A screening failure is a person who is checked against the eligibility criteria and turns out not to qualify. Each one costs time and money and produces no usable data, so screening failures are the most wasteful part of recruitment. They happen when the criteria are applied loosely at the first pass: a broad search flags many people who look plausible but do not actually fit.
Matching is the step that reduces them. Rather than checking whether a person satisfies each criterion in isolation, matching compares the person's full profile against the trial's requirements together, and only surfaces people who satisfy the combination. The effect is visible as a trade-off. Loosen the criteria and the shortlist grows, but so does the number of people who will fail screening. Tighten them and the shortlist shrinks, but the people on it are more likely to qualify. The useful setting is the one that keeps the shortlist large enough to fill the trial while keeping screening failures low.
This is a balance rather than a fixed answer, because the right setting depends on how many eligible people exist and how many the trial needs. AI helps by making the consequences of a given set of criteria visible before the search is run, so the choice is made with evidence instead of by trial and error.