How many clicks should a keyword get before you decide it does not work?

It depends on the conversion rate, and the arithmetic is simpler than people make it. If your site converts at 3 percent, then roughly one visitor in 33 converts. At 33 clicks with no conversion you have seen about one expected conversion's worth of traffic, which is not evidence of anything - a keyword performing exactly at your account average has a decent chance of showing zero at that point. At 100 clicks you would have expected three, and zero starts to look like a real difference rather than bad luck. So: three times your break-even click count is where I start taking it seriously, and I pause rather than delete so the decision is reversible. The other half of the answer is cost, not clicks. A keyword at £8 a click deserves a decision much sooner than one at 40p, because the cost of being patient is completely different. Judge on money spent against your target cost per conversion, and use clicks as the sanity check.