One of the odd things about science is that you spend a good amount of time trying to prove that you don't know what you're talking about. At least, if you're doing it right, you should.
Take the first part of a drug discovery project, for instance. Most of them have a "primary assay", which is usually done against a purified protein in vitro, under fairly artificial conditions. Compounds that meet some standard of activity against that target then move on to the secondary assay, which is supposed to be aimed at the same process, but now it's done in living cells. That's a much tougher test. (It's a big leap from pure proteins to cells, about the same size as the leap from cells to whole animals.)
The hope is that the two assays will correlate with each other, but it's never a perfect fit. Generally, what you see is some of the active compounds dropping out for no apparent reason in the cell assay. If your target is in the cytoplasm, then there's always the possibility that these compounds don't penetrate into the cell as well as the others. Or they make it in, but are pumped right back out before they can get anything accomplished. Or perhaps they find some other (even tighter) binding site once they're inside, on some protein unrelated to the readout of your assay. There are always plenty of ways to explain these misfires.
And that's fine, as far as it goes. But if you don't double back and check these things out occasionally, you run the risk of fooling yourself. If your two assays don't correlate very well, it might be that cell penetration is lousing things up, sure - and it might also be that your assays aren't measuring the same thing. Or it could be that your target from the first assay isn't as important as you thought it was. These are the sorts of thing you really ought to be sure about.
So you need to keep yourself honest. Take some of your not-so-good compounds, the ones you'd normally discard after the first cut, and take them on to the cell assay regardless. They'd better not work! Test some of the compounds on a closely related cell line that doesn't have your target in it, if you've got some - is your target really the reason for the activity you're seeing?
Most of the time, you'll find that things are just fine. The inactive compounds really are inactive all the way through. But I've seen the exceptions occur, and more than once. You don't want to wait any longer than necessary to find out that your project is a dud. And worse yet, you really don't want someone else to find out for you. It leads to some of those awkward scenes we'd all rather avoid.