Garbage in, garbage out. Or is it garbage out, garbage in?
Why poor data keeps itself going, and how to break the cycle.

In the data world, from BI to data science to AI, everyone has heard the term garbage in, garbage out. In fact, we hear it too often. It usually comes with a shrug: the data wasn't good, we worked with it anyway, and that's the end of the matter.
Yet that conclusion isn't right.
So that's that dealt with
Garbage in, garbage out is mostly used as an end point. The data was messy, so the result is messy too, and as soon as the report, dashboard or forecast is delivered, the responsibility has been handed over. The problem is acknowledged, and then let go.
That attitude seems harmless. It gives a team room to carry on working without having to reopen the discussion. It isn't harmless, because the output it justifies doesn't disappear. It comes back in somewhere else.

Output is always someone else's input
No output stays where it was created. A report is read by a manager. A dashboard steers a decision. An AI forecast triggers a follow-up action. The garbage out of one process is, almost always, the garbage in of the next.
That's why the reversal matters at least as much as the original. Garbage out is garbage in.
From operational to strategic, and back again
Take an operational report with incomplete numbers. That report feeds a tactical overview, put together by someone who can no longer see the underlying mess. That overview in turn feeds a strategic decision, taken by someone even further from the source.
At every step the data gets a layer more abstract, a layer more convincing and a layer less visibly wrong. Nobody in that chain can see the original garbage any more. Everyone sees only the input they receive themselves, and takes it seriously.
Why nobody feels like the owner
That's exactly the problem with how we use the term. Whoever produces the output doesn't feel responsible for what happens to it after delivery. They've done their part.
Whoever receives the output assumes in turn that it has already been checked by whoever supplied it. They have no reason to doubt it, especially when the output is slickly presented. A sleek dashboard or a tidy report looks reliable, and that look is soon confused with the reliability of the data behind it. The form convinces, not the content.
Between those two assumptions, quality stalls. Nobody is lying, nobody is negligent, and yet nobody checks.
AI makes the chain faster, not shorter
With AI, that effect gets bigger, not smaller. A model uses output from one system as input for the next, often with no human in between to assess the intermediate step. Agentic AI even takes that next step on its own, based on what it has just produced itself.
So the garbage travels further through the organisation, and faster, with less chance of anyone noticing it along the way before it serves as input somewhere else.
A warning, not a licence
So garbage in, garbage out shouldn't be an excuse to use poor data anyway, with the idea that you're done with it after the output. It should be a warning that travels with that output, all the way through the process.
Anyone who realises that asks an extra question at every delivery. Is the data good enough for this purpose, and also: where will this output come back in, and with whom?
Garbage out is garbage in!
The next time someone shrugs and says garbage in, garbage out to justify poor data, the answer is simple. The conversation isn't over at the output. It starts again there, for someone else, at the next level.
Garbage out is garbage in. Until proven otherwise.
Written by Ruud Kuil
Co-founder of MAD-Quality. Helps leadership teams move from data chaos to data they dare to manage by. Want to know more about data quality? Follow Ruud on LinkedIn.



