Most data quality work happens as a project. The project ends, and the data slowly goes bad again. The fix is a weekly rhythm, with named owners and monitoring on the few data paths that matter most.
The cleanup that does not last
You have probably seen this happen. A report is wrong in front of an executive. Someone starts a data quality project.
The team scrubs duplicates. They fix the broken joins. They write new rules for the source systems. Everyone feels better for a while.
Then the project closes. The team moves on to the next priority. The source systems keep changing, and nobody is watching them.
A few months later, the numbers are wrong again. Someone starts a new cleanup project.
Why cleanups fade
A cleanup fixes the data as it was on one day. It does not fix the process that made the data bad.
Data changes all the time:
- A source system adds or renames a field.
- A team changes how they enter orders.
- A vendor changes a feed format.
- A new product line does not fit the old categories.
None of these changes waits for your next project. Each one starts to break the data on the day it happens.
That is why I think of data quality as an operating cadence. It is a habit you run every week, not a job you finish.
Pick the few paths that matter
You cannot watch everything. If you try, you will watch nothing well.
Start with the data paths that your business depends on. These are usually the ones behind:
- The numbers your leaders review each week.
- The reports that go to customers or regulators.
- The data that feeds your models and AI tools.
Write the list down. Keep it short. If a path is on the list, it gets an owner and monitoring. If it is not on the list, it waits.
Give each path a named owner
Every path on the list needs one person who owns it. Not a team. Not a channel. A person with a name.
The owner does not fix every problem alone. The owner makes sure someone looks at each problem and that it gets a decision.
When nobody owns a path, every alert belongs to everyone. In practice, that means it belongs to no one.
Monitor before people notice
Monitoring should tell you about a problem before your leaders do. For each path, check the basics:
- Did the data arrive on time?
- Did the row counts change in a way you did not expect?
- Are key fields empty or out of range?
- Do the totals match the source system?
Send each alert to the owner of that path. An alert with no owner is just noise.
This practice has a name: data observability. You do not need to write every check by hand. Tools such as Monte Carlo automate these checks and give you constant information about the data quality status of each path. The tool watches the data all the time. The owner still decides what to do when it finds a problem.
Run the weekly review
Once a week, hold a short review with the owners. Keep the agenda the same every time:
- What broke this week, and on which path?
- What did we fix, and what is still open?
- What changed upstream that we need to watch?
- Does any path need to go on or off the list?
Write down the decisions. Next week, start with the open items from last week.
This meeting is not exciting. That is the point. Boring and steady is what keeps the numbers right.
Do this before the next big initiative
Many teams want to buy a new catalog or start an AI roadmap. Those tools depend on data you can trust.
If the data on your critical paths is not owned and watched, a new tool will not fix it. It will show the same bad data in a new place.
Start the weekly rhythm first. Pick the paths. Name the owners. Turn on the checks. Then build on top of data that stays right.