Cost per lead

Because of the direction it moves when things go wrong. Real leads cost money to acquire. Fake ones cost approximately nothing to produce. So any contamination of the denominator pushes CPL down, and a spam wave presents as the best month the campaign has ever had.

Two sentences, in that order. The metric improved and the business got worse, and the metric had no way to say so.

Never report CPL alone. Report it next to something the junk cannot fake:

• Cost per qualified lead, using whatever your sales team already means by qualified — see MQL vs SQL.
• [MQL→SQL rate](/glossary/mql-vs-sql) by campaign. If CPL falls and this falls with it, you did not get cheaper. You got worse.
• Cost per closed deal, on the cohort that has had time to close — which depends on your time to disposition.

The pair is the point. CPL alone cannot distinguish a better campaign from a contaminated one, and the two look identical on the chart.

A target CPA or CPL instructs an ad platform to find you more conversions at a lower price. The cheapest conversions available are the ones that cost nobody anything to produce, so the optimiser finds them, reliably, because you asked.

The fix is not to abandon the target. It is to change what counts as a conversion, so the cheap thing stops qualifying — offline conversion import is the standard route, and it is what competent teams already do.