The instinct is exactly right. The problem is where the check lives.

This CEO's process is manual, so it runs only when he has an hour to spare. It's private, so when the check passes, nobody else gets the confidence, and when it fails, nobody finds out until he walks into someone's office and starts questioning numbers. It also sits in the wrong chair. The most expensive person in the company has become the last line of defense for revenue accuracy.

What you're looking at is a trust gap, and he's closing it with the only tool he has. Don't talk him out of the spreadsheet. Take his check seriously enough to automate it. Build the same reconciliation into the pipeline: BI revenue against the close figure, a tolerance SLA agreed upon with Finance, and a named owner who gets the alert when the check fails.

He might keep the spreadsheet for a few months, and that's fine. Once the automated check has matched his manual one enough months in a row, he'll stop opening it. Nobody has to ask.

Paving Plan — Promote the Personal Check

When a leader runs their own reconciliation, don't take it personally! Treat it as a requirements document. Ask what they compare, which source they compare it against, and how small the gap has to be before no one is worried. That gives you your test logic, your system of record, and your tolerance for free.

Build the check as an automated test, and make the Finance data owner responsible for it (not the data team, and not the CEO). Publish the result where leadership will see it, such as a certified badge on the dataset or a "reconciled to close on" date in the report footer. In controls language, you've turned a manual, private detective control into an automated, visible one.

person using MacBook Pro
Photo by Campaign Creators / Unsplash

Path Two: The Analyst Who Parameterized Everything

Right action, low transparency.

A senior analyst on the finance team inherited a monthly board pack that had to be rebuilt by hand. Every month she had to open the file, change the dates in six places, update the targets in four measures, and hope she hadn't missed one. She got tired of hoping. She moved the fiscal period dates into parameters and put the targets in a small table. Now the report refreshes on schedule without anyone opening Desktop.

This is good work. I've written before about the weekend I lost changing a server name in 55 reports, and parameters are exactly the cure. They harden the pipeline. They also serve as a kind of documentation, because anyone who opens the file can see which values drive the report and where they're set.

Access and transparency is the problem here. In order to learn when the fiscal year starts or what the Q3 margin target is, you need access to the PBIX sitting on her desktop, and the knowledge that you should look there in the first place (who knew?). That's documentation with a locked front door.

The second problem is change control. Fiscal period boundaries and budget targets aren't technical settings. They're business rules, and somebody in Finance owns them. When the CFO revises a target in July, who updates the parameter? Did anyone know that Sally the senior finance analyst manages these parameters needed to be informed?

When Operations builds its own report next quarter, will they copy her values or type their own? Once a second report carries its own copy of the fiscal calendar, you have two versions of the truth, and they might agree on day one, but that benefit has an expiration date.

Governance draws a line here between two kinds of metadata. Technical metadata describes how something is built: the parameter name and the step that reads it. Business metadata describes what something means and who decides it: "The fiscal year starts February 1, and the Controller owns that rule." Sally the Senior Analyst captured the first kind beautifully. The second kind still lives in her head, and more dangerously, in no one else's head.

Paving Plan — A Parameter Is Not a Record

Connection strings and environment settings belong in report parameters; that's engineering hygiene. A parameter that holds a business rule, such as a fiscal calendar, a target, or a threshold, is different. It needs an owner and a shared home outside any one file. Fiscal periods belong in a single date dimension that every model uses. Targets and thresholds belong in a governed reference table.

A dbt seed works well here, because it's small, version-controlled, and reviewed on every change. Then add the rule and its owner to the glossary, so the next person can find it without opening Desktop.

A woman sitting at a desk with a laptop and papers
Photo by JUICE / Unsplash

Path Three: The Platform Lead With the Lookup Spreadsheet

Right action, wrong system.

Every company has this person: the platform lead who has been around long enough to know which PINs belong to which category, even though the source system has no field for it. When someone asks for analysis on that category, they open a spreadsheet on their laptop and paste the list into a WHERE clause. The file has a few hundred rows, one column of PINs and one of categories, and it gets updated only when the stakeholders get noisy.

They have done something valuable. They captured logic the business uses constantly and the system never recorded. Without this list, every request would start from zero, and every analyst would build their own version. That's how the feudal Excel castles go up.

Now look at where the list lives: one laptop, with one maintainer. It gets updated when the lead remembers, and only if they hear about a new PIN at all. There's no history of what changed, no approval step when a PIN moves categories, and no way for anyone else to find the list. The week this person goes on vacation, the category effectively stops existing.

In governance terms, that spreadsheet is reference data: a controlled list of values used to classify other data. Reference data is among the most commonly governed assets in any organization. Dozens of reports depend on it, and it changes slowly enough that nobody notices when it drifts.

