Attrition is the steady loss of drug candidates as they move through development, and it is severe. For every compound that enters early testing, only a small number reach human trials at all, and of those, only a fraction complete the full sequence to approval. The losses are not evenly distributed. In terms of raw count, most candidates are eliminated early, during laboratory and animal testing, because they fail basic safety or efficacy checks. But in terms of cost and time, the losses that matter most happen late.
A failure in a late-phase trial is damaging for three reasons. First, the money already spent is gone: years of salaries, manufacturing, site payments, and monitoring cannot be recovered. Second, the time is gone: a decade may have passed, and the patent clock has been running the whole time, leaving less protected market time if the drug had succeeded. Third, the failure is often ambiguous. A late trial can fail because the drug truly does not work, or because the trial was poorly designed, enrolled the wrong patients, or could not measure the effect reliably. That ambiguity makes it hard to know whether to abandon the approach or try again differently.
This is why the funnel shape matters. The wide top represents many cheap early candidates; the narrow bottom represents few, very expensive late candidates. The cost of a mistake rises sharply as you move down, so the process is designed to be cautious precisely where it is most expensive to be wrong.