You were hired to answer questions. Then the engineer left, or there never was one, and now the warehouse is yours. Nobody gave you a test suite. Most warehouses don’t have one.
That’s a familiar spot. Analysts already do the ad hoc data engineering themselves, in SQL, Alteryx, and Excel, pulling the ten tables that matter out of hundreds. Testing them is the next job, and it needs less code than you’d think.
Profile it before you’re blamed for it
Before you change anything, profile every table and keep the dated result. The First Move When You Inherit Someone Else’s Data makes the case: within weeks, every problem in that warehouse becomes yours to explain, including the ones that were there before you. A profile dated before your first day records the null rates, schema violations, and broken joins you walked into.
What you can do without an engineer
| Job | Who does it | What you need |
|---|---|---|
| Profile every table | You | A read connection to the warehouse |
| Freshness, volume, and schema monitors | You | The same connection; monitors learn from history |
| Baseline tests on key tables | You, reviewing generated tests | A tool that generates tests from the profile |
| Business rules for key metrics | You, because you know them | Plain-language rules, or an AI tool to draft them |
| Stop the pipeline when a test fails | An engineer or IT | A test task added inside the orchestrator |
| Fix the code a failed test points at | An engineer, or you with help | Access to the pipeline code |
The top four rows are most of the value, and they’re yours.
Code-first test frameworks help engineers write tests faster. They still start every test as code. A tool that profiles the data and generates the tests turns your job into review: keep what’s right, delete what isn’t. DataOps TestGen is one such tool. It profiles each column on 55 characteristics, flags issues with 32 hygiene detectors, and generates tests from 34 test types. The open source version is free for one user, one connection, and one project, and the install page puts setup at two commands.
Which comes first, monitors or tests covers the order. Monitors go on everything. Tests go where a wrong number costs money.
Your advantage is the business rules
Generated tests learn what the data already looks like, including what’s already wrong. In DataKitchen’s NYC taxi exercise, 62 generated tests passed on a 300,000-row sample, because the minimum-fare test set its floor at negative $367. Seven business rules failed on that same sample. One of them said fares are never negative.
An engineer can’t write that rule without asking someone like you. Say you know that refunds never exceed the original charge, or that each territory rolls up to exactly one region. Write rules like those in plain sentences. An AI tool connected to TestGen over MCP can turn them into tests, as it did in the taxi exercise. Can AI write data quality tests explains where you still have to check its work.
What still needs help
Testing tells you a table is wrong. It doesn’t stop the next step from reading it. That takes a tripwire, a test task inside the pipeline that exits with an error, and adding one means changing the pipeline. Ask IT for that once your tests have earned some trust.
If the pipelines are scripts you wrote with an AI tool, and you’re now the only one running them, that’s its own problem. You Built the Thing. Now You’re Stuck Running It. is about exactly that.