The problem with "garbage in, garbage out" | Sunbeam Consulting I’ve always squirmed at the phrase “garbage in, garbage out.”

I’ve heard it dozens of times, always with good intention: someone arguing that we need better data quality because the data’s a mess.

But my problem with it is that data is never perfect . There is always some degree of garbage. If you wait for clean data, you’ll wait forever. And if you spend all your time cleaning up the garbage, you’ll waste a bunch of time and over-clean.

It’s a phrase that sounds smarter than it is. Nobody wants to argue for garbage, so it’s hard to push back on. But it’s misleading: it usually just results in premature optimization.

The data is likely good enough to get going. Then improve quality incrementally, and only where there’s evidence it will actually improve the outcome.

This is also the bet behind a project I’m building right now: it takes “garbage” unstructured data and incrementally adds structure to improve ML models. It embraces the mess, cleaning up only what’s needed for a specific target outcome.