The right home for the list has an owner, a change process, and a spotlight, so anyone can find it, even in the dark. If the source system can carry the category as a real field, that's the best answer. Capture it at entry, make it a required field, and every downstream consumer inherits it. Plenty of source systems can't be changed, though, whether they're vendor platforms or stuck behind an IT backlog measured in years. In that case, the list belongs in a governed mapping table in the warehouse, under version control, with the platform lead as its steward instead of its only copy.

Then, if you're serious about governance, add a test that fails when a new PIN appears with no category. If you read about my ideal Data Governance team, this is the Monitor phase and it's crucial to keeping the governance cycle afloat. This test is also what keeps the list honest after your platform lead moves on to their next role.

Paving Plan — Reference Data Needs a Front Door

A mapping list that lives in one person's files is a single point of failure. Move it into the source system if you can add the field; otherwise, move it into a version-controlled mapping table in the warehouse. Name the person who built it as the steward. Add a test that flags any new, uncategorized value, so the list can't quietly fall behind.

One more check before you move it: if those identifiers are sensitive (personal, patient, or account numbers), the loose spreadsheet is also a security exposure. Its new home needs the same access controls as the source.

white computer keyboard showing white button
Photo by Marcus Urbenz / Unsplash

Path Four: The HR Coordinator With the Delete Key

Right instinct, wrong layer.

Every month, an HR coordinator exports the headcount report, opens it in Excel, deletes the salary column, and emails the file to department heads. Nobody told her to. She simply knew that managers shouldn't see each other's payroll.

That's sensitive data classification and least-privilege access. Right instinct, yes? Sure, but done by hand? Risky!

Those governance functions also depend entirely on her remembering, every single month. The report itself still exposes salary to anyone who has access to it. The export is a copy that can be forwarded anywhere. One missed column and payroll lands in fourteen inboxes.

Better strategy, replace the emailed attachment with a secured report. The data sensitivity instinct becomes a policy that still holds on the Friday she's out sick and the HR intern takes over the process.

Paving Plan — Security Belongs Above the Delete Key

If you're using sensitivity labels in Microsoft, these carry some policy protection into Excel when someone exports the data. But, careful! O365 labels might propagate this to XLS, PPT, PDF, PBIX, but what about CSV?

Classify the column as sensitive, then enforce that classification in the platform. Row-Level-Security in the Power BI model or a masking policy in the underlying database keeps the column away from people who don't need it.

Paving the Paths

Look back at these paths: validation, business rules, reference tables, access control... Together they cover a solid slice of what the Data Management Association (DAMA) body of knowledge treats as governance territory: data quality, metadata, reference data, and security. None of these people attended a governance council, and yet they all had instincts that magically moved them along the path of governance work.

What each of these examples lacked wasn't effort. It wasn't bad intentions or even laziness. It lacked connective tissue. Each person solved their own problem, alone, in their own tool. Nobody formally owned the rule, nobody else could find it, and nobody had the authority to decide how it should change. DAMA defines data governance as the exercise of authority, control, and shared decision-making over data assets, which is a primary component of what is missing in these examples.

This isn't my idea alone. Robert Seiner built an entire methodology on it, called Non-Invasive Data Governance. It starts from the premise that people are already governing data informally, and that a program's job is to recognize and formalize that accountability rather than invent it from scratch.

The idea also maps onto the cycle I've written about before: Define, Apply, Monitor, Discover. The CEO is doing Monitor. The platform lead did Define, alone, on their laptop. The HR Coordinator Applies security with every sensitive column delete. What they don't have is the hub. That means a small group with the authority to decide, owners who carry each decision back to their domains, stewards who watch for drift, and a queue that catches the next desire path before it wears through the grass.

So before you write a charter or buy a catalog tool, maybe walk the lawn? In every department, wrestle with these kinds of questions:

  1. What do you check before you trust a number?
  2. What do you keep in a spreadsheet that the system doesn't know?
  3. What do you remove before you share something?

Each answer is a governance requirement someone has already written for you, and each person who answers is a candidate steward. Recognize them! People support a program that turns out to be what they were already doing, and they resist one that arrives as a new obligation. Many times, that difference decides whether governance sticks or ends up as a charter and a logo.


If you walk through the McCormick Tribune Campus Center today, you're following routes that students wore across a parking lot in the 1990s. Nobody posted a sign telling them where to go. The architect's job was to notice where they were already going and build something worth walking through.

That's the goal for every path in this article: pavement laid under the route people were already taking, done so well that the dirt paths fade from memory. The CEO stops opening his Sunday spreadsheet. The platform lead's list outlives his tenure. The HR coordinator's delete key becomes a policy. Nobody announces any of it. People keep walking the routes they always did, and now those routes are on the map.

You're already doing data governance. Now go find out who else is